A Sensitive and Specific Neural Signature Robustly Predicts Graded Computations of Moral Wrongness

A Sensitive and Specific Neural Signature Robustly Predicts Graded Computations of Moral Wrongness | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article A Sensitive and Specific Neural Signature Robustly Predicts Graded Computations of Moral Wrongness Frederic Hopp, Sungbin Youk, Walter Sinnott-Armstrong, René Weber This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7935407/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Humans universally condemn what they see as moral violations, yet the perceived wrongness of distinct moral transgressions varies across individuals, neurodiverse populations, and cultures. We currently do not know how and to what extent the human brain universally computes and translates moral wrongness into subjective, graded moral judgments. Here, we combined fMRI with pattern recognition techniques to identify and evaluate a neural signature predictive of graded moral wrongness judgments. Drawing on to date’s largest database for studying the neural basis of moral judgment, spanning independent, multi-culture fMRI datasets of moral vignettes for discovery (n = 64), validation (n = 30), replication (n = 27), and generalization (n = 30) analyses (n total =151), we demonstrate that accurate prediction of graded moral wrongness relies on a distributed neural circuit, with important contributions from cortical and subcortical areas. We further evaluate common and domain-isolated (e.g., care, fairness, purity) brain systems for graded moral wrongness and demonstrate shared and dissociable neural representations with negative affect and subjective disgust. Together, we find that graded moral wrongness judgments are robustly computed via a shared and distributed neural code and provide a sensitive and specific moral wrongness biomarker for future studies. Biological sciences/Psychology/Human behaviour Biological sciences/Psychology Biological sciences/Neuroscience/Cognitive neuroscience Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction The discernment of gradations in transgressive behaviors provides a compass for regulating human life. Indeed, the relative severity of moral violations is a fundamental issue in ethics and is codified in systems like the US Penal Code, which distinguishes between infractions, misdemeanors, and felonies, and provides statutes for the legal ramifications of different types of transgressions 1-2 . How the human brain computes the graded reprehensibility of varied moral transgressions remains a compelling and enduring puzzle for neuroscience 3-5 . Decades of research demonstrate that moral judgment and decision making recruit domain-general neural circuits, from affective 6 and value-based systems 7-8 to mentalizing networks 9 . Meta-analyses of moral brain mapping studies further document that different kinds of moral judgements are reliably orchestrated by the default mode network (DMN) 10-12 , with robust contributions from dorsomedial and ventromedial prefrontal cortex (PFC), temporoparietal junction (TPJ), as well as posterior cingulate cortex (PCC) and precuneus (PC). Neural activation patterns in the DMN contain information concerning the nature of the observed moral behaviors, enabling the flexible distinction between intentional and accidental harms 9 or varied moral violations (e.g., betrayal, cheating, etc.) 13-16 . However, we still know little about the neural circuit supporting the precise computation of graded moral wrongness and whether individual, neurologic, and cultural differences in brain representations of moral severity predict variation in momentary, self-reported, and graded moral judgments 2 . We herein capitalized on recent advances in multivariate pattern analysis (MVPA)-based neural decoding 17 to develop a multivariate brain model that can predict the subjective experience of degrees of moral wrongness. In contrast to extant univariate moral neuroimaging studies that express brain activation as a function of the underlying binary moral judgment (i.e., right versus wrong) 6,18-20 , our developed multivariate brain model permits the inference of continuous (graded) moral wrongness judgments given brain activity 21 . Multivariate brain models have already yielded large effect sizes in brain-outcome associations for somatic and vicarious 22 as well as future pain 23 , negative affect 24,25 and pleasure 26 , as well as fear 27 , threat anticipation 28 , and disgust 29 . Furthermore, quantitative predictions about outcomes can be empirically falsified, tested, and validated across studies, populations, and scanner settings, which promotes generalizability and reproducibility 30 . Notably, individuals may not always have reliable insights into their moral cognitions 31 or succumb to social desirability response biases, whereas predictive brain models can yield objective neurobiological read-outs of diverse mental states 32 . Finally, examining for which individuals and under which conditions the model predictions fail can provide novel insights into aberrant moral cognition and serve as targets for interventions 33 . In this work, we therefore developed a functional magnetic resonance imaging (fMRI)-based model, or neurologic signature, of graded moral wrongness judgments. Using distributed blood-oxygen-level-dependent (BOLD)-based information, we investigate whether (1) it is possible to develop a sensitive and specific neural signature of severity ratings of subjective moral wrongness on the population level, (2) this neural signature robustly generalizes across independent samples, MRI systems, and two different cultures, and (3) can predict the graded severity of momentary (trialwise) moral wrongness judgments on the individual level, (4) the neural signature in brain regions (e.g., the dorsomedial prefrontal cortex; dmPFC) and networks (e.g., DMN) implicated in moral judgment is sufficient to capture the continuous rating of degrees of moral wrongness, and (5) the neural representation of momentary graded moral wrongness judgments is distinct (i.e., sensitive and specific) from brain representations of general emotionally aversive states and subjective disgust. More specifically, we trained a linear support vector regression (SVR) algorithm 27-29 in healthy subjects (n = 64) from a previous study 13 to identify the brain signature that predicts the graded severity of trial-by-trial subjective moral wrongness ratings elicited by validated and standardized moral foundations vignettes (MFV) 34 . We then evaluated the performance of the established moral judgment signature (MJS) in (1) a novel U.S.-based validation cohort using the identical MRI system and MFV task but with additional vignettes (n= 30); (2) a U.S.-based conceptual replication cohort from a previous study 14 at a different campus with a different MRI system and using a modified MFV task (n=27); and (3) a novel generalization cohort using a validated Dutch version of the MFV 36 in the Netherlands with a different MRI system and Dutch subjects (n = 30). Furthermore, partial least squares regression (PLS-R) provided a framework for jointly estimating both generalized (common) and domain-specific representations of graded moral wrongness across physical and emotional care, fairness 1 , liberty, loyalty, authority, sanctity, and social (conventional) norms, allowing us to examine how generalized and domain-specific representations jointly contribute to degrees of moral wrongness judgments 24,36,37 . To extend the perspective from a population to an individual level, we probed whether the MJS can predict trial-wise graded moral wrongness ratings for each subject in discovery, validation, replication, and generalization cohorts separately. We further systematically identified brain regions that are associated with (forward model, i.e., expressing the observed data as functions of underlying variables) and predictive of (backward model, i.e., expressing variables of interest as functions of the data) graded moral wrongness judgments 38 and examined to what extent single brain systems or networks can capture degrees of moral wrongness severity. Moreover, we probed the sensitivity and specificity of the developed MJS for graded judgments of moral wrongness. Considering the intricate role of affect for moral judgment 31,39-42 , we determined the functional specificity of the neural moral judgment signature by comparing the spatial and functional similarities between the MJS with predictive brain markers for general negative emotional experience 25 and disgust 29 . Finally, we validated the MJS in novel contexts and paradigms, including intentional versus accidental moral transgressions 9 , socio-moral images 35,43 , and unfair social offers 29 . Results Moral vignettes elicited a robust range of moral wrongness judgments Moral wrongness judgments were elicited by validated moral foundations vignettes (MFV) 34 shown to reliably induce varying levels of perceived moral wrongness 13-14,35 . The MFV span 120, one-sentence descriptions detailing the violation of one (and only one) of seven moral foundations: Physical care, emotional care, fairness, liberty, loyalty, authority, and sanctity. The vignettes also contain a non-moral, social norm transgression category. Subjects were explicitly instructed to vividly imagine each scenario and were asked to rate the moral wrongness of each vignette on a 4-point Likert scale ranging from 1 (not morally wrong) to 4 (extremely morally wrong) (Fig. 1a; Extended Data Fig. 1). As the vignette appeared on screen, subjects could immediately provide a graded moral wrongness judgment to better capture moral judgments’ hypothesized intuitive nature 31 . We initially tested whether the vignette stimuli elicited meaningful and varying levels of perceived moral wrongness. To this end, we plotted the number of each selected moral wrongness level (across subjects and categories) for each run (Extended Data Fig. 2a) and for each vignette category (across subjects and runs, Extended Data Fig. 2b). We found that the stimuli induced sufficient levels of moral wrongness in the discovery cohort (n = 64) which was used to develop the neural signature of moral wrongness (see below for details), such that between 9% (Authority) and 45% (Physical Care) of trials of each moral vignette were rated as 4 (reflecting that they induced strong moral condemnation). Confirming previous work 34-35 , violations of physical care were rated as most morally wrong, whereas transgressions of conventional norms received the expected lowest moral wrongness ratings (Extended Data Fig. 2b). Individual differences in self-reported sensitivity to moral foundations were correlated with moral wrongness ratings in a domain-consistent way ( r = 0.35–0.47; Supplementary Table 2). Self-reported moral wrongness judgments were generally evenly distributed across runs and all 64 subjects used all 4 levels of moral wrongness ratings at least once. A brain signature for graded computations of moral wrongness (MJS) We used brain activity during the vignette presentation and moral judgment phase (Fig. 1a) across all moral foundations to develop a multivariate model capable of predicting graded moral wrongness ratings. We trained a support vector regression (SVR) algorithm to predict the selected graded moral wrongness ratings using the smoothed and standardized, whole-brain parametric maps for each moral judgment level (1–4) and each subject as input features (Fig. 1b). The resulting pattern of voxel weights was considered as the moral judgment signature (MJS) (Fig. 2a). We assessed the performance of each pattern in predicting the graded severity of moral wrongness with leave-one-subject-out (LOSO) cross-validation in which the same SVR was iteratively trained on the parametric activation maps from all but one subject and tested on the activity maps of the held-out subject. The analysis indicated that across subjects, the developed MJS accurately predicted graded degrees of moral wrongness judgments. Specifically, for individual subjects the average within-subject correlation between predicted and actual graded moral wrongness ratings (4 pairs of scalar values per subject) was r = 0.94 ± 0.02 (standard error (SE)), the average root mean squared error (RMSE) was 0.68 ± 0.05 and the overall (between- and within-subjects) prediction-outcome (i.e. 256 pairs) correlation coefficient was 0.78 (averaged across 64 repetitions). Testing the MJS model developed in the discovery cohort, with no further model fitting, in the validation, replication, and generalization cohorts yielded significant graded moral wrongness predictions (Fig. 2c–e; Supplementary Table 3), indicating a robust and replicable neurologic signature for graded computations of moral wrongness. To further determine the sensitivity of the MJS to predict graded degrees of moral wrongness, a two-alternative forced-choice test was applied, comparing all possible pairs of activation maps within each subject and choosing the one with higher MJS response as more morally wrong. In the (LOSO cross-validated) discovery cohort (Fig. 2b; Supplementary Table 4), the MJS response classified not morally wrong (1) versus extremely morally wrong (4) judgments with 98% accuracy; not morally wrong (1) versus very morally (3) with 98% accuracy; and moderately morally wrong (2) versus extremely morally wrong (4) with 98% accuracy. Moreover, the MJS response could distinguish each successive pair of moral wrongness rating levels (e.g., rating 2 versus 3) with ≥ 91% accuracy, which was significantly better than chance level (50%; p < 0.001) Similar performance was achieved across validation, replication, and generalization cohorts, again demonstrating an accurate neural signature for graded moral wrongness predictions. Retraining the MJS excluding the entire occipital lobe 44 revealed slightly lower, but significant prediction accuracies ( r = 0.91 ± 0.03 for within-subject prediction-outcome correlations and r = 0.71, p < .001, for the overall prediction-outcome correlation; Extended Data Fig. 3), suggesting that graded computations of moral wrongness are informed by the visual system 45-46 , but also draw on additional distributed brain systems. In addition, we applied the MJS to the vignette time series data using dot product across discovery (LOSO cross-validated), validation, replication, and generalization cohorts (Fig. 2 f–i) to evaluate the chronometry of the moral wrongness judgment pattern. Visual inspection of the MJS reactivity at each timepoint following stimulus onset indicated that the MJS response began approximately 4sec following vignette onset and increased with increasing levels of reported degrees of moral wrongness during approximately 5–9 sec. These findings align with previous work demonstrating a temporal sequence of moral deliberation followed by moral verdict 47 and confirmed that the MJS dynamically tracked graded computations of moral wrongness judgment. Within-subject trial-wise prediction The subjective feeling of disapproval that comes with experiencing a moral transgression suggests that moral judgments are momentary, and to some degree individually constructed states 31,48 . Thus, a key question is to what extent the population-level model (i.e., the MJS), which is a statistical summary of a highly variable set of instances, can predict momentary (trial-wise) graded moral wrongness judgments for each subject (on the individual level). To this end, we performed single-trial analyses using the Least Squares All (LSA) approach 49 to obtain a graded moral wrongness beta map for each vignette item for each subject across discovery (~120 beta maps per subject), validation (~72 beta maps per subject), replication (~120 beta maps per subject), and generalization (~120 beta maps per subject) cohorts. The MJS was next applied to these beta maps to calculate the pattern expressions which were further correlated with the true graded moral wrongness ratings for each vignette and subject separately. The statistical significance was evaluated by prediction-outcome Pearson correlation for each subject separately. We found that the MJS could significantly predict graded trial-by-trial moral wrongness ratings for 61 out of 64 subjects (95.3%) in the discovery cohort (cross-validated) and for 25 out of 30 (83.3%) subjects in the validation cohort, for 11 out of 27 subjects (40.7%) in the replication cohort, and for 21 out of 30 subjects (70%) in the generalization cohort. The mean prediction-outcome correlations were 0.44 ± 0.15 (discovery; p < .0001), 0.34 ± 0.11 (validation; p < .0001), 0.17 ± 0.12 (replication; p < .0001), and 0.22 ± 0.14 (generalization; p < .0001) (Fig. 2 j–m; two-sided P values based on a 10,000 samples bootstrap test of within-subject r values). Moreover, the MFV are controlled on a range of dimensions (e.g., syntactic structure, word and character length) but deliberately vary in the type of transgressions they portray (e.g., physical care, fairness, loyalty, etc.). Thus, we explored whether the MJS pattern could predict the graded moral wrongness ratings for each individual vignette item. Here, we first averaged the within-subject trial-wise MJS pattern expressions and graded moral wrongness ratings for each vignette. We then computed the Pearson correlation between the z -scored and averaged MJS pattern expression and average graded moral wrongness rating for each vignette item. We found strongly positive and statistically significant linear relationships between the average ratings and pattern responses across all vignettes in the discovery (cross-validated) cohort r = .88, p < .001, the validation cohort r = .83, p < .001, the replication cohort r = .55, p < .001, and the generalization cohort r = .60, p < .001 (Fig. 2n–q; Supplementary Table 5). Compellingly, the averaged MJS pattern expressions for each vignette item in the (cross-validated) discovery cohort correlated strongly positively and significantly ( r = .79; p < .001) with the averaged graded vignette moral wrongness ratings from an out-of-sample U.S. population study (n = 510) 34 . Analogously, the averaged MJS pattern expressions for each vignette item in the generalization cohort correlated positively and significantly ( r = .58; p < .001) with the averaged graded vignette moral wrongness ratings from an out-of-sample, Dutch population study (n = 586) 35 . Together, these findings suggest that the MJS reliably predicts computations of graded moral wrongness across both individuals, cohorts, and vignette items. Common and domain-specific brain representations of moral wrongness A central question for moral neuroscience is whether moral judgments of different types of moral behaviors are captured by common or domain-specific neural representations. Recent work demonstrates that moral judgment of different moral transgressions elicits dissociable neural activations 13-16,19 , but it is not clear to what extent the representation of graded moral wrongness for variable moral violations is robustly instantiated via common and domain-specific brain representations. Inspired by recent findings revealing common and stimulus-type-specific brain representations of negative affect 24 , we used PLS-R on the discovery cohort data (Fig. 3a) for jointly estimating both common and domain-specific (care, fairness, loyalty, etc.) graded moral wrongness brain representations (see Methods and Extended Data Fig. 4). This produced nine multivariate patterns: one for each of the eight vignette conditions and one for common moral wrongness. Models were tested using LOSO cross-validation. To evaluate model predictions, we examined whether (i) average graded moral wrongness predictions (i.e., model responses) were significantly higher for each matched model-domain pair and (ii) there was a significant correlation, expressed as mean within-subject r ± standard error (s.e.), for each model and graded moral wrongness rating. First, each of the eight domain-isolated models exhibited the highest average model response for its matched target domain (Fig. 3b, Supplementary Table 6), providing further evidence that computations of graded moral wrongness for distinct moral transgressions rely on dissociable neural representations. Second, the common model predicted graded moral wrongness ratings for each moral vignette condition, but not social norms, as evidenced by significant associations between observed and predicted ratings (Fig. 3c–d; Supplementary Table 7). The domain-isolated Physical Care ( r = 0.36, p = 0.001), Fairness ( r = 0.62, p = 0.001), Liberty ( r = 0.41, p < 0.001), and Sanctity ( r = 0.43, p < 0.001) models were also significantly sensitive towards graded moral wrongness computations in their respective target domains, and Emotional Care ( r = 0.17, p = 0.09) as well as Loyalty ( r = 0.18, p = 0.06) approached statistical significance. Of note, none of the domain-isolated models were specific to their target condition, as evidenced by statistically significant cross-prediction of off-target graded moral wrongness ratings (Fig. 3c, Supplementary Table 7). Taken together, these results provided strong support for common coding of graded moral wrongness, and mixed evidence for domain-specific graded moral wrongness representations. Moral wrongness is associated with and predicted by distributed neural systems In view of the evidence for domain-general computation of graded moral wrongness, we next systematically determined individual brain regions that were associated with subjective graded moral wrongness ratings and that provided consistent and reliable contributions to the whole-brain moral judgment decoding model (MJS) using different analytic strategies. We first examined regions that made reliable contributions to the graded moral wrongness prediction within the MJS itself by applying a bootstrap test to identify regions with significant, consistent model weights ( q < 0.05, false discovery rate (FDR) corrected). Given that some brain features could contribute to controlling for noise in the data 38 rather than the moral judgment computation per se, we next transformed the population-level MJS into ‘activation pattern’ (‘structure coefficient’; for details, see Methods). The results showed that a set of distributed brain systems exhibited significant model weights ( q < 0.05, FDR corrected; Fig. 4a) and structure coefficients ( q < 0.05, FDR corrected; Fig. 4b), including dmPFC, the precuneus, as well as visual and supplementary motor area (Fig. 4c). Brain regions associated with and predictive of moral wrongness computations on the individual level determined with convergent univariate and multivariate approaches identified a similar set of broadly distributed regions (Supplementary Methods, Supplementary Results and Extended Data Fig. 5). The broad conclusion is that the neural representation of graded moral wrongness is not limited to a single or a set of focal regions (e.g., the dmPFC), but rather includes a broad set of regions spanning multiple systems. MJS outperforms prediction based on local systems Due to continuing debates on the contribution of specific brain regions 5,50 , such as dmPFC 13,19 , vmPFC 51 or the default mode network (DMN) 10,52-53 to human moral judgment, (1) both searchlight- and parcellation-based analyses were employed to determine local brain regions that were predictive of subjective moral wrongness severity, and (2) models were trained on single brain region and networks to examine to what extent these models could predict moral wrongness severity compared to the whole-brain MJS. As shown in Fig. 5a,b, moral wrongness severity could be significantly predicted by distributed regions, including dmPFC and PCC as well as visual cortex and supplementary motor area (SMA) ( P < 0.001, average across leave-one-subject-out cross-validation procedure). However, none of the local models predicted moral wrongness to the extent the MJS did, suggesting that moral wrongness representations are distributed across regions, and can best be captured in whole-brain but not local analyses. We then re-trained predictive SVR models restricted to activations in (1) the bilateral dmPFC, (2) the bilateral vmPFC, (3) a meta-analytic “moral” map 54 (Extended Data Fig. 6), (4) a consciousness network, (5) a subcortical network, and (6) each of seven large-scale cerebral networks (Methods). Our findings showed that the dmPFC (prediction–outcome correlation r = 0.18 and 0.38 for discovery (cross-validation)), the vmPFC (prediction–outcome correlation r = 0.27 for discovery (cross-validation)), and other brain networks (Fig. 5c–e) could, to some extent, predict graded moral wrongness judgments. Nonetheless, although statistically significant ( p < 0.001), the effect sizes in terms of prediction–outcome correlations (including searchlight- and parcellation-based predictions) were substantially smaller than those obtained from the MJS, which used features spanning multiple brain systems. To control for potential effects of the number of features/voxels in prediction analyses (that is, the whole-brain model contains many more features), we randomly selected voxels (repeated 1,000 times) from a uniform distribution spanning the entire brain (black; Fig. 5e), consciousness network (light orange), subcortical (brown) or individual large-scale cerebral networks (averaged over 1,000 iterations) 27-29 . The asymptotic prediction when sampling from all brain systems as we did with the MJS (black line in Fig. 5e) was substantially higher than the asymptotic prediction within individual networks (coloured lines in Fig. 5e). Notably, only the visual system showed an initially higher prediction-outcome correlation compared to the whole-brain model when sampling fewer than 1000 voxels, which further underlines the relative importance of the occipital lobe for representing moral 13,45-46 and emotional 55 concepts. Nevertheless, this analysis thus demonstrated that whole-brain models have much larger effect sizes than those using features from a single network. Moreover, model performance was optimized (that is, reaching asymptote) when approximately 10,000 voxels were randomly sampled across the whole brain, as long as voxels were drawn from multiple brain systems, further confirming that information about moral wrongness is contained in patterns of activity that span multiple systems. Together, converging lines of evidence from the above systematic analyses point to the fact that subjective, graded moral wrongness judgments are encoded in distributed neural patterns that span multiple systems, adding to increasing evidence that morality is represented in distributed brain systems rather than single brain regions or networks. Separable signatures of moral wrongness, negative affect, and disgust Sentimentalist theories of moral cognition assert that moral judgments are fundamentally rooted in emotional responses 31,39-42 , but the extent to which neural representations for computing graded moral wrongness overlap with brain representations of graded (negative) emotional experience remains unclear. Accordingly, we determined whether (i) neural biomarkers of general negative emotion experience (PINES) 25 and specific socio-moral emotions such as disgust (VIDS) 29 can predict graded moral wrongness judgments and (ii) whether the MJS is sensitive and specific to graded moral wrongness judgments. To address these questions, we performed a series of analyses. First, we investigated spatial similarities between stable decoding maps and a set of regions of interest (ROIs) commonly involved in moral cognition and emotion as well as networks 10 . Speaking to the distributed nature of moral wrongness computations, the MJS was the only model that contained stable predictive voxels in all regions (Fig. 6a). Amygdala, dmPFC, anterior Insula, and precuneus contained stable predictive voxels in all three models, but the visually induced disgust signature (VIDS) showed larger contributions to the amygdala, anterior Insula, and precuneus than MJS and picture-induced negative emotion signature (PINES). In contrast, the MJS exhibited the largest contributions in dlPFC and pSTS, while dlPFC, vmPFC, thalamus, and pSTS exhibited a degree of specificity for MJS and VIDS. In line with previous work 29 , all networks showed stable predictive voxels across the three models, but the relative contributions of each network to each model varied, such that the visual network contributed more strongly to the MJS than to VIDS and PINES; the ventral attention and limbic networks contributed more strongly to VIDS than to MJS and PINES; and the somatomotor network strongly contributed to PINES (Fig. 6b). Second, we examined functional similarities between the MJS, PINES and VIDS, respectively. The results showed that the MJS was more sensitive and specific to predict high versus low moral wrongness as compared with PINES or VIDS, as reflected by larger effect sizes for MJS in (cross-validated) discovery (2.18–6.95 times larger), validation (3.76–10.44 times larger), replication (2.73–3.74 times larger), and generalization (2.02–3.34 times larger) cohorts (Fig. 6c). PINES was more sensitive to predict high versus low negative emotion with effect sizes 4.17–48.0 times higher than those for MJS or VIDS (Fig. 6d), and VIDS more accurately predicted high versus low disgust with effect sizes 3.13–27.32 higher than those for PINES or MJS (Fig. 6e). The above findings were further substantiated by the comparisons of the overall and within-subject prediction–outcome correlations of the three decoders across four datasets (Supplementary Table 8). Finally, we employed multi-level mediation models, which examined whether the covariance between two variables ( X and Y ) can be explained by a third variable ( M ), to determine the neurofunctional relationship between the representations of moral wrongness encoded in MJS, PINES, and VIDS. While PINES could to some degree track moral wrongness (discovery cohort: r = 0.46; validation cohort: r = 0.16; replication cohort r = 0.32; generalization cohort r = 0.22), the MJS response partially mediated the effect of the PINES response on moral wrongness ratings in the discovery cohort, and in the validation cohort the MJS response fully mediated the effect of the PINES response on moral wrongness ratings (Fig. 6f). The MJS response also partially mediated the effect of the PINES response on moral wrongness ratings in the replication cohort, whereas in the generalization cohort the MJS response did not mediate the effect of the PINES response on moral wrongness ratings (Extended Data Fig. 6a). In contrast, the PINES response failed to mediate the effect of the MJS response on wrongness ratings in discovery and validation cohorts (Fig. 6g) as well as in replication and generalization cohorts (Extended Data Fig. 6b). VIDS did not reliably predict graded moral wrongness ratings (discovery cohort: r = 0.15; validation cohort: r = 0.05; replication cohort r = 0.13; generalization cohort r = 0.12). Together, these findings underscore that neural representations of graded moral wrongness engage shared yet distinct neural representations underlying subjective experiences of negative emotions. Sensitivity and specificity of the MJS for intentional versus accidental harms Humans’ moral wrongness judgments are reliably informed by inferring the beliefs and motives of observed agents 56 . Accordingly, across different types of moral transgressions, intentional (versus accidental) harms are robustly judged as more morally wrong 9,20 . Notably, in neurotypical populations (NT), the right TPJ exerts distinct neural activation patterns when judging intentional versus accidental harms, which has been shown to predict individual differences in subjects’ moral judgments 9 . However, clinical populations characterized by difficulties with social interactions (e.g., Autism Spectrum Disorder, ASD) do not show this neural dissociation between intentional and accidental harms, underlining a disproportionate impairment on moral judgment tasks that require recursive mentalizing 9,57-58 . We explored whether the MJS responds more strongly to intentional as opposed to accidental harms, and whether this effect is diminished in individuals with ASD compared to NT populations. To this end, we performed predictions by applying the MJS pattern along with the PINES and VIDS to another independent fMRI dataset 9 where NT and ASD subjects judged the moral wrongness of intentional versus accidental harm scenarios (study 7, n = 39; Supplementary Table 1). We found that only the MJS could significantly predict intentional versus accidental harms in NT individuals (n=25; accuracy 76% (±15% s.e.m.), Cohen’s d = 0.43, sensitivity 73%, specificity 73%, two-sided binomial test p = 0.01). In contrast, PINES (accuracy 60% (±12% s.e.m.), Cohen’s d = 0.09, sensitivity 54%, specificity 54%, two-sided binomial test p = 0.42) as well as VIDS (accuracy 36% (±7% s.e.m.), Cohen’s d = 0.05, sensitivity 53%, specificity 53%, two-sided binomial test p = 0.23) could not significantly predict intentional versus accidental harms. Compellingly, neither the MJS (n=14; accuracy 50% (±13% s.e.m.), Cohen’s d = 0.09, sensitivity 57%, specificity 57%, two-sided binomial test p = 1.00), nor the PINES (accuracy 57% (±15% s.e.m.), Cohen’s d = 0.13, sensitivity 61%, specificity 61%, two-sided binomial test p = 0.79) or VIDS (accuracy 57% (±15% s.e.m.), Cohen’s d = 0.17, sensitivity 59%, specificity 59%, two-sided binomial test p = 0.79) could distinguish between intentional versus accidental harms in individuals with ASD, providing further evidence that clinical populations characterized by difficulties with social interactions rely on fundamentally different neural systems when evaluating moral transgressions as compared to NT individuals. Validating the moral judgment signature in visual scenes and social contexts Finally, we examined whether the moral judgment signature can predict graded moral wrongness judgments in visual scenes and social contexts. Humans frequently encounter morally relevant situations that they directly perceive 59 , even to an extent that moral stimuli “pop-out” in early visual perception 60 . Accordingly, we tested whether the MJS, which was developed solely on text-based moral vignettes, also captures graded moral judgments of real-world visual scenes. To this end, we performed predictions by applying the MJS pattern along with PINES and VIDS to two independent fMRI datasets (study 8, n = 30; study 9, n = 30) acquired while subjects morally judged photographic images sampled representatively from the socio-moral image database 35,43 (Supplementary Methods and Supplementary Table 1). Contrary to the vignette paradigms (studies 1–4) in which stimulus presentation was not decoupled from graded moral judgment ratings to capture subjects’ intuitive moral response, both visual scene tasks presented stimuli separately from the moral judgment period to better decouple the motor and moral judgment response (Extended Data Fig. 8). Furthermore, we deliberately included images depicting morally good actions and modified the graded moral judgment rating scale to range from very moral (1) to neutral (3) to very immoral (5), allowing us to further probe whether the MJS, which was only trained on sociomoral transgressions, can also discriminate between morally right and wrong stimuli. As shown in Fig. 8a,b, the MJS could accurately predict graded moral image ratings in both samples: for individual subjects the average within-subject correlation between predicted and actual morality ratings (5 pairs of scalar values per subject) was r = 0.36 ± 0.11 (standard error (SE), n= 30, study 8) and r = 0.41 ± 11 (standard error (SE), n=30, study 9) and the overall (between- and within-subjects) prediction outcome was 0.14 (study 8, p = .08) and 0.31 (study 9, p = .0001). In addition, the MJS could predict moral versus immoral image ratings with high accuracy in both samples: study 8, accuracy 83% (±15% s.e.m.), Cohen’s d = 0.33, sensitivity 78%, specificity 78%, two-sided binomial test p < 0.001; study 9, accuracy 83% (±15% s.e.m.), Cohen’s d = 0.68, sensitivity 86%, specificity 86%, two-sided binomial test p < 0.001). In contrast, PINES (study 8, accuracy 43% (±8% s.e.m.), Cohen’s d = –0.07, sensitivity 45%, specificity 45%, two-sided binomial test p = 0.58; study 9, accuracy 73% (13% s.e.m.), Cohen’s d = 0.4, sensitivity 78%, specificity 78%, two-sided binomial test p = 0.02) and VIDS (study 8, accuracy 57% (±10% s.e.m.), Cohen’s d = 0.07, sensitivity 55%, specificity 55%, two-sided binomial test p = 0.58; study 9, accuracy 57% (10% s.e.m.), Cohen’s d = 0.15, sensitivity 61%, specificity 61%, two-sided binomial test p = 0.58) could not reliably predict moral versus immoral image ratings across both studies. These results indicated that the MJS captures neural representations of graded moral wrongness that generalize from judgments of text-based vignettes to visual photographic scenes and can discriminate between morally right and wrong stimuli. To further test whether the MJS extends beyond third-party contexts captured by the moral vignettes and images and is also sensitive to second-party transgressions in social contexts, we applied the MJS to another independent fMRI dataset 29 that acquired neural responses during a series of unfair offers in a social exchange (ultimatum game) task (study 10, N = 43; Supplementary Table 1). As shown in Fig. 8b, the MJS did not predict high versus low unfairness (n=43, accuracy 35% (±5% s.e.m.), Cohen’s d = –0.17, sensitivity 44%, specificity 44%, two-sided binomial test p = 0.07). However, replicating previous work 29 , PINES could predict high versus low unfairness (accuracy 70% (±11% s.e.m.), Cohen’s d = 0.73, sensitivity 71%, specificity 71%, two-sided binomial test p = 0.01), as could VIDS (accuracy 74% (±11% s.e.m.), Cohen’s d = 0.81, sensitivity 74%, specificity 74%, two-sided binomial test p < 0.001). Combined, these findings suggest that the MJS is not sensitive to unfair offers in a social context, yet it is relatively specific for moral judgments of visual scenes. Discussion The relative severity of moral transgressions provides a fundamental compass for regulating human behavior, from everyday life 59,61-62 to the courtroom 1 , yet how the human brain represents and computes gradations of moral wrongness remains a longstanding neuroscientific question 2,5,50,63-65 . Evolutionary frameworks postulate an innate neural circuitry which evolved to detect and punish moral transgressors 1,3,66-67 . On the other hand, decades of research demonstrate that the subjective experience of graded moral wrongness for distinct moral violations varies considerably across individuals 13,68-69 , neurodiverse populations 9,58 , and cultures 70-71 . Thus, it remains unclear whether brain representations of graded moral wrongness are instantiated via a shared or idiographic neural code, and which neural systems reliably contribute to graded computations of moral wrongness. Here we combined fMRI with predictive modeling approaches designed to uncover whether neural representations of graded moral wrongness are (a) generalizable across individuals, MRI systems, two different cultures, and neurodiverse populations, (b) common or domain-specific across different kinds of moral norm transgressions, (c) sensitive and specific to moral wrongness versus general negative affect or subjective disgust, and (d) robust across multimodal experimental paradigm variations. In study 1, we applied SVR to identify and evaluate a distributed, whole-brain neural signature (MJS) for predicting graded moral wrongness judgments, with reliable contributions from cortical (e.g., dmPFC, vmPFC, and insula) and subcortical (e.g., amygdala, thalamus, and caudate) regions previously linked to human moral cognition 5,10-12 . By jointly estimating common (general) and domain-specific representations of moral wrongness via PLS-R across eight theory-derived categories of (socio)moral transgressions 34 , we found that computations of graded moral wrongness are encoded in a combination of general (common) and domain-specific representations. Studies 2–4 provided further evidence that brain representations of moral wrongness severity are generalizable across different kinds of moral norm violations, as well as individuals, cohorts, MRI systems, and two different cultures. Moreover, our findings contribute to ongoing debates concerning the role of affect in human moral cognition 31,39-42,72 , demonstrating that neural representations of moral wrongness and general negative emotion (study 5) and subjective disgust (study 6) exhibit shared yet separable representations, with the MJS response mediating the association between the PINES response and graded moral wrongness judgments. Attesting to the potential clinical relevance of our moral judgment biomarker for future translational applications, we found that the MJS could accurately discriminate between moral judgments of intentional versus accidental harms in neurotypical adults, but not in individuals diagnosed with autism spectrum disorder (study 7). Finally, from a biomarker perspective, it is imperative that a neuroaffective signature captures the respective mental process across variations of experimental contexts 73 . Accordingly, we showed that brain representations of graded moral wrongness derived from text-based moral vignettes generalize to graded moral judgment of visual scenes (studies 8–9), albeit not to unfair offers in second-party social interactions probed via the ultimatum game (UG; study 10). It remains a matter of debate to what extent the UG probes moral preferences and decisions as it has been argued that the essence of a moral transgression is an intentional agent causing harm to a suffering moral patient 74-75 ; features that are salient across the vignette scenarios used for developing the MJS. Although the UG likely captures intentional decisions, what is intended is gain to the player and not necessarily intended harm to the recipient. Thus, whether this task induces suffering and is construed as moral is debatable: Given that the worst possible outcome for a recipient in an ultimatum game is to receive nothing, and even putatively “unfair” transfers in the ultimatum game (i.e., < 50%) yield benefits for the recipient, it may be inappropriate to construe the ultimatum game as a moral task paradigm 76 . Concerning the neural basis of moral cognition, it is now commonly understood that moral judgment recruits multiple domain-general systems that are distributed throughout the brain 50,64 . Aligned with this distributed account, in a series of analyses combining both univariate and multivariate analyses, our findings confirm that graded computations of moral wrongness require concerted engagement of brain-wide distributed representations with comparably strong contributions of cortical systems engaged in mentalizing, valuation, and mental imagery such as dorsomedial and ventromedial prefrontal cortex, as well as precuneus and visual cortex 8,9,13,77 , and subcortical regions involved in rapid threat detection, punishment, and avoidance responses (amygdala, thalamus and caudate) 3,78-79 . While no single network reached the predictive performance of the whole-brain MJS model for predicting graded computations of moral wrongness, the visual network exhibited comparably strong contributions, underlining the importance of visual imagery for moral judgment 45-46 and representations of affective stimuli 55 . Consistent with previous predictive models for subjective emotional experiences 27-29 , our analyses revealed that approximately 10,000 voxels that were randomly sampled across the whole brain could lead to high predictive performance for moral wrongness predictions. Together, our results complement a growing body of research from both univariate 6,8,19 and multivariate perspectives 13-16 demonstrating that capturing moral cognition requires integration across multiple distributed neural systems. Analogously, the distributed representation perspective aligns with modular 80-81 and constructionist 82 theories of morality that propose that shared but also distinct distributed functional assemblies integrate to facilitate subjective moral judgment. Moreover, sentimentalist and rationalist perspectives disagree whether affect serves as input, output, or constituent element of moral cognition 6,31,39,41-42,72 . Our results show that computations of moral wrongness exhibit shared yet separable neural representations with established predictive models for the subjective experience of non-specific negative affect or disgust, such that all models engaged subcortical regions and regions implicated in emotional awareness and appraisal, whereas the neural signature of moral wrongness robustly mediated the response of the PINES on subjective moral wrongness ratings but not vice versa. Although the three neural signatures showed a certain extent of similarity in the range of intense emotional experiences (for a similar observation, see also 29 ), the MJS (as well as PINES and VIDS) were more sensitive to the target construct in the low versus high intensity range. However, the MJS outperformed other brain markers in predicting moral judgments in response to both intentional versus accidental harm scenarios and socio-moral images, while it did not track responses to unfair offers. Applications of our moral judgment signature to new studies include (a) characterizing individual differences in brain representations of moral judgement severity related to sociodemographics, hormonal fluctuations, disorders, and subgroups, (b) predicting or monitoring the development and progression of moral wrongness representations over time, (c) establishing the sensitivity and specificity of the MJS with regard to alternative moral judgment types, such as norm and blame judgments 4 as well as additional mental processes known to modulate moral judgment, including empathic care and distress 83 , and (d) decoding spontaneous (implicit) moral condemnation in naturalistic settings 84-85 . Further validation will help refine the use cases and boundary conditions for such applications, with a particular focus on their implications for neuroethics 86-88 . Our study has several limitations. First, the primary datasets used for developing and validating the MJS (studies 1­4) only included subjects from the United States and the Netherlands, raising the question whether the MJS can also predict subjective moral judgments in individuals from cultures that are not Western, Educated, Industrialized, Rich, and Democratic (WEIRD) 70 . Second, although the experimental vignettes used for developing the MJS cover a broad range of different moral scenarios, they feature decontextualized scenarios of “raceless, genderless strangers” 89 from a third-party perspective and hence do not exhaust humans’ rich everyday moral experience. Future studies must consider an array of additional contextual factors that may modulate brain representations of graded moral wrongness, including the (social) identities of agents and victims as well as their intentions, motives, and character 56 . In conclusion, we show that graded computations of moral severity are neurally encoded in a sensitive, specific, and distributed neural signature, which robustly generalizes across individuals, cohorts, experimental variations, and MRI systems. Our resulting brain markers for graded moral wrongness judgments provide robust measures that can serve to understand pathological moral cognition, track the development of brain representations of moral severity over time and across two Western cultures, and advance the precision and generalizability of moral neuroscience. They also lead to further basic and translational research questions. One area for future development concerns the identification of cerebral sources of individual differences in graded moral wrongness representations 90-91 , which may provide potential targets for personalized assessment of aberrant moral cognition 92 and serve as interventions for morally rooted neural polarization 93 . Likewise, illuminating how graded moral wrongness is represented and integrated across modalities, from written scenarios and photographic images to auditory stories and audiovisual movies 34,84-85,94 may reveal how degrees of moral wrongness are hierarchically computed across sensory modalities, helping to explain, for example, where in the brain pathologies interfere with moral cognition 95-96 . Considering that condemnation of what are seen as moral transgressions is a human universal 1,3 this systematic evaluation provides answers to fundamental questions in the cognitive neurosciences, namely that there exists a sensitive and specific neural code for computing graded moral wrongness, which is distributed throughout the brain and robustly tracks moral severity across varied scenarios, individuals, and cultures. Methods Ethics The present study includes 10 datasets (Supplementary Table 1). Among them, study 2 (validation cohort), study 4 (generalization cohort), study 9 (visual scenes 1), and study 10 (visual scenes 2) were new and original experiments designed and implemented by the authors. Informed consent was obtained before each of these experiments. Corresponding experimental protocols were approved by the University of California at Santa Barbara (study 2 and study 8 with the approval numbers 21-17-0123 and 32-25-0079) and the University of Amsterdam (study 4 and study 9 with the approval number FMG-3172). Subjects in study 2 and study 8 were compensated with $50 USD, those in study 4 and study 9 with 25€. The current work also includes secondary analysis of previously acquired experiments based on anonymized data (that is, the remaining six studies; Supplementary Table 1). All subjects in these studies provided informed consent in line with local ethics and institutional review boards. Detailed descriptions of the ethics approval and information on subject compensation are available through the corresponding references. Subjects Discovery cohort . Details of the discovery cohort were reported in a previous study 13 . Briefly, 64 native English speakers were recruited from the University of California Santa Barbara (UCSB) Department of Communication subject pools and from the local Santa Barbara community. Exclusion criteria included a history of systemic or neurological disorders, psychiatric disorders, psychoactive medication or drug use and pregnancy (33 males; mean ± s.d. age 20.78 ± 2.45 years). Validation cohort . 31 healthy volunteers were recruited from the University of California, Santa Barbara. Exclusion criteria included a history of systemic or neurological disorders, psychiatric disorders, psychoactive medication or drug use and pregnancy. One subject was excluded due to excessive head motion, resulting in a final sample of 30 (11 males; mean ± s.d. age 20.46 ± 2.33 years). Replication cohort . Details of the replication cohort were reported in a previous study 14 . Briefly, 30 native English speakers were recruited from the Durham, North Carolina area. Exclusion criteria included a history of psychiatric or neurological disorders. Data from the first three subjects was excluded from analysis because of an error in the experiment script, resulting in a final sample of 27 (14 male, mean ± s.d. age 24.65 ± 4.21 years). Generalization cohort . Healthy volunteers were recruited from the University of Amsterdam’s subject pool and the local Amsterdam community. Exclusion criteria included a history of systemic or neurological disorders, psychiatric disorders, psychoactive medication or drug use and pregnancy. For the fMRI study, we recruited 31 Dutch subjects, and data from 1 subject was excluded due to errors with stimulus presentation, resulting in a final sample of 30 (12 males; mean ± s.d. age 26.4 ± 9.17 years). Stimuli and paradigm used in discovery, validation, replication, and generalization cohorts Subjects in the discovery and replication cohorts were presented with the original set of Moral Foundations Vignettes (MFV) 34 while undergoing fMRI, whereas subjects in the generalization cohort were presented with the validated Dutch version 35 of the MFV (Extended Data Fig. 1). The MFV span 120, one sentence descriptions (14–17 words) detailing the violation of one (and only one) of seven moral foundations: physical care, emotional care, fairness, liberty, loyalty, authority and sanctity. The vignettes also contain a non-moral, social norm transgression category. Each of the eight conditions features 15 vignettes. In the validation cohort, subjects rated a total of 72 vignettes, of which 42 were drawn from the MFV (6 per category). The remaining 30 vignettes were newly created by the authors to depict violations of six moral domains drawn from the Morality as Cooperation (MAC) 97 framework, including family, reciprocity, bravery, property, respect, and group values (5 per category). Vignettes were organized in an event-related design, randomly distributed over three ~8 min functional runs. Subjects viewed one vignette at a time and were instructed to vividly imagine the described scene. While the vignette was on screen, subjects were asked to make a judgment of how morally wrong the action described in the vignette was using an MRI-safe button box (not morally wrong (1) to extremely morally wrong (4)). Subjects had ~8s to read and make judgments of each vignette. Detailed paradigm variations across datasets are reported in Extended Data Fig. 1. MRI data acquisition and preprocessing Discovery and validation cohort fMRI scanning was performed on a 3T Siemens Magnetom Prisma with a Siemens head coil, at the Brain Imaging Centre of the University of California, Santa Barbara. Functional images were taken using a multiband echo-planar gradient sequence (TR, 720 ms; echo time, 37 ms; flip angle, 52°; field of view, 208 mm; acceleration factor, 8). Volumes consisted of 72 interleaved slices (2 mm isotropic) acquired with an angle of ~20° relative to the AC–PC plane, so that the slices are acquired more dorsally near the eyes relative to the back of the brain (in that fashion we were able to acquire the entire brain volume including the cerebellum for every subject). High-resolution T1-weighted whole-brain acquisitions were collected before functional image acquisition (TR, 2,500 ms; echo time, 2.22 ms; flip angle, 7°; field of view, 241 mm; 0.9 mm, isotropic resolution). All data was preprocessed using fMRIprep 98 version 24.1.1 (Supplementary Methods). Replication cohort Scanning was conducted on a research-dedicated 3T GE MR750 scanner with an 8-channel head coil. The scanning session began with a high-resolution T1-weighted structural scan followed by a five-minute resting state scan and three runs of the moral judgment task. Functional scans were collected using a whole-brain spiral-in sequence (TR = 2s, TE = 30 ms, flip angle = 70 o ). Slices were acquired in an interleaved fashion, and subjects' heads were kept in place with cushions to limit head motion. Each run began and ended with 10 seconds (5 TRs) of fixation that were dropped from analysis. The scanning session concluded with five minutes of a localizer task for emotional faces. Data from the resting state and localizer task are not reported here. All data was preprocessed using fMRIprep 98 version 24.1.1 (Supplementary Methods). Generalization cohort Data for this sample was collected on a 3T Philips Achieva scanner, with dStream architecture and 32-channel head coil at the Spinoza center located at the University of Amsterdam’s Roeterseiland Campus. High-resolution T1-weighted whole-brain acquisitions were collected before functional image acquisition (TR, 8.2 ms; echo time, 3.7 ms; flip angle, 8°; field of view, 240 mm). Functional MRI data were acquired using a multiband echo-planar gradient sequence (TR, 720 ms; echo time, 30 ms; flip angle, 55°; field of view, 216 mm; acceleration factor, 4) and all volumes consisted of 44 axial (“ascending”) slices. The task was projected into the scanner and viewed by subjects with a mirror placed above the head coil. The experimental task was programmed using Psychopy (version 2022.1.2) 99 , which was also used to collect behavioral responses on an MRI-safe 4-button box. All data was preprocessed using fMRIprep 98 version 24.1.1 (Supplementary Methods). First-level fMRI analysis used in the discovery, validation, replication, and generalization cohorts Whole-brain univariate GLM (general linear model) analyses were conducted. Each run started with a tail of 11 repetition times (TRs) which were discarded. Thereafter, preprocessed images were spatially smoothed using a Gaussian filter (full-width half-maximum, 8 mm kernel). We conducted two separate subject-level GLM analyses in which the three runs were modeled by separate regressors in the same GLM. To account for residual variance, the temporal derivative of each condition regressor was added in both GLMs as well as a constant regressor for each entire run. The first GLM model was used to obtain beta images for the prediction analysis. In this model we included four separate boxcar regressors time-logged to the presentations of vignettes (7.92s) in each rating (i.e., 1–4), which allowed us to model brain activity in response to each moral judgment level separately. The second GLM modeled the vignette viewing period and the design matrix also included moral wrongness ratings (1−4) reported for each vignette as a parametric modulator for the vignette viewing period. All task regressors were convolved with the canonical haemodynamic response function and a standard high-pass filter (90s cutoff) was applied to exclude low-frequency drifts. Regressors of non-interest (nuisance variables) included (1) six head movement parameters and their squares, their derivatives and squared derivatives (leading to 24 motion-related nuisance regressors in total) and (2) vectors indicating motion outlier timepoints. Multivariate pattern analysis. We applied whole-brain multivariate machine-learning pattern analysis to obtain a pattern of brain activity that best predicted subjects’ self-reported graded moral wrongness ratings. We employed support vector regression (linear kernel with C=1) 27-29 implemented in the NLTools Python package (v.0.4.5) 100 with individual beta maps (one per rating for each subject) as features to predict subjects’ continuous moral wrongness ratings of the grouped vignettes they viewed while undergoing fMRI. We only used data from the discovery cohort to develop the MJS. To evaluate the performance of our algorithm, we used a leave-one-subject-out (LOSO) cross-validation procedure, ensuring that every subject served as both training and testing data. This allowed us to evaluate how a model trained on 63 subjects could predict the rating level associated with each of the four beta maps from the left-out subject 22,25 . To facilitate a robust determination of the predictive accuracy of the neurofunctional signature we employed various metrics including correlation, RMSE, and forced-choice classification accuracy. Specifically, we used overall (between- and within-subjects; 256 pairs in total) and within-subject (4 pairs per subject) Pearson correlations between the cross-validated predictions and the actual ratings to indicate the effect sizes and the RMSE to illustrate overall prediction error. In addition, we assessed classification accuracy of the MJS using a forced-choice test, where signature responses were compared for two conditions tested within the same individual, and the higher was chosen as more morally wrong. We also applied the moral judgment-predictive pattern (trained on the whole discovery cohort) to the validation, replication, and generalization cohorts to obtain a signature response for each map (that is, the dot-product of the MJS weight map and the test image plus the intercept) to assess the prediction performance of the MJS. Within-subject trial-wise prediction Here we tested whether the MJS could predict individual trial-by-trial moral wrongness judgments. To this end we performed a single-trial analysis, which was achieved by specifying a GLM design matrix with separate regressors for each stimulus (vignette). Each task regressor was convolved with the canonical hemodynamic response function. Nuisance regressors and high-pass filters were identical to the above GLM analyses. Next, we calculated the MJS pattern expressions of these single-trial beta maps (i.e., the dot-product of vectorized activation images with the MJS weights), which were finally correlated with the true ratings for each subject separately. For subjects in the discovery cohort we again used the LOSO cross-validation procedure to obtain the MJS response of each single-trial beta map for each subject. PLS-R for common and domain-specific moral wrongness model development We developed common and domain-specific predictive models of moral wrongness ratings from brain activity across the eight vignette conditions using PLS-R 36-37 . PLS-R estimates a set of latent brain components (voxel-wise spatial maps) and a set of latent moral wrongness rating factors that are optimized to be maximally intercorrelated (that is, maximal variance in ratings explained by brain patterns). Compared to standard predictive brain models that typically characterize a single outcome at a time, PLS-R jointly estimates multiple solutions (that is, separate brain patterns for common and domain-specific outcomes) simultaneously, which is why it is capable of predicting multiple (correlated) stimulus types, as is the case with our data 24 . The predictors (brain activity) are stored in the input matrix X and the outcome variables (ratings) are stored in the matrix Y . By an iterative application of a singular value decomposition algorithm, which factorizes (decomposes) the cross-product matrix of the two input matrices, PLS-R finds latent variables, also called component scores, that model X (for example, brain activity) and simultaneously predict Y (for example, ratings). Each run of the singular value decomposition algorithm produces orthogonal latent variables and corresponding regression weights for predictions. By estimating different latent sources, PLS-R can provide improved estimates of common and specific patterns (versus single-outcome models such as SVR), but these are not necessarily fully independent of each other. PLS-R was conducted using ‘PLSRegression’ in sklearn (version 1.6.1) 101 . Predictors ( X ) constituted 1,468 whole-brain activation maps associated with subjects (64) x moral wrongness level (~4) x vignette condition (8), aggregated into an images x voxels matrix (stacked across subjects), and split into training and test sets using leave-one-subject-out cross-validation. The activation maps were derived via univariate GLM analysis in which we modeled the wrongness ratings of each vignette condition against all other event regressors and averaged across trials of each rating level of each vignette condition to obtain ~32 contrast images for each subject. Because not every subject used every wrongness rating option for every vignette condition, the final number of whole-brain activation maps (1,468) was smaller than the theoretical maximum (2,048). We constructed the outcome ( Y ) matrix to include moral wrongness ratings across all vignette conditions ( Y 1 ) as well as domain-specific wrongness ( Y 2 – Y 9 , ratings for each vignette condition separately, with values of 0 for other stimulus types). By setting the Y value of other stimulus types to 0 we constrained each pattern to be domain specific. The linear combination of latent brain factors that explains Y 1 reflects a common model of moral wrongness across vignette conditions. Likewise, brain patterns predictive of Y 2 – Y 9 are models optimized to be selective to violations of physical care, emotional care, fairness, liberty, loyalty, authority, sanctity, and social norms. Each model was then projected into a single predictive spatial map (see also Extended Data Fig. 4). To evaluate the models’ performance, we compared the average model response (predicted wrongness rating) of its target outcome (that is, the four average ratings per vignette condition) with the average model response of off-target outcomes via t tests. In addition, we estimated in each subject the Pearson correlation ( r ) between observed and cross-validated predicted wrongness ratings and tested whether the output of one model predicted the specific type of wrongness it was trained to predict (sensitivity) and not other types (specificity). Determining brain regions associated with and predictive of moral wrongness judgments To determine which brain areas made reliable contributions to graded moral wrongness computations and to threshold voxel weights for interpretation and display, we constructed 10,000 bootstrap samples (with replacement) consisting of paired brain and wrongness data from the discovery cohort and performed SVR on each sample. The z-scores at each voxel were estimated based on the mean and standard error of the bootstrap distributions, and the statistical map was thresholded based on the corresponding P values. The corresponding map was thresholded voxel-wise at q < 0.05 (FDR-corrected). Next, model encoding (‘structure coefficient’) maps were computed for each subject by regressing the SVR model predictions on voxel-wise fMRI activation maps (four maps per person, corresponding to averages for each moral wrongness level). Structure coefficients identify voxels individually associated with the model’s output, mapping individual voxels to the overall multivariate model prediction 38 . The analysis was performed using a standard summary statistics-based mixed-effects GLM, with robust regression at the second level, thresholded at FDR q < 0.05 corrected for multiple comparisons. The core system map for graded moral wrongness computations was derived via a conjunction of the model weight map thresholded at FDR q < 0.05 created during the bootstrap and the model encoding map thresholded at FDR q < 0.05. Next, we performed a one-sample t test on the first-level univariate parametric modulation beta maps to see which brain regions’ activation was associated with moral wrongness ratings. In addition to evaluating the population-level model (MJS), we also probed the consistency of each weight for every voxel in the brain across within-subject multivariate classifiers. To this end we performed a prediction analysis (linear SVR with C = 1) for each subject in the discovery cohort separately using their single-trial data (10-fold cross-validated). We then used a one-sample t test (FDR q < 0.05) on the resulting subject-wise weightmaps to identify which voxels reliably contribute to moral wrongness predictions across subjects. We again created model encoding maps (structure coefficients) for the within-subject models by regressing the within-subject SVR model predictions on voxel-wise fMRI activation maps (~120 maps per person, corresponding to each vignette). This analysis was again performed using a standard summary statistics-based mixed-effects GLM, with robust regression at the second level, thresholded at FDR q < 0.05 corrected for multiple comparisons. The core system map for within-subject graded moral wrongness computations was derived via a conjunction of the model weight map thresholded at FDR q < 0.05 obtained from the one-sample t -test across within-subject weightmaps and the model encoding map thresholded at FDR q < 0.05 obtained from subjects’ trial-wise beta maps. Furthermore, we asked whether moral wrongness computations could be reducible to activations in a single brain region (e.g., dmPFC) or network (e.g., DMN). To examine this hypothesis, we employed whole-brain searchlight (three-voxel radius spheres)—and parcellation (279 cortical and subcortical regions) 102 —based analyses to identify local regions predictive of moral wrongness and compared model performances of local regions with the whole-brain model (i.e., the MJS). In addition, we compared prediction performances of dmPFC (based on a whole-brain parcellation of the coactivation patterns of activations across over 10,000 published studies available in the Neurosynth database; available at https://neurovault.org/images/39711/) and large-scale networks to the whole-brain approach. The networks of interest included seven resting-state functional networks 103 , a subcortical network (including the striatum, thalamus, hippocampus and amygdala) and a ‘consciousness network’ 104 . To reduce potential biases arising from different atlases, we continued to use the modified 279-region version of the Brainnetome Atlas (which also combined Yeo’s seven networks) to extract the nine networks. For these analyses we trained and tested a model for each searchlight sphere, parcellation, brain region or network separately using the discovery data (LOSO cross-validated). Comparing the performance of MJS with the PINES and VIDS Previous studies have developed and evaluated whole-brain emotional decoders for general negative emotion experience (PINES) 25 and subjective disgust experience (VIDS) 29 . To compare the performance of MJS with the PINES and the VIDS, we applied the three decoders to the discovery, validation, replication, generalization, PINES holdout test 25 (study 5, n = 61; see Supplementary Table 1 for details) and VIDS validation 29 (study 6, n = 30; Supplementary Table 1) cohorts and assessed the overall as well as within-subject prediction–outcome correlations between the pattern expressions and the true ratings. Two-alternative forced-choice classification accuracies between the separate moral wrongness judgment levels (and general negative emotion as well as disgust) based on the pattern expressions were further calculated. Spatial similarity between stable decoding maps and a priori ROIs as well as networks of interest River plots were created to illustrate spatial similarity between stable decoding maps derived from bootstrap tests and a priori ROIs previously documented as regions linking to moral judgment, negative affect processes, and disgust. We further depicted spatial similarity between stable decoding maps and seven large-scale cerebral networks, the subcortical and consciousness networks. In line with a recent study 24 , spatial similarity was computed as cosine similarity between the ROI or network and the thresholded MJS, PINES, and VIDS (FDR q < 0.05, retaining positive values) from bootstrap tests. Multi-level two-path mediation analysis To explore the relationship between MJS response, moral wrongness rating and PINES response, multi-level two-path mediation analyses were performed using the Mediation Toolbox, available via MediationToolbox 105 (see also 27-29,106 ). Briefly, the mediation analysis examines whether the observed covariance between the independent/predictor variable ( X ) and the dependent/outcome variable ( Y ) can be explained by the third variable ( M , also mediator). The predictor–mediator relation, mediator–outcome relation and predictor–outcome relation before and after controlling for the mediator are characterized by paths a , b , c and c′ , respectively. Specifically, the total effect of the predictor on the outcome (path c ) is the sum of direct/non-mediation effect (path c′ ) and indirect/mediation effect (the product of the path coefficients of path a and path b , that is, a × b ). A significant mediation effect is obtained when a , b and a × b are all significant. Furthermore, when c′ is significant, M (that is, the mediator) is considered to have a partial mediation effect; otherwise, M plays a full mediation role. In this study, we constructed two multi-level mediation analyses: (1) the trial-by-trial PINES responses were entered as predictors ( X ), moral wrongness ratings were entered as outcomes ( Y ), and the trial-by-trial MJS responses were entered as mediators ( M ); (2) the trial-by-trial MJS responses were entered as predictors ( X ), moral wrongness ratings were entered as outcomes ( Y ), and the trial-by-trial PINES responses were entered as mediators ( M ). To do this, the MJS and PINES responses were calculated by the dot product of the single-trial beta maps with the MJS and PINES patterns, respectively, for each subject across discovery (cross-validated), validation, replication, and generalization cohorts. Bootstrap tests with 10,000 iterations were used to assess the statistical significance of mediation effects. If the bootstrapped 95% CI does not include zero, the effect will be considered to be significant ( P < 0.05). Validation in intentional vs. accidental harm scenarios and neurodiverse populations We examined whether the MJS can discriminate between intentional versus accidental harms in both neurotypical (NT) and individuals diagnosed with autism spectrum disorder (ASD) by applying the MJS, PINES, and VIDS pattern to another independent fMRI dataset 9 during which subjects read 60 vignettes in the second-person point of view that either portrayed intentional or accidental harm violations (n = 39; NT = 25, ASD = 14; Supplementary Methods and Supplementary Table 1). For each brain model (MJS, PINES, and VIDS), we computed the classification accuracy between intentional and accidental harms from receiver operating characteristic curves using forced-choice classification (average of all intentional harms versus average of all accidental harms). P values were calculated using a two-sided independent binomial test. Validation in visual scenes To test whether the MJS—developed on text-based vignettes—can generalize to visual scenes, we designed and implemented two new fMRI experiments (study 8, n = 30; and study 9, n = 30; Supplementary Methods and Supplementary Table 1) employing photographic stimuli 35,43 (design based on previous similar studies 27-29 ). Next, we applied the MJS—as well as the PINES and VIDS—to the visual scenes fMRI data. Specifically, we calculated Pearson correlation coefficients across actual and predicted moral judgments as well as forced-choice classification accuracies between moral and immoral image ratings based on the pattern responses. Validation in the social context First, to examine whether the MJS—developed during third-party moral wrongness judgments of concrete moral transgressions—could be extended into a second-party social context frequently linked to socio-moral disgust (for example, unfairness 29,107-108 ), we applied the MJS pattern to another independent fMRI dataset during which subjects were confronted with a series of unfair offers in an ultimatum game task (study 10, n = 43; Supplementary Methods and Supplementary Table 1). Specifically, we calculated forced-choice classification accuracies between high (unfairness level 5) and low (unfairness level 1) unfairness based on the pattern responses. Declarations Data availability fMRI data (beta maps) used to train and validate the signature are available via figshare at https://figshare.com/articles/dataset/Discovery_dataset_mjs/29423726?file=55720082 (ref. 109) (study 1); https://figshare.com/articles/dataset/Validation_dataset_mjs/29423789?file=55721972 (ref. 110) (study 2); https://figshare.com/articles/dataset/Replication_dataset_mjs/29423966?file=55724255 (ref. 111) (study 3); and https://figshare.com/articles/dataset/Generalization_dataset_mjs/29423981?file=55724291 (ref. 112) (study 4). The data of study 5 are from a previous study 25 and are available via NeuroVault at https://neurovault.org/collections/1964 (ref. 113). The data of study 6 are from a previous study 29 and are available at figshare https://figshare.com/articles/dataset/validation_dataset_disgust/22841117 (ref. 114). The data of study 7 are from a previous study 9 and are available via OpenNeuro at https://openneuro.org/datasets/ds000212/versions/1.0.0 (ref. 115). The data of study 8 are available via figshare at https://figshare.com/articles/dataset/Visualscenes1_dataset_mjs/29424164?file=55725344 (ref. 116) and the data of study 9 are available via figshare at ataset/Visualscenes2_dataset_mjs/29424176?file=55725368 (ref. 117). The data from the ultimatum game (study 10) were provided by the authors of a previous study (ref. 118). The MJS and the thresholded statistical maps are available via figshare at https://figshare.com/articles/dataset/Brain_models_and_maps/29424206?file=55725515 (ref. 119). Code availability The custom code that supports the findings of this study is available at https://github.com/Moral-Computing-Lab/hopp_mjs. Data were analyzed using the NLTools (v.0.4.5) 100 Python package available at https://github.com/cosanlab/nltools and CanlabCore Tools 120 available at https://github.com/canlab/CanlabCore. Acknowledgements We thank Ting Xu for sharing the ultimatum game data (study 10) with us. The acquisition of the discovery dataset was funded by a grant from the Army Research Lab to R.W. (W911NF-15-2-0115); the validation data was funded by a grant from the John Templeton Foundation to R.W. (W911NF-15-2-0115); the replication dataset was partially supported by the Duke Institute for Brain Sciences incubator grant awarded to W.S.A. and additional support from the Duke Institute for Brain Sciences; the generalization dataset received funding from the Amsterdam School of Communication Research awarded to F.R.H. (ASCoR-u-2023-Hopp; ASCoR-u-2024-Hopp). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Author contributions F.R.H. and R.W. conceived of this work. F.R.H., R.W., and S.Y. undertook data curation. R.W. conducted the investigation for the discovery dataset, S.Y. and R.W. conducted the investigation for the validation dataset, W.S.A. conducted the investigation for the replication dataset, and F.R.H. conducted the investigation for the generalization dataset. F.R.H. did the formal analysis in consultation with R.W., produced the visualizations, and wrote the original paper. S.Y., W.S.A., and R.W. were responsible for validation. S.Y., W.S.A., and R.W. reviewed and edited the final paper. F.R.H. and R.W. undertook project administration and funding acquisition. References Mikhail, J. (2007). Universal moral grammar: Theory, evidence and the future. Trends in Cognitive Sciences , 11 (4), 143-152. Powell, D., & Horne, Z. (2017). Moral severity is represented as a domain-general magnitude. Experimental Psychology . Buckholtz, J. W., & Marois, R. (2012). The roots of modern justice: cognitive and neural foundations of social norms and their enforcement. Nature Neuroscience , 15 (5), 655-661. Malle, B. F. (2021). Moral judgments. Annual Review of Psychology , 72 , 293-318. Moll, J., Zahn, R., de Oliveira-Souza, R., Krueger, F., & Grafman, J. (2005). The neural basis of human moral cognition. Nature Reviews Neuroscience , 6(10), 799-809. Greene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M., & Cohen, J. D. (2001). An fMRI investigation of emotional engagement in moral judgment. Science , 293 (5537), 2105-2 Crockett, M. J., Siegel, J. Z., Kurth-Nelson, Z., Dayan, P., & Dolan, R. J. (2017). Moral transgressions corrupt neural representations of value. Nature Neuroscience , 20 (6), 879-885. Shenhav, A., & Greene, J. D. (2010). Moral judgments recruit domain-general valuation mechanisms to integrate representations of probability and magnitude. Neuron , 67 (4), 667-677. Koster-Hale, J., Saxe, R., Dungan, J., & Young, L. L. (2013). Decoding moral judgments from neural representations of intentions. Proceedings of the National Academy of Sciences , 110 (14), 5648-5653. Eres, R., Louis, W. R., & Molenberghs, P. (2018). Common and distinct neural networks involved in fMRI studies investigating morality: an ALE meta-analysis. Social Neuroscience , 13 (4), 384-398. Han, H. (2017). Neural correlates of moral sensitivity and moral judgment associated with brain circuitries of selfhood: A meta-analysis. Journal of Moral Education , 46 (2), 97-113. Sevinc, G., & Spreng, R. N. (2014). Contextual and perceptual brain processes underlying moral cognition: a quantitative meta-analysis of moral reasoning and moral emotions. PloS one , 9 (2), e87427. Hopp, F. R., Amir, O., Fisher, J., Grafton, S., Sinnott-Armstrong, W., & Weber, R. (2023). Moral foundations elicit shared and dissociable cortical activation modulated by political ideology. Nature Human Behavior . Khoudary, A., Hanna, E., O’Neill, K., Iyengar, V., Clifford, S., Cabeza, R., ... & Sinnott-Armstrong, W. (2022). A functional neuroimaging investigation of Moral Foundations Theory. Social Neuroscience , 17 (6), 491-507. Tsoi, L., Dungan, J. A., Chakroff, A., & Young, L. L. (2018). Neural substrates for moral judgments of psychological versus physical harm. Social Cognitive and Affective Neuroscience , 13 (5), 460-470. Wasserman, E. A., Chakroff, A., Saxe, R., & Young, L. (2017). Illuminating the conceptual structure of the space of moral violations with searchlight representational similarity analysis. NeuroImage , 159 , 371-387. Peelen, M., & Downing, P. (2023). Testing cognitive theories using multivariate pattern analysis of neuroimaging data. Nature Human Behavior . Moll, J., Eslinger, P. J., & Oliveira-Souza, R. D. (2001). Frontopolar and anterior temporal cortex activation in a moral judgment task: preliminary functional MRI results in normal subjects. Arquivos de neuro-psiquiatria , 59 , 657-664. Parkinson, C., Sinnott-Armstrong, W., Koralus, P. E., Mendelovici, A., McGeer, V., & Wheatley, T. (2011). Is morality unified? Evidence that distinct neural systems underlie moral judgments of harm, dishonesty, and disgust. Journal of Cognitive Neuroscience , 23 (10), 3162-3180. Schaich Borg, J., Hynes, C., Van Horn, J., Grafton, S., & Sinnott-Armstrong, W. (2006). Consequences, action, and intention as factors in moral judgments: An fMRI investigation. Journal of Cognitive Neuroscience , 18 (5), 803-817. Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data?. Trends in Cognitive Sciences , 10 (2), 59-63. Wager, T. D., Atlas, L. Y., Lindquist, M. A., Roy, M., Woo, C. W., & Kross, E. (2013). An fMRI-based neurologic signature of physical pain. New England Journal of Medicine , 368 (15), 1388-1397. Coll, M. P., Slimani, H., Woo, C. W., Wager, T. D., Rainville, P., Vachon-Presseau, É., & Roy, M. (2022). The neural signature of the decision value of future pain. Proceedings of the National Academy of Sciences , 119 (23), e2119931119. Čeko, M., Kragel, P. A., Woo, C. W., López-Solà, M., & Wager, T. D. (2022). Common and stimulus-type-specific brain representations of negative affect. Nature Neuroscience , 25 (6), 760-770. Chang, L. J., Gianaros, P. J., Manuck, S. B., Krishnan, A., & Wager, T. D. (2015). A sensitive and specific neural signature for picture-induced negative affect. PLoS biology , 13 (6), e1002180. Kragel, P. A., Treadway, M. T., Admon, R., Pizzagalli, D. A., & Hahn, E. C. (2023). A mesocorticolimbic signature of pleasure in the human brain. Nature Human Behaviour , 7 (8), 1332-1343. Zhou, F., Zhao, W., Qi, Z., Geng, Y., Yao, S., Kendrick, K. M., ... & Becker, B. (2021). A distributed fMRI-based signature for the subjective experience of fear. Nature Communications , 12 (1), 6643. Liu, X., Jiao, G., Zhou, F., Kendrick, K. M., Yao, D., Gong, Q., ... & Becker, B. (2024). A neural signature for the subjective experience of threat anticipation under uncertainty. Nature Communications , 15 (1), 1544. Gan, X., Zhou, F., Xu, T., Liu, X., Zhang, R., Zheng, Z., ... & Becker, B. (2024). A neurofunctional signature of subjective disgust generalizes to oral distaste and socio-moral contexts. Nature Human Behaviour , 1-20. Kragel, P. A., Koban, L., Barrett, L. F., & Wager, T. D. (2018). Representation, pattern information, and brain signatures: from neurons to neuroimaging. Neuron , 99 (2), 273. Haidt, J. (2001). The emotional dog and its rational tail: a social intuitionist approach to moral judgment. Psychological Review , 108 (4), 814. Kragel, P. A., Kano, M., Van Oudenhove, L., Ly, H. G., Dupont, P., Rubio, A., ... & Wager, T. D. (2018). Generalizable representations of pain, cognitive control, and negative emotion in medial frontal cortex. Nature Neuroscience , 21 (2), 283-289. Woo, C. W., Chang, L. J., Lindquist, M. A., & Wager, T. D. (2017). Building better biomarkers: brain models in translational neuroimaging. Nature Neuroscience , 20 (3), 365-377. Clifford, S., Iyengar, V., Cabeza, R., & Sinnott-Armstrong, W. (2015). Moral foundations vignettes: A standardized stimulus database of scenarios based on moral foundations theory. Behavior Research Methods , 47 (4), 1178–1198. Hopp, F. R., Jargow, B., Kouwen, E., & Bakker, B. N. (2024). The Dutch moral foundations stimulus database: A validation of moral foundations vignettes and sociomoral images. Judgment and Decision Making . McIntosh, A. R., & Lobaugh, N. J. (2004). Partial least squares analysis of neuroimaging data: applications and advances. Neuroimage , 23 , S250-S263. Wold, S., Sjöström, M., & Eriksson, L. (2001). PLS-regression: a basic tool of chemometrics. Chemometrics and intelligent laboratory systems , 58 (2), 109-130. Haufe, S., Meinecke, F., Görgen, K., Dähne, S., Haynes, J. D., Blankertz, B., & Bießmann, F. (2014). On the interpretation of weight vectors of linear models in multivariate neuroimaging. Neuroimage , 87 , 96-110. Damasio, A.R., 1994. Descartes’ Error: Emotion, Reason, and The Human Brain. Avon, New York Nichols, S. (2004). Sentimental rules: On the natural foundations of moral judgment . Oxford University Press. Prinz, J. (2007). The emotional construction of morals . Oxford University Press. Prinz, J. (2016). Sentimentalism and the moral brain. Moral brains: The neuroscience of morality , 45-73. Crone, D. L., Bode, S., Murawski, C., & Laham, S. M. (2018). The Socio-Moral Image Database (SMID): A novel stimulus set for the study of social, moral and affective processes. PloS one , 13 (1), e0190954. Lancaster, J. L., Woldorff, M. G., Parsons, L. M., Liotti, M., Freitas, C. S., Rainey, L., ... & Fox, P. T. (2000). Automated Talairach atlas labels for functional brain mapping. Human brain mapping , 10 (3), 120-131. Amit, E., & Greene, J. D. (2012). You see, the ends don’t justify the means: Visual imagery and moral judgment. Psychological Science , 23 (8), 861-868. De Freitas, J., & Alvarez, G. A. (2018). Your visual system provides all the information you need to make moral judgments about generic visual events. Cognition , 178 , 133-146. Schaich Borg, J., Sinnott-Armstrong, W., Calhoun, V. D., & Kiehl, K. A. (2011). Neural basis of moral verdict and moral deliberation. Social Neuroscience , 6 (4), 398-413. Decety, J., & Cacioppo, S. (2012). The speed of morality: a high-density electrical neuroimaging study. Journal of Neurophysiology , 108 (11), 3068-3072. Rissman, J., Gazzaley, A., & D'Esposito, M. (2004). Measuring functional connectivity during distinct stages of a cognitive task. Neuroimage , 23 (2), 752-763. Young, L., & Dungan, J. (2012). Where in the brain is morality? Everywhere and maybe nowhere. Social Neuroscience , 7 (1), 1-10. Greene, J. D. (2007). Why are VMPFC patients more utilitarian? A dual-process theory of moral judgment explains. Trends in Cognitive Sciences , 11 (8), 322-323. Chiong, W., Wilson, S. M., D’Esposito, M., Kayser, A. S., Grossman, S. N., Poorzand, P., ... & Rankin, K. P. (2013). The salience network causally influences default mode network activity during moral reasoning. Brain , 136 (6), 1929-1941. Reniers, R. L., Corcoran, R., Völlm, B. A., Mashru, A., Howard, R., & Liddle, P. F. (2012). Moral decision-making, ToM, empathy and the default mode network. Biological Psychology , 90 (3), 202-210. Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., & Wager, T. D. (2011). Large-scale automated synthesis of human functional neuroimaging data. Nature methods , 8 (8), 665-670. Kragel, P. A., Reddan, M. C., LaBar, K. S., & Wager, T. D. (2019). Emotion schemas are embedded in the human visual system. Science Advances , 5 (7), eaaw4358. Carlson, R. W., Bigman, Y. E., Gray, K., Ferguson, M. J., & Crockett, M. J. (2022). How inferred motives shape moral judgements. Nature Reviews Psychology , 1 (8), 468-478. Grant C.M., Boucher J., Riggs K.J., Grayson A. (2005). Moral understanding in children with autism. Autism 9(3):317–331 Moran, J. M., Young, L. L., Saxe, R., Lee, S. M., O'Young, D., Mavros, P. L., & Gabrieli, J. D. (2011). Impaired theory of mind for moral judgment in high-functioning autism. Proceedings of the National Academy of Sciences , 108 (7), 2688-2692. Hofmann, W., Wisneski, D. C., Brandt, M. J., & Skitka, L. J. (2014). Morality in everyday life. Science , 345 (6202), 1340-1343. Gantman, A. P., & Van Bavel, J. J. (2015). Moral perception. Trends in Cognitive Sciences , 19 (11), 631-633. Elmore, J. D., Lewis, J. A., Himmelberger, Z. M., & Sherwood, J. A. (2024). Everyday moral transgressions (EMTs): Investigating the morality of everyday behaviors. Current Psychology , 43 (11), 10484-10493. Yudkin, D. A., Goodwin, G. P., Reece, A., Gray, K., & Bhatia, S. (2025). A large-scale investigation of everyday moral dilemmas. PNAS Nexus , 4 (5), pgaf119. Dolan, R. J. (1999). On the neurology of morals. Nature Neuroscience , 2 (11), 927-929. Greene, J., & Haidt, J. (2002). How (and where) does moral judgment work?. Trends in Cognitive Sciences , 6 (12), 517-523. Greene, J. D. (2009). The cognitive neuroscience of moral judgment. The cognitive neurosciences , 4 , 1-48. Krebs, D. L. (2008). Morality: An evolutionary account. Perspectives on Psychological Science , 3 (3), 149-172. Moll, J., de Oliveira-Souza, R., & Eslinger, P. J. (2003). Morals and the human brain: a working model. Neuroreport , 14 (3), 299-305. Moore, A. B., Stevens, J., & Conway, A. R. (2011). Individual differences in sensitivity to reward and punishment predict moral judgment. Personality and Individual Differences , 50 (5), 621-625. van Baar, J. M., Chang, L. J., & Sanfey, A. G. (2019). The computational and neural substrates of moral strategies in social decision-making. Nature Communications , 10 (1), 1483. Atari, M., Haidt, J., Graham, J., Koleva, S., Stevens, S. T., & Dehghani, M. (2023). Morality beyond the WEIRD: How the nomological network of morality varies across cultures. Journal of Personality and Social Psychology , 125 (5), 1157. Awad, E., Dsouza, S., Kim, R., Schulz, J., Henrich, J., Shariff, A., ... & Rahwan, I. (2018). The moral machine experiment. Nature , 563 (7729), 59-64. Horberg, E. J., Oveis, C., & Keltner, D. (2011). Emotions as moral amplifiers: An appraisal tendency approach to the influences of distinct emotions upon moral judgment. Emotion Review , 3 (3), 237-244. Kiar, G., Mumford, J. A., Xu, T., Vogelstein, J. T., Glatard, T., & Milham, M. P. (2024). Why experimental variation in neuroimaging should be embraced. Nature Communications , 15 (1), 9411. Gray, K., Waytz, A., & Young, L. (2012). The moral dyad: A fundamental template unifying moral judgment. Psychological Inquiry, 23(2), 206–215. http://doi.org/10.1080/10478 40X.2012.686247 Gray, K.,Young, L., & Waytz, A. (2012). Mind perception is the essence of morality. Psychological Inquiry , 23(2), 101–124. http://doi.org/10.1080/1047840X.2012.651387 Crockett, M. J. (2016). Computational modeling of moral decisions. In The social psychology of morality (pp. 71-90). Routledge. Zoh, Y., Chang, S. W., & Crockett, M. J. (2022). The prefrontal cortex and (uniquely) human cooperation: a comparative perspective. Neuropsychopharmacology , 47 (1), 119-133. Krueger, F., & Hoffman, M. (2016). The emerging neuroscience of third-party punishment. Trends in Neurosciences , 39 (8), 499-501. Seymour, B., Singer, T., & Dolan, R. (2007). The neurobiology of punishment. Nature Reviews Neuroscience , 8 (4), 300-311. Haidt, J., & Joseph, C. (2004). Intuitive ethics: How innately prepared intuitions generate culturally variable virtues. Daedalus , 133 (4), 55-66. Sinnott-Armstrong, W., & Wheatley, T. (2012). The disunity of morality and why it matters to philosophy. The Monist , 95 (3), 355-377. Cameron, C. D., Lindquist, K. A., & Gray, K. (2015). A constructionist review of morality and emotions: No evidence for specific links between moral content and discrete emotions. Personality and Social Psychology Review , 19 (4), 371-394. Ashar, Y. K., Andrews-Hanna, J. R., Dimidjian, S., & Wager, T. D. (2017). Empathic care and distress: predictive brain markers and dissociable brain systems. Neuron , 94 (6), 1263-1273. Weber, R., Hopp, F. R., Eden, A., Fisher, J. T., & Lee, H. E. (2024). Vicarious punishment of moral violations in naturalistic drama narratives predicts cortical synchronization. NeuroImage , 292 , 120613. Kaplan, J. T., Gimbel, S. I., Dehghani, M., Immordino-Yang, M. H., Sagae, K., Wong, J. D., ... & Damasio, A. (2017). Processing narratives concerning protected values: A cross-cultural investigation of neural correlates. Cerebral Cortex , 27 (2), 1428-1438. Gazzaniga, M. S. (2005). The ethical brain . Dana press. Illes, J., & Bird, S. J. (2006). Neuroethics: a modern context for ethics in neuroscience. Trends in Neurosciences , 29 (9), 511-517. Roskies, A. (2002). Neuroethics for the new millenium. Neuron , 35 (1), 21-23. Hester, N., & Gray, K. (2020). The moral psychology of raceless, genderless strangers. Perspectives on Psychological Science , 15 (2), 216-230. Kohoutová, L., Atlas, L. Y., Büchel, C., Buhle, J. T., Geuter, S., Jepma, M., ... & Woo, C. W. (2022). Individual variability in brain representations of pain. Nature Neuroscience , 25 (6), 749-759. Lewis, G. J., Kanai, R., Bates, T. C., & Rees, G. (2012). Moral values are associated with individual differences in regional brain volume. Journal of Cognitive Neuroscience , 24 (8), 1657-1663. Anderson, S. W., & Bechara, A. & Damasio, A., R.(1999) Impairment of social and moral behavior related to early damage in human prefrontal cortex. Nature Neuroscience , 2 (11), 1032-1037. Leong, Y. C., Chen, J., Willer, R., & Zaki, J. (2020). Conservative and liberal attitudes drive polarized neural responses to political content. Proceedings of the National Academy of Sciences , 117 (44), 27731-27739. McCurrie, C. H., Crone, D. L., Bigelow, F., & Laham, S. M. (2018). Moral and Affective Film Set (MAAFS): A normed moral video database. PloS one , 13 (11), e0206604. Koenigs, M., Young, L., Adolphs, R., Tranel, D., Cushman, F., Hauser, M., & Damasio, A. (2007). Damage to the prefrontal cortex increases utilitarian moral judgements. Nature , 446 (7138), 908-911. Young, L., Bechara, A., Tranel, D., Damasio, H., Hauser, M., & Damasio, A. (2010). Damage to ventromedial prefrontal cortex impairs judgment of harmful intent. Neuron , 65 (6), 845-851. Curry, O. S., Chesters, M. J., & Van Lissa, C. J. (2019). Mapping morality with a compass: Testing the theory of ‘morality-as-cooperation’with a new questionnaire. Journal of Research in Personality , 78 , 106-124. Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, A. I., Erramuzpe, A., ... & Gorgolewski, K. J. (2019). fMRIPrep: a robust preprocessing pipeline for functional MRI. Nature Methods , 16 (1), 111-116. Peirce, J. W. (2007). PsychoPy—psychophysics software in Python. Journal of neuroscience methods , 162 (1-2), 8-13. Chang, L., Jolly, E., Cheong, J. H., Burnashev, A. & Chen, A. cosanlab/nltools: 0.3.11 Zenodo https://zenodo.org/ record/2229813(2018). Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... & Duchesnay, É. (2011). Scikit-learn: Machine learning in Python. the Journal of machine Learning research , 12 , 2825-2830. Fan, L., Li, H., Zhuo, J., Zhang, Y., Wang, J., Chen, L., ... & Jiang, T. (2016). The human brainnetome atlas: a new brain atlas based on connectional architecture. Cerebral Cortex , 26 (8), 3508-3526. Yeo, B. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., ... & Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology , 106 (3), 1125-1165. LeDoux, J. E., & Pine, D. S. (2016). Using neuroscience to help understand fear and anxiety: a two-system framework. American journal of psychiatry , 173 (11), 1083-1093. MediationToolbox. GitHub https://github.com/canlab/ MediationToolbox (2014). Woo, C. W., Schmidt, L., Krishnan, A., Jepma, M., Roy, M., Lindquist, M. A., ... & Wager, T. D. (2017). Quantifying cerebral contributions to pain beyond nociception. Nature Communications , 8 (1), 14211. Chapman, H. A., Kim, D. A., Susskind, J. M., & Anderson, A. K. (2009). In bad taste: Evidence for the oral origins of moral disgust. Science , 323 (5918), 1222-1226. Sanfey, A. G., Rilling, J. K., Aronson, J. A., Nystrom, L. E., & Cohen, J. D. (2003). The neural basis of economic decision-making in the ultimatum game. Science , 300 (5626), 1755-1758. Moral judgment discovery cohort dataset. figshare https://figshare.com/articles/dataset/Discovery_dataset_mjs/29423726?file=55720082 Moral judgment validation cohort dataset. figshare https://figshare.com/articles/dataset/Validation_dataset_mjs/29423789?file=55721972 Moral judgment replication cohort dataset. figshare https://figshare.com/articles/dataset/Replication_dataset_mjs/29423966?file=55724255 Moral judgment generalization cohort dataset. figshare https://figshare.com/articles/dataset/Generalization_dataset_mjs/29423981?file=55724291 PINES holdout test dataset. NeuroVault https://neurovault.org/collections/1964 (2015). Disgust validation cohort dataset. figshare https://figshare.com/ articles/dataset/validation_dataset_disgust/22841117 (2023). Young, L., Chakroff, A., Wasserman, E., Saxe, R., Dungan, J., Brown, A., & Koster-Hale, J. (2019). Moral judgments of intentional and accidental moral violations across Harm and Purity domains. OpenNeuro.[Dataset] , 10 . Moral judgment visual scenes 1 dataset figshare https://figshare.com/articles/dataset/Visualscenes1_dataset_mjs/29424164?file=55725344 Moral judgment visual scenes 2 dataset figshare https://figshare.com/articles/dataset/Visualscenes2_dataset_mjs/29424176?file=55725368 Xu, T., Zhang, L., Zhou, F., Fu, K., Gan, X., Chen, Z., ... & Becker, B. (2025). Distinct neural computations scale the violation of expected reward and emotion in social transgressions. Communications Biology , 8 (1), 106. The MJS and thresholded statistical maps. figshare https://figshare.com/articles/dataset/Brain_models_and_maps/29424206?file=55725515 Cognitive and Affective Neuroscience Laboratory, CANLab. GitHub https://github.com/canlab (2014). Footnotes Recent theoretical advancements separate fairness into equality and proportionality 70 , whereas we examined fairness in its traditional, holistic form. Additional Declarations There is NO Competing Interest. Supplementary Files suppinfofinal.docx Supplemental Materials ExtendedDataFig.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7935407","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":548770386,"identity":"e45321ec-fe64-48b6-a0bc-4762bd145b13","order_by":0,"name":"Frederic Hopp","email":"","orcid":"https://orcid.org/0000-0002-0866-064X","institution":"Leibniz Institute for Psychology","correspondingAuthor":false,"prefix":"","firstName":"Frederic","middleName":"","lastName":"Hopp","suffix":""},{"id":548770387,"identity":"96564b1e-a78d-4077-b990-3b35a5a360cb","order_by":1,"name":"Sungbin Youk","email":"","orcid":"","institution":"Ewha Womans University","correspondingAuthor":false,"prefix":"","firstName":"Sungbin","middleName":"","lastName":"Youk","suffix":""},{"id":548770388,"identity":"5c0b1d49-f9df-417a-92fd-bbde9bc68a0f","order_by":2,"name":"Walter Sinnott-Armstrong","email":"","orcid":"https://orcid.org/0000-0003-2579-9966","institution":"Duke University","correspondingAuthor":false,"prefix":"","firstName":"Walter","middleName":"","lastName":"Sinnott-Armstrong","suffix":""},{"id":548770385,"identity":"c128579e-42a2-41e7-984e-c33087c61e47","order_by":3,"name":"René Weber","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxklEQVRIiWNgGAWjYNCCCjjrABAnEKPlDMlaGNtI0WLe3nvwMe88OzndaQfYJH78ucPAz55jgFeLzJlzyca825KNzW4nsEn2tj1jkOx5g1+LhESOmXTutgOJ24BapBkbDjMY3CBgi4T8G/PfuXOgWhj+HGawJ6hFgseMObcBpoUNaIsEIS08OcbSf46B/JLYbAn0C4/EmWcF+LWwnzH8OKPGTs7sdvLBG8AQk+NvT96AVwsSYGwAkTzEKh8Fo2AUjIJRgAcAABRsRPWdiyl7AAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-8247-7341","institution":"University of California Santa Barbara","correspondingAuthor":true,"prefix":"","firstName":"René","middleName":"","lastName":"Weber","suffix":""}],"badges":[],"createdAt":"2025-10-23 22:25:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7935407/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7935407/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96711033,"identity":"52862434-c2fa-449a-92bf-1e5bfaa39f42","added_by":"auto","created_at":"2025-11-25 10:11:32","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11347090,"visible":true,"origin":"","legend":"","description":"","filename":"moralsignaturefinal.docx","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/a84cb78b7d7723946e63c880.docx"},{"id":96710658,"identity":"c68854e0-3bdc-407d-88f2-e28daaf965af","added_by":"auto","created_at":"2025-11-25 10:11:02","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5819,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS2589726T.json","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/d475891ae7a6e214d98b66c7.json"},{"id":96710689,"identity":"4f59cb3d-9c49-44d6-aa31-3535e3a8cd95","added_by":"auto","created_at":"2025-11-25 10:11:04","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4278257,"visible":true,"origin":"","legend":"","description":"","filename":"suppinfofinal.docx","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/a0008888705ad1ca78428555.docx"},{"id":96689595,"identity":"5abf9822-e1ec-43b0-94b7-a65e23adfabe","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":282101,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS2589726T0enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/5137c0769350bd6091ef283d.xml"},{"id":96710944,"identity":"f6c3bc88-0061-4f86-a291-92a5ec8b1ec5","added_by":"auto","created_at":"2025-11-25 10:11:26","extension":"eps","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/008938d824e64d51b3dd8a3f.eps"},{"id":96710680,"identity":"c03e83a7-20ef-44db-9752-d33378544fbe","added_by":"auto","created_at":"2025-11-25 10:11:03","extension":"eps","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/921e4eb091d808bd875f54df.eps"},{"id":96689548,"identity":"901bc3c3-adb6-4d05-af36-051091df1e6d","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"eps","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/a5b69d2801e118b14e382fda.eps"},{"id":96710088,"identity":"57bd8f74-e73b-4a7f-a7fd-69ae04edb235","added_by":"auto","created_at":"2025-11-25 10:10:03","extension":"eps","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/7e7c29fddea5df1420bf3889.eps"},{"id":96689552,"identity":"cfe3dbdc-9197-435f-950a-6140bb917604","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"eps","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/89a78f62d6f7a30d354f1311.eps"},{"id":96710860,"identity":"6e6277a8-6100-47be-9e9d-e2c8db74e270","added_by":"auto","created_at":"2025-11-25 10:11:17","extension":"eps","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/f477990f2321a0f446f6ca9a.eps"},{"id":96710947,"identity":"87ce9f60-29ff-40eb-9d8d-7f60c121212e","added_by":"auto","created_at":"2025-11-25 10:11:26","extension":"eps","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/a2746d445338a672f1a7c72c.eps"},{"id":96710126,"identity":"8d469f08-9f78-49ae-9b6b-2f026f79e899","added_by":"auto","created_at":"2025-11-25 10:10:09","extension":"eps","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":438,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/117317ad585e23085896864f.eps"},{"id":96689566,"identity":"f8b71818-1156-42cb-9748-9e98a6f4cfae","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":787859,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/a253ef1e76b2ad465b242f1f.jpeg"},{"id":96711608,"identity":"551abbb9-e066-4ed9-b44e-3bab922c7fe1","added_by":"auto","created_at":"2025-11-25 10:12:18","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":739406,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/03832ffce937819b64bed1c1.png"},{"id":96711517,"identity":"264caf33-bca9-46e1-97cb-ef66fd135fdd","added_by":"auto","created_at":"2025-11-25 10:12:08","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":222041,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/c2e6f38a2682747aa8596cbc.png"},{"id":96689557,"identity":"1e2c5032-cc02-4451-af25-0b74af49e647","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":887598,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/be2a2af55a34d105a2dbf667.png"},{"id":96689564,"identity":"b49d1fcd-4d48-4527-ba0e-65a13481e327","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1281893,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/045bc0eb581c4e8767c2ef78.png"},{"id":96689562,"identity":"10492a6d-900a-4576-88c9-5b80c4759794","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":820363,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/41d58a34aac7f4bda8814811.png"},{"id":96689571,"identity":"dd30829a-7662-41ae-b7b5-99bed3dc655a","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":480958,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/0ddf962be0a996b974c20750.png"},{"id":96711038,"identity":"8472dbc9-0a48-41ff-99e6-72dc2b1daee9","added_by":"auto","created_at":"2025-11-25 10:11:32","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":117898,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/de5372eb4caaae322a1d9cbd.png"},{"id":96711040,"identity":"29a40e55-a6e3-4d7d-8d0d-511232640e3a","added_by":"auto","created_at":"2025-11-25 10:11:32","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1453678,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/8a88ced41e254f111e198829.png"},{"id":96710880,"identity":"8cb1987b-85c3-4c22-b9f0-52e32ec1d1c8","added_by":"auto","created_at":"2025-11-25 10:11:18","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":844244,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/db7f8a6e674750c6dea22b91.png"},{"id":96710926,"identity":"b57649cb-1d18-4236-8c1a-679fed7d739b","added_by":"auto","created_at":"2025-11-25 10:11:25","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":952480,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/5806a6c46cd7d8e9bb664f73.png"},{"id":96689597,"identity":"2fe749bf-64b2-4edb-b7da-3dd75e7cbe95","added_by":"auto","created_at":"2025-11-25 06:15:40","extension":"jpeg","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":556386,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/3e8d6ab735c707b698bd27d1.jpeg"},{"id":96710565,"identity":"41f6fac2-caa4-4a43-b6fe-bb9209a5400d","added_by":"auto","created_at":"2025-11-25 10:10:55","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1000157,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/f4c38cc490c94626c5365760.png"},{"id":96710199,"identity":"699f094e-1cc8-4dca-b604-18d3412c2732","added_by":"auto","created_at":"2025-11-25 10:10:18","extension":"jpeg","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":280078,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/6e366f1f350f1c3fa31d6d67.jpeg"},{"id":96711210,"identity":"59e8baa8-8311-488e-94f9-6c3e4e6cf9f6","added_by":"auto","created_at":"2025-11-25 10:11:46","extension":"jpeg","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":318276,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/9f5b6d0c260bd2ce72577cf8.jpeg"},{"id":96689590,"identity":"c0b290ec-59a5-42c3-aa2e-0b5468d0d1e8","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":498473,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/e5cb7e720c78ed425771ff4d.png"},{"id":96689572,"identity":"5dd14290-3c6e-488d-942e-fbe816b1fba5","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":161016,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/53f2046da35ae51bd0759a94.png"},{"id":96689574,"identity":"172bda9b-b1cb-4545-9da0-85ed928e7e63","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":142972,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/68dcea8d273dcaaa0ff857ef.png"},{"id":96689569,"identity":"2b292db3-0ebc-4f2d-8ed8-864714cb9489","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":61058,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/8729c4f090062050fd940a00.png"},{"id":96689577,"identity":"e8b5b84b-1d0e-451c-92d2-d90ed48bce39","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":136071,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/ee450a001bf501ba7f081559.png"},{"id":96710730,"identity":"3bb0234a-62fc-4ce6-adbb-30af4fc96acb","added_by":"auto","created_at":"2025-11-25 10:11:07","extension":"png","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":198359,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/6e882e6e4b2fde512214b7b9.png"},{"id":96689575,"identity":"1ed0437b-3c6d-4902-a272-1206d7a0519f","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":94253,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/a6ad1e0c0613fc23fbc94155.png"},{"id":96710704,"identity":"b487f3ae-2f8d-4da4-8a23-ba5e4e46e937","added_by":"auto","created_at":"2025-11-25 10:11:06","extension":"png","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":90348,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/82b6b28e4240ee9da8822834.png"},{"id":96689581,"identity":"9ab2d119-17c2-4ea5-a516-2449c75e15ec","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":35,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24654,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/22fa11624a912172bff99aeb.png"},{"id":96689579,"identity":"e5a0e06a-939f-4e0f-bcf5-fe4c5db83d4d","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":36,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":266418,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/c556d7ab2353839f2424974d.png"},{"id":96689576,"identity":"41596fa8-b45c-4540-ac70-0d4a05b314c3","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":37,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":156723,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/fbc16933bbef527cefcb806a.png"},{"id":96711063,"identity":"36a6c7d2-322b-42b0-bd58-97b28fc9c306","added_by":"auto","created_at":"2025-11-25 10:11:34","extension":"png","order_by":38,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":149243,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/6655a47398813a4426c1d6e9.png"},{"id":96689593,"identity":"c2d9f8f5-5f82-4314-a2bd-fc86f24090c8","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":39,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":133254,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/5c8e45aad0fe23b45ac8c341.png"},{"id":96689596,"identity":"62878333-7c54-4bdc-9b35-f2af72864424","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":40,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":181437,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/ffdcdbb0728bc470c226d0f4.png"},{"id":96689583,"identity":"f7412f47-0000-4ef8-b475-889f973b6eac","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":41,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":58432,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/3157c4cfb73dbd12381b237e.png"},{"id":96710708,"identity":"7dfb08d9-1b1e-4595-83b2-50779d3a11f5","added_by":"auto","created_at":"2025-11-25 10:11:06","extension":"png","order_by":42,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":70283,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/0f738d4eb10f40cec6c6ce65.png"},{"id":96689587,"identity":"53bf4cf6-0064-4524-9c53-0ef19903971e","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"png","order_by":43,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":136439,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/af722f9345d014fabf3444a7.png"},{"id":96710948,"identity":"10cb34ac-3eab-4061-9005-f344e9ea6cb5","added_by":"auto","created_at":"2025-11-25 10:11:26","extension":"xml","order_by":44,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":275655,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS2589726T0structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/bc93feb2388b2823a4109269.xml"},{"id":96689585,"identity":"2b6be1f6-ef33-4259-beab-720621b06a77","added_by":"auto","created_at":"2025-11-25 06:15:39","extension":"html","order_by":45,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":304219,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/5b25dd9fad7ed701fb4e370f.html"},{"id":96689541,"identity":"97d266c3-a927-4714-b4f9-b74701f99062","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":964426,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMoral wrongness evaluation model, task design, and analytic workflow.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, Schematic representation of a moral foundations vignettes trial. Across discovery, validation, replication, and generalization cohorts, subjects evaluated the moral wrongness of validated vignettes. \u003cstrong\u003eb\u003c/strong\u003e, Feature determination, model training and evaluation. Voxel-level brain maps (beta images) were used as features in the prediction analysis. A whole-brain multivariate pattern predictive of the degree of moral wrongness (MJS) was trained on the discovery sample (n = 64) using SVR and further evaluated in discovery (cross-validated), validation (n = 30), replication (n = 27), and generalization (n = 30) cohorts. \u003cstrong\u003ec\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eWithin-subject trial-wise prediction by applying the population-level MJS to individual-level trial-wise graded moral wrongness judgments in discovery, validation, replication, and generalization cohorts. \u003cstrong\u003ed\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eIdentification and evaluation of generalized (common) and domain-isolated moral wrongness brain representations across physical and emotional care, fairness, liberty, loyalty, authority, sanctity, and social norm transgressions. \u003cstrong\u003ee\u003c/strong\u003e, Systematic tests of distributed versus localized moral wrongness circuits hypotheses. Univariate and multivariate approaches were employed to determine the contribution of specific brain systems to predict subjective moral wrongness judgments. Multiple prediction analyses were next conducted to test the performance of isolated brain regions or systems in predicting subjective moral wrongness experience. \u003cstrong\u003ef\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eTesting the specificity of the MJS in terms of distinguishable representations compared with general negative affect (PINES) and subjective disgust (VIDS). \u003cstrong\u003eg–i\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eValidation of MJS in intentional versus accidental harm scenarios (\u003cstrong\u003eg\u003c/strong\u003e), visual scenes (\u003cstrong\u003eh\u003c/strong\u003e), and social contexts (\u003cstrong\u003ei\u003c/strong\u003e). ITI intertrial interval, MJS moral judgment signature, SVR support vector regression, PINES picture induced negative affect signature, VIDS visually induced disgust signature.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/aec66759dea858ee0913b998.png"},{"id":96711067,"identity":"262aac90-8815-4743-aebf-a890dca1ec50","added_by":"auto","created_at":"2025-11-25 10:11:35","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1773253,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMoral judgment signature (MJS). a\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eThe MJS pattern thresholded at \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 (FDR corrected, 10,000-sample bootstrap). \u003cstrong\u003eb–e \u003c/strong\u003edepicts the predicted moral wrongness judgments (mean ± SE) compared to the actual severity of moral wrongness for the discovery cohort (cross-validated), and for independent validation, replication, and generalization cohorts, respectively. Accuracies reflect forced-choice comparisons based on two-sided binomial tests. \u003cem\u003er\u003c/em\u003e indicates the Pearson correlation coefficient between predicted and true ratings. \u003cstrong\u003ef–i\u003c/strong\u003e depict an average peristimulus plot (\u003cem\u003ez\u003c/em\u003e-scored mean ± SE) of the MJS response to the discovery cohort (cross-validated) and the independent validation, replication, and generalization cohorts. This reflects the average MJS response at every repetition time (TR) in the time series separated by the moral wrongness ratings. \u003cstrong\u003ej-m\u003c/strong\u003e. The range of the within-subject trial-wise prediction–outcome correlation coefficients for each subject across discovery (cross-validated), validation, replication, and generalization cohorts; note that the grids represent the prediction performance on the individual level, with each cell representing prediction performance (in terms of prediction–outcome correlation) for an individual subject. \u003cstrong\u003en–q \u003c/strong\u003edepicts the relationship between the \u003cem\u003ez\u003c/em\u003e-scored and averaged MJS pattern response and average vignette item rating using the within-subject trial-wise beta maps across discovery (cross-validated), validation, replication, and generalization cohorts. Dots represent vignette items and colors reflect the vignette category with red (blue) colors indicating higher (lower) average moral wrongness ratings.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/f620152e7798704932d9d93d.png"},{"id":96689543,"identity":"352cc1c4-20d9-4c31-919b-82d7faffebf5","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":997991,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCross-decoding of graded moral wrongness computations. a\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eDistribution of training data (subject-level beta maps for each vignette condition and wrongness rating option) used for developing common and domain-specific moral wrongness models. \u003cstrong\u003eb\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eAverage model response (that is, predicted moral wrongness rating) for each model (\u003cem\u003ey\u003c/em\u003e-axis) and vignette condition (\u003cem\u003ex\u003c/em\u003e-axis). Domain-specific models consistently exhibited the highest average model response for their matched target domain. \u003cstrong\u003ec\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eCross-prediction of ratings across domain types tested by Pearson correlation between the predicted and the observed outcomes for each train–test domain pair. \u003cstrong\u003ed\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eRelationship between observed and predicted ratings (that is, model response). Data are shown as mean values across subjects for each vignette condition. Error bars reflect within-subject s.e.m. The common model, trained on all vignette conditions, significantly predicted ratings to each moral (but not social norm) vignette condition. Domain-specific models, optimized for specificity by setting other conditions at 0 during training, significantly predicted ratings for four target (color-matched) domains. \u003cem\u003er\u003c/em\u003e, mean within-subject Pearson correlation between predicted and observed ratings; two-sided P values based on a 10,000 samples bootstrap test of within-subject \u003cem\u003er\u003c/em\u003e values.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/307d1411bef05f05a4ceef8b.png"},{"id":96689545,"identity":"329a6c40-feb8-4d14-9598-02434e22225f","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":914107,"visible":true,"origin":"","legend":"\u003cp\u003eComputations of moral wrongness are associated with and predicted by distributed brain regions. \u003cstrong\u003ea. \u003c/strong\u003eThresholded MJS; \u003cstrong\u003eb.\u003c/strong\u003e Thresholded transformed MJS ‘activation pattern’. \u003cstrong\u003ec. \u003c/strong\u003eThe conjunction between MJS and transformed ‘activation pattern’. Images thresholded at \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05, FDR corrected. Hot colour indicates positive weights (\u003cstrong\u003ea\u003c/strong\u003e) or associations (\u003cstrong\u003eb\u003c/strong\u003e), whereas cold colour indicates negative weights (\u003cstrong\u003ea\u003c/strong\u003e) or associations (\u003cstrong\u003eb\u003c/strong\u003e).\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/6bef8f14742348772df1c41b.png"},{"id":96689555,"identity":"ab90a47e-db6a-49a5-8017-8a60f05ee064","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":810739,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eLocal brain region and network predictions in the discovery cohort. a,b\u003c/strong\u003e, Brain regions that significantly predict graded moral wrongness judgments revealed by searchlight (\u003cstrong\u003ea\u003c/strong\u003e) and parcellation-based (\u003cstrong\u003eb\u003c/strong\u003e) analyses, respectively. Statistical significance was evaluated by prediction–outcome correlation (Pearson; two-sided). Uncorrected P values equivalent to \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 might be considered lenient; therefore, brain regions that survived \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001 uncorrected (corresponding to \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.005 and 0.004, FDR corrected, for searchlight- (\u003cstrong\u003ea\u003c/strong\u003e) and parcellation-based (\u003cstrong\u003eb\u003c/strong\u003e) predictions, respectively) are displayed. Histograms: cross-validated predictions (correlations) from local searchlights (\u003cstrong\u003ea\u003c/strong\u003e) or parcellations (\u003cstrong\u003eb\u003c/strong\u003e). The red lines indicate the prediction–outcome correlation from MJS. \u003cstrong\u003ec,d\u003c/strong\u003e, Cross-validated predictions (mean ± s.e.m.) from dmPFC- (\u003cstrong\u003ec\u003c/strong\u003e) and vmPFC-based (\u003cstrong\u003ed\u003c/strong\u003e) models. \u003cstrong\u003ee\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003edemonstrates that the information about moral wrongness is distributed across multiple systems. Model performance was evaluated as increasing numbers of voxels/features (x axis) were used to predict moral wrongness judgments in different regions of interest including the entire brain (black), consciousness network (orange), subcortical regions (brown) or large-scale resting-state networks. The y axis denotes the crossvalidated prediction-outcome correlation. Colored dots indicate the mean correlation coefficients, solid lines indicate the mean parametric fit and shaded regions indicate standard deviation. Model performance is optimized when approximately 10,000 voxels are randomly sampled across the whole-brain. MJS moral judgment signature.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/29982db44e0972bc95672c25.png"},{"id":96711073,"identity":"60bc583e-fbb5-4cab-84eb-18ebb56c7700","added_by":"auto","created_at":"2025-11-25 10:11:35","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1242447,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparing neural signatures for moral wrongness, negative affect and disgust (MJS, PINES, VIDS). a\u003c/strong\u003e, River plots showing spatial similarity (cosine similarity) between stable decoding maps and selected ROIs. \u003cstrong\u003eb\u003c/strong\u003e, River plots showing spatial similarity (cosine similarity) between stable decoding maps and selected networks. In \u003cstrong\u003ea\u003c/strong\u003e and \u003cstrong\u003eb\u003c/strong\u003e, ribbons are normalized by the max cosine similarity across all ROIs (\u003cstrong\u003ea\u003c/strong\u003e) and networks (\u003cstrong\u003eb\u003c/strong\u003e). Stable decoding models were thresholded at FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 and positive voxels were retained only for similarity calculation and explanation. Ribbon locations in relation to the boxes are arbitrary. The pie charts show the relative contributions of each model to each ROI (a) or network (b) (that is, the percentage of voxels with the highest cosine similarity for each map). \u003cstrong\u003ec\u003c/strong\u003e,\u003cstrong\u003ed\u003c/strong\u003e,\u003cstrong\u003ee\u003c/strong\u003e, MJS, PINES, and VIDS more accurately (shown as overall prediction-outcome Pearson correlation, forced-choice accuracy, and Cohen’s \u003cem\u003ed\u003c/em\u003e) predict the targeted constructs at high versus low intensities, respectively. Only the MJS was sensitive and specific to moral wrongness ratings. Pattern expressions display \u003cem\u003ez\u003c/em\u003e-scored and averaged model responses. \u003cstrong\u003ef\u003c/strong\u003e, MJS response partially mediates the association between PINES response and moral wrongness ratings in the discovery cohort, and MJS response fully mediates the PINES response–wrongness rating association in the validation cohort. \u003cstrong\u003eg\u003c/strong\u003e, PINES response does not mediate the effect of MJS response on wrongness ratings in the discovery cohort, and partially mediates the effect of MJS response on wrongness ratings in the validation cohort. In \u003cstrong\u003ef\u003c/strong\u003e and \u003cstrong\u003eg\u003c/strong\u003e, the mediation analysis examines whether the observed covariance between the independent variable (\u003cem\u003eX\u003c/em\u003e) and the dependent variable (\u003cem\u003eY\u003c/em\u003e) can be explained by the third variable (\u003cem\u003eM\u003c/em\u003e, also mediator); for details, see Methods; two-sided \u003cem\u003eP\u003c/em\u003evalues are based on bootstrap tests with 10,000 samples, uncorrected.\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/0078ec3da25f77330a1aac2c.png"},{"id":96710968,"identity":"a23299a3-781d-4f76-9850-8e5877038081","added_by":"auto","created_at":"2025-11-25 10:11:27","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":321714,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSensitivity of MJS for intentional versus accidental harms. \u003c/strong\u003eMJS could predict intentional versus accidental harms in neurotypical adults (left) (shown as forced-choice classification accuracy and Cohen’s \u003cem\u003ed\u003c/em\u003e), while PINES and VIDS failed to predict intentional versus accidental harms. In individuals with autism spectrum disorder (right), none of the neural biomarkers could predict intentional versus accidental harms. \u003cem\u003eP\u003c/em\u003evalues were based on binomial tests, two-sided (uncorrected). Pattern expressions display the \u003cem\u003ez\u003c/em\u003e-scored and averaged model responses for MJS, PINES, and VIDS, respectively.\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/8eb8d3d286702065cfdcb662.png"},{"id":96689561,"identity":"21485c8c-e458-4d8a-8565-a9841de5cc8b","added_by":"auto","created_at":"2025-11-25 06:15:38","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":402401,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eValidation tests of MJS in visual scenes and social contexts. \u003c/strong\u003ePlots depict the average pattern expression (\u003cem\u003ez\u003c/em\u003e-scored; mean ± SE) compared to the actual level of moral image ratings for the visual scenes datasets (study 9–10) (\u003cstrong\u003ea\u003c/strong\u003e) and a series of unfair offers (study 11) (\u003cstrong\u003eb\u003c/strong\u003e). Accuracies (Cohens’\u003cem\u003ed\u003c/em\u003e) reflect forced-choice comparisons between moral vs. immoral image ratings (\u003cstrong\u003ea\u003c/strong\u003e) and unfair vs. fair offers (\u003cstrong\u003eb\u003c/strong\u003e) based on two-sided binomial tests. \u003cem\u003er\u003c/em\u003eindicates the overall (between- and within-subject prediction-outcome correlation.\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/023cdbf99e69beb1ee1c119a.png"},{"id":105752063,"identity":"e7250841-99e8-458b-bf1e-ea3a8037db91","added_by":"auto","created_at":"2026-03-30 15:54:17","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":8299767,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/21456932-072f-409e-bd4e-d2654a3f1731.pdf"},{"id":96710612,"identity":"d7181639-0dc3-48ce-978e-386206e8494d","added_by":"auto","created_at":"2025-11-25 10:10:59","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":4278257,"visible":true,"origin":"","legend":"Supplemental Materials","description":"","filename":"suppinfofinal.docx","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/1ae6abc415a3307bbd35f2c2.docx"},{"id":96710244,"identity":"98a02c0b-d8ac-4bb4-86f8-1a7440dd1028","added_by":"auto","created_at":"2025-11-25 10:10:20","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":3044454,"visible":true,"origin":"","legend":"","description":"","filename":"ExtendedDataFig.docx","url":"https://assets-eu.researchsquare.com/files/rs-7935407/v1/85a54e40649d59e1350f42f5.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"A Sensitive and Specific Neural Signature Robustly Predicts Graded Computations of Moral Wrongness","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe discernment of gradations in transgressive behaviors provides a compass for regulating human life. Indeed, the relative severity of moral violations is a fundamental issue in ethics and is codified in systems like the US Penal Code, which distinguishes between infractions, misdemeanors, and felonies, and provides statutes for the legal ramifications of different types of transgressions\u003csup\u003e1-2\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHow the human brain computes the graded reprehensibility of varied moral transgressions remains a compelling and enduring puzzle for neuroscience\u003csup\u003e3-5\u003c/sup\u003e. Decades of research demonstrate that moral judgment and decision making recruit domain-general neural circuits, from affective\u003csup\u003e6\u003c/sup\u003e and value-based systems\u003csup\u003e7-8\u003c/sup\u003e to mentalizing networks\u003csup\u003e9\u003c/sup\u003e. Meta-analyses of moral brain mapping studies further document that different kinds of moral judgements are reliably orchestrated by the default mode network (DMN)\u003csup\u003e10-12\u003c/sup\u003e, with robust contributions from dorsomedial and ventromedial prefrontal cortex (PFC), temporoparietal junction (TPJ), as well as posterior cingulate cortex (PCC) and precuneus (PC). Neural activation patterns in the DMN contain information concerning the nature of the observed moral behaviors, enabling the flexible distinction between intentional and accidental harms\u003csup\u003e9\u003c/sup\u003e or varied moral violations (e.g., betrayal, cheating, etc.)\u003csup\u003e13-16\u003c/sup\u003e. However, we still know little about the neural circuit supporting the precise computation of \u003cem\u003egraded\u003c/em\u003e moral wrongness and whether individual, neurologic, and cultural differences in brain representations of moral severity predict variation in momentary, self-reported, and graded moral judgments\u003csup\u003e2\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe herein capitalized on recent advances in multivariate pattern analysis (MVPA)-based neural decoding\u003csup\u003e17\u003c/sup\u003e to develop a multivariate brain model that can predict the subjective experience of degrees of moral wrongness. In contrast to extant univariate moral neuroimaging studies that express brain activation as a function of the underlying binary moral judgment (i.e., right versus wrong)\u003csup\u003e6,18-20\u003c/sup\u003e, our developed multivariate brain model permits the inference of continuous (graded) moral wrongness judgments given brain activity\u003csup\u003e21\u003c/sup\u003e. Multivariate brain models have already yielded large effect sizes in brain-outcome associations for somatic and vicarious\u003csup\u003e22\u003c/sup\u003e as well as future pain\u003csup\u003e23\u003c/sup\u003e, negative affect\u003csup\u003e24,25\u003c/sup\u003e and pleasure\u003csup\u003e26\u003c/sup\u003e, as well as fear\u003csup\u003e27\u003c/sup\u003e, threat anticipation\u003csup\u003e28\u003c/sup\u003e, and disgust\u003csup\u003e29\u003c/sup\u003e. Furthermore, quantitative predictions about outcomes can be empirically falsified, tested, and validated across studies, populations, and scanner settings, which promotes generalizability and reproducibility\u003csup\u003e30\u003c/sup\u003e. Notably, individuals may not always have reliable insights into their moral cognitions\u003csup\u003e31\u003c/sup\u003e or succumb to social desirability response biases, whereas predictive brain models can yield objective neurobiological read-outs of diverse mental states\u003csup\u003e32\u003c/sup\u003e. Finally, examining for which individuals and under which conditions the model predictions fail can provide novel insights into aberrant moral cognition and serve as targets for interventions\u003csup\u003e33\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eIn this work, we therefore developed a functional magnetic resonance imaging (fMRI)-based model, or neurologic signature, of graded moral wrongness judgments. Using distributed blood-oxygen-level-dependent (BOLD)-based information, we investigate whether (1) it is possible to develop a sensitive and specific neural signature of severity ratings of subjective moral wrongness on the population level, (2) this neural signature robustly generalizes across independent samples, MRI systems, and two different cultures, and (3) can predict the graded severity of momentary (trialwise) moral wrongness judgments on the individual level, (4) the neural signature in brain regions (e.g., the dorsomedial prefrontal cortex; dmPFC) and networks (e.g., DMN) implicated in moral judgment is sufficient to capture the continuous rating of degrees of moral wrongness, and (5) the neural representation of momentary graded moral wrongness judgments is distinct (i.e., sensitive and specific) from brain representations of general emotionally aversive states and subjective disgust.\u003c/p\u003e\n\u003cp\u003eMore specifically, we trained a linear support vector regression (SVR) algorithm\u003csup\u003e27-29\u0026nbsp;\u003c/sup\u003ein healthy subjects (n = 64) from a previous study\u003csup\u003e13\u003c/sup\u003e to identify the brain signature that predicts the graded severity of trial-by-trial subjective moral wrongness ratings elicited by validated and standardized moral foundations vignettes (MFV)\u003csup\u003e34\u003c/sup\u003e. We then evaluated the performance of the established moral judgment signature (MJS) in (1) a novel U.S.-based validation cohort using the identical MRI system and MFV task but with additional vignettes (n= 30); (2) a U.S.-based conceptual replication cohort from a previous study\u003csup\u003e14\u003c/sup\u003e at a different campus with a different MRI system and using a modified MFV task (n=27); and (3) a novel generalization cohort using a validated Dutch version of the MFV\u003csup\u003e36\u003c/sup\u003e in the Netherlands with a different MRI system and Dutch subjects (n = 30). Furthermore, partial least squares regression (PLS-R) provided a framework for jointly estimating both generalized (common) and domain-specific representations of graded moral wrongness across physical and emotional care, fairness\u003ca href=\"#_ftn1\" name=\"_ftnref1\" title=\"\"\u003e\u003c/a\u003e\u003csup\u003e1\u003c/sup\u003e, liberty, loyalty, authority, sanctity, and social (conventional) norms, allowing us to examine how generalized and domain-specific representations jointly contribute to degrees of moral wrongness judgments\u003csup\u003e24,36,37\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo extend the perspective from a population to an individual level, we probed whether the MJS can predict trial-wise graded moral wrongness ratings for each subject in discovery, validation, replication, and generalization cohorts separately. We further systematically identified brain regions that are associated with (forward model, i.e., expressing the observed data as functions of underlying variables) and predictive of (backward model, i.e., expressing variables of interest as functions of the data) graded moral wrongness judgments\u003csup\u003e38\u003c/sup\u003e and examined to what extent single brain systems or networks can capture degrees of moral wrongness severity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMoreover, we probed the sensitivity and specificity of the developed MJS for graded judgments of moral wrongness. Considering the intricate role of affect for moral judgment\u003csup\u003e31,39-42\u003c/sup\u003e, we determined the functional specificity of the neural moral judgment signature by comparing the spatial and functional similarities between the MJS with predictive brain markers for general negative emotional experience\u003csup\u003e25\u003c/sup\u003e and disgust\u003csup\u003e29\u003c/sup\u003e. Finally, we validated the MJS in novel contexts and paradigms, including intentional versus accidental moral transgressions\u003csup\u003e9\u003c/sup\u003e, socio-moral images\u003csup\u003e35,43\u003c/sup\u003e, and unfair social offers\u003csup\u003e29\u003c/sup\u003e.\u003c/p\u003e"},{"header":"Results","content":"\u003ch2\u003e\u003cstrong\u003eMoral vignettes elicited a robust range of moral wrongness judgments\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eMoral wrongness judgments were elicited by validated moral foundations vignettes (MFV)\u003csup\u003e34\u003c/sup\u003e shown to reliably induce varying levels of perceived moral wrongness\u003csup\u003e13-14,35\u003c/sup\u003e. The MFV span 120, one-sentence descriptions detailing the violation of one (and only one) of seven moral foundations: Physical care, emotional care, fairness, liberty, loyalty, authority, and sanctity. The vignettes also contain a non-moral, social norm transgression category. Subjects were explicitly instructed to vividly imagine each scenario and were asked to rate the moral wrongness of each vignette on a 4-point Likert scale ranging from 1 (not morally wrong) to 4 (extremely morally wrong) (Fig. 1a; Extended Data Fig. 1). As the vignette appeared on screen, subjects could immediately provide a graded moral wrongness judgment to better capture moral judgments\u0026rsquo; hypothesized intuitive nature\u003csup\u003e31\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eWe initially tested whether the vignette stimuli elicited meaningful and varying levels of perceived moral wrongness. To this end, we plotted the number of each selected moral wrongness level (across subjects and categories) for each run (Extended Data Fig. 2a) and for each vignette category (across subjects and runs, Extended Data Fig. 2b). We found that the stimuli induced sufficient levels of moral wrongness in the discovery cohort (n = 64) which was used to develop the neural signature of moral wrongness (see below for details), such that between 9% (Authority) and 45% (Physical Care) of trials of each moral vignette were rated as 4 (reflecting that they induced strong moral condemnation). Confirming previous work\u003csup\u003e34-35\u003c/sup\u003e, violations of physical care were rated as most morally wrong, whereas transgressions of conventional norms received the expected lowest moral wrongness ratings (Extended Data Fig. 2b). Individual differences in self-reported sensitivity to moral foundations were correlated with moral wrongness ratings in a domain-consistent way (\u003cem\u003er\u003c/em\u003e = 0.35\u0026ndash;0.47; Supplementary Table 2). Self-reported moral wrongness judgments were generally evenly distributed across runs and all 64 subjects used all 4 levels of moral wrongness ratings at least once.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eA brain signature for graded computations of moral wrongness (MJS)\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eWe used brain activity during the vignette presentation and moral judgment phase (Fig. 1a) across all moral foundations to develop a multivariate model capable of predicting graded moral wrongness ratings. We trained a support vector regression (SVR) algorithm to predict the selected graded moral wrongness ratings using the smoothed and standardized, whole-brain parametric maps for each moral judgment level (1\u0026ndash;4) and each subject as input features (Fig. 1b). The resulting pattern of voxel weights was considered as the moral judgment signature (MJS) (Fig. 2a). We assessed the performance of each pattern in predicting the graded severity of moral wrongness with leave-one-subject-out (LOSO) cross-validation in which the same SVR was iteratively trained on the parametric activation maps from all but one subject and tested on the activity maps of the held-out subject.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe analysis indicated that across subjects, the developed MJS accurately predicted graded degrees of moral wrongness judgments. Specifically, for individual subjects the average within-subject correlation between predicted and actual graded moral wrongness ratings (4 pairs of scalar values per subject) was \u003cem\u003er\u003c/em\u003e = 0.94 \u0026plusmn; 0.02 (standard error (SE)), the average root mean squared error (RMSE) was 0.68 \u0026plusmn; 0.05 and the overall (between- and within-subjects) prediction-outcome (i.e. 256 pairs) correlation coefficient was 0.78 (averaged across 64 repetitions). Testing the MJS model developed in the discovery cohort, with no further model fitting, in the validation, replication, and generalization cohorts yielded significant graded moral wrongness predictions (Fig. 2c\u0026ndash;e; Supplementary Table 3), indicating a robust and replicable neurologic signature for graded computations of moral wrongness.\u003c/p\u003e\n\u003cp\u003eTo further determine the sensitivity of the MJS to predict graded degrees of moral wrongness, a two-alternative forced-choice test was applied, comparing all possible pairs of activation maps within each subject and choosing the one with higher MJS response as more morally wrong. In the (LOSO cross-validated) discovery cohort (Fig. 2b; Supplementary Table 4), the MJS response classified not morally wrong (1) versus extremely morally wrong (4) judgments with 98% accuracy; not morally wrong (1) versus very morally (3) with 98% accuracy; and moderately morally wrong (2) versus extremely morally wrong (4) with 98% accuracy. Moreover, the MJS response could distinguish each successive pair of moral wrongness rating levels (e.g., rating 2 versus 3) with \u0026ge; 91% accuracy, which was significantly better than chance level (50%; p \u0026lt; 0.001)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSimilar performance was achieved across validation, replication, and generalization cohorts, again demonstrating an accurate neural signature for graded moral wrongness predictions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRetraining the MJS excluding the entire occipital lobe\u003csup\u003e44\u003c/sup\u003e revealed slightly lower, but significant prediction accuracies (\u003cem\u003er\u003c/em\u003e = 0.91 \u0026plusmn; 0.03 for within-subject prediction-outcome correlations and \u003cem\u003er\u003c/em\u003e = 0.71, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001, for the overall prediction-outcome correlation; Extended Data Fig. 3), suggesting that graded computations of moral wrongness are informed by the visual system\u003csup\u003e45-46\u003c/sup\u003e, but also draw on additional distributed brain systems.\u003c/p\u003e\n\u003cp\u003eIn addition, we applied the MJS to the vignette time series data using dot product across discovery (LOSO cross-validated), validation, replication, and generalization cohorts (Fig. 2 f\u0026ndash;i) to evaluate the chronometry of the moral wrongness judgment pattern. Visual inspection of the MJS reactivity at each timepoint following stimulus onset indicated that the MJS response began approximately 4sec following vignette onset and increased with increasing levels of reported degrees of moral wrongness during approximately 5\u0026ndash;9 sec. These findings align with previous work demonstrating a temporal sequence of moral deliberation followed by moral verdict\u003csup\u003e47\u003c/sup\u003e and confirmed that the MJS dynamically tracked graded computations of moral wrongness judgment.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eWithin-subject trial-wise prediction\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe subjective feeling of disapproval that comes with experiencing a moral transgression suggests that moral judgments are momentary, and to some degree individually constructed states\u003csup\u003e31,48\u003c/sup\u003e. Thus, a key question is to what extent the population-level model (i.e., the MJS), which is a statistical summary of a highly variable set of instances, can predict momentary (trial-wise) graded moral wrongness judgments for each subject (on the individual level). To this end, we performed single-trial analyses using the Least Squares All (LSA) approach\u003csup\u003e49\u003c/sup\u003e to obtain a graded moral wrongness beta map for each vignette item for each subject across discovery (~120 beta maps per subject), validation (~72 beta maps per subject), replication (~120 beta maps per subject), and generalization (~120 beta maps per subject) cohorts. The MJS was next applied to these beta maps to calculate the pattern expressions which were further correlated with the true graded moral wrongness ratings for each vignette and subject separately. The statistical significance was evaluated by prediction-outcome Pearson correlation for each subject separately. We found that the MJS could significantly predict graded trial-by-trial moral wrongness ratings for 61 out of 64 subjects (95.3%) in the discovery cohort (cross-validated) and for 25 out of 30 (83.3%) subjects in the validation cohort, for 11 out of 27 subjects (40.7%) in the replication cohort, and for 21 out of 30 subjects (70%) in the generalization cohort. The mean prediction-outcome correlations were 0.44 \u0026plusmn; 0.15 (discovery; \u003cem\u003ep\u003c/em\u003e \u0026lt; .0001), 0.34 \u0026plusmn; 0.11 (validation; \u003cem\u003ep\u003c/em\u003e \u0026lt; .0001), 0.17 \u0026plusmn; 0.12 (replication; \u003cem\u003ep\u003c/em\u003e \u0026lt; .0001), and 0.22 \u0026plusmn; 0.14 (generalization; \u003cem\u003ep\u003c/em\u003e \u0026lt; .0001) (Fig. 2 j\u0026ndash;m; two-sided \u003cem\u003eP\u003c/em\u003e values based on a 10,000 samples bootstrap test of within-subject \u003cem\u003er\u003c/em\u003e values).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMoreover, the MFV are controlled on a range of dimensions (e.g., syntactic structure, word and character length) but deliberately vary in the type of transgressions they portray (e.g., physical care, fairness, loyalty, etc.). Thus, we explored whether the MJS pattern could predict the graded moral wrongness ratings for each individual vignette item. Here, we first averaged the within-subject trial-wise MJS pattern expressions and graded moral wrongness ratings for each vignette. We then computed the Pearson correlation between the \u003cem\u003ez\u003c/em\u003e-scored and averaged MJS pattern expression and average graded moral wrongness rating for each vignette item. We found strongly positive and statistically significant linear relationships between the average ratings and pattern responses across all vignettes in the discovery (cross-validated) cohort \u003cem\u003er\u003c/em\u003e = .88, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001, the validation cohort \u003cem\u003er\u003c/em\u003e = .83, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001, the replication cohort \u003cem\u003er\u003c/em\u003e = .55, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001, and the generalization cohort \u003cem\u003er\u003c/em\u003e = .60, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001 (Fig. 2n\u0026ndash;q; Supplementary Table 5). Compellingly, the averaged MJS pattern expressions for each vignette item in the (cross-validated) discovery cohort correlated strongly positively and significantly (\u003cem\u003er\u003c/em\u003e = .79; \u003cem\u003ep\u003c/em\u003e \u0026lt; .001) with the averaged graded vignette moral wrongness ratings from an out-of-sample U.S. population study (n = 510)\u003csup\u003e34\u003c/sup\u003e. Analogously, the averaged MJS pattern expressions for each vignette item in the generalization cohort correlated positively and significantly (\u003cem\u003er\u003c/em\u003e = .58; \u003cem\u003ep\u003c/em\u003e \u0026lt; .001) with the averaged graded vignette moral wrongness ratings from an out-of-sample, Dutch population study (n = 586)\u003csup\u003e35\u003c/sup\u003e. Together, these findings suggest that the MJS reliably predicts computations of graded moral wrongness across both individuals, cohorts, and vignette items.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eCommon and domain-specific brain representations of moral wrongness\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003e\u0026nbsp;A central question for moral neuroscience is whether moral judgments of different types of moral behaviors are captured by common or domain-specific neural representations. Recent work demonstrates that moral judgment of different moral transgressions elicits dissociable neural activations\u003csup\u003e13-16,19\u003c/sup\u003e, but it is not clear to what extent the representation of graded moral wrongness for variable moral violations is robustly instantiated via common and domain-specific brain representations. Inspired by recent findings revealing common and stimulus-type-specific brain representations of negative affect\u003csup\u003e24\u003c/sup\u003e, we used PLS-R on the discovery cohort data (Fig. 3a) for jointly estimating both common and domain-specific (care, fairness, loyalty, etc.) graded moral wrongness brain representations (see Methods and Extended Data Fig. 4). This produced nine multivariate patterns: one for each of the eight vignette conditions and one for common moral wrongness. Models were tested using LOSO cross-validation. To evaluate model predictions, we examined whether (i) average graded moral wrongness predictions (i.e., model responses) were significantly higher for each matched model-domain pair and (ii) there was a significant correlation, expressed as mean within-subject \u003cem\u003er\u003c/em\u003e \u0026plusmn; standard error (s.e.), for each model and graded moral wrongness rating.\u003c/p\u003e\n\u003cp\u003eFirst, each of the eight domain-isolated models exhibited the highest average model response for its matched target domain (Fig. 3b, Supplementary Table 6), providing further evidence that computations of graded moral wrongness for distinct moral transgressions rely on dissociable neural representations. Second, the common model predicted graded moral wrongness ratings for each moral vignette condition, but not social norms, as evidenced by significant associations between observed and predicted ratings (Fig. 3c\u0026ndash;d; Supplementary Table 7). The domain-isolated Physical Care (\u003cem\u003er\u003c/em\u003e = 0.36, \u003cem\u003ep\u003c/em\u003e = 0.001), Fairness (\u003cem\u003er\u003c/em\u003e = 0.62, \u003cem\u003ep\u003c/em\u003e = 0.001), Liberty (\u003cem\u003er\u003c/em\u003e = 0.41, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001), and Sanctity (\u003cem\u003er\u003c/em\u003e = 0.43, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001) models were also significantly sensitive towards graded moral wrongness computations in their respective target domains, and Emotional Care (\u003cem\u003er\u003c/em\u003e = 0.17, \u003cem\u003ep\u003c/em\u003e = 0.09) as well as Loyalty (\u003cem\u003er\u003c/em\u003e = 0.18, \u003cem\u003ep\u003c/em\u003e = 0.06) approached statistical significance. Of note, none of the domain-isolated models were specific to their target condition, as evidenced by statistically significant cross-prediction of off-target graded moral wrongness ratings (Fig. 3c, Supplementary Table 7). Taken together, these results provided strong support for common coding of graded moral wrongness, and mixed evidence for domain-specific graded moral wrongness representations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMoral wrongness is associated with and predicted by distributed neural systems\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn view of the evidence for domain-general computation of graded moral wrongness, we next systematically determined individual brain regions that were associated with subjective graded moral wrongness ratings and that provided consistent and reliable contributions to the whole-brain moral judgment decoding model (MJS) using different analytic strategies. We first examined regions that made reliable contributions to the graded moral wrongness prediction within the MJS itself by applying a bootstrap test to identify regions with significant, consistent model weights (\u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05, false discovery rate (FDR) corrected). Given that some brain features could contribute to controlling for noise in the data\u003csup\u003e38\u003c/sup\u003e rather than the moral judgment computation per se, we next transformed the population-level MJS into \u0026lsquo;activation pattern\u0026rsquo; (\u0026lsquo;structure coefficient\u0026rsquo;; for details, see Methods). The results showed that a set of distributed brain systems exhibited significant model weights (\u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05, FDR corrected; Fig. 4a) and structure coefficients (\u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05, FDR corrected; Fig. 4b), including dmPFC, the precuneus, as well as visual and supplementary motor area (Fig. 4c). Brain regions associated with and predictive of moral wrongness computations on the individual level determined with convergent univariate and multivariate approaches identified a similar set of broadly distributed regions (Supplementary Methods, Supplementary Results and Extended Data Fig. 5). The broad conclusion is that the neural representation of graded moral wrongness is not limited to a single or a set of focal regions (e.g., the dmPFC), but rather includes a broad set of regions spanning multiple systems.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMJS outperforms prediction based on local systems\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDue to continuing debates on the contribution of specific brain regions\u003csup\u003e5,50\u003c/sup\u003e, such as dmPFC\u003csup\u003e13,19\u003c/sup\u003e, vmPFC\u003csup\u003e51\u0026nbsp;\u003c/sup\u003eor the default mode network (DMN)\u003csup\u003e10,52-53\u003c/sup\u003e to human moral judgment, (1) both searchlight- and parcellation-based analyses were employed to determine local brain regions that were predictive of subjective moral wrongness severity, and (2) models were trained on single brain region and networks to examine to what extent these models could predict moral wrongness severity compared to the whole-brain MJS. As shown in Fig. 5a,b, moral wrongness severity could be significantly predicted by distributed regions, including dmPFC and PCC as well as visual cortex and supplementary motor area (SMA) (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001, average across leave-one-subject-out cross-validation procedure). However, none of the local models predicted moral wrongness to the extent the MJS did, suggesting that moral wrongness representations are distributed across regions, and can best be captured in whole-brain but not local analyses.\u003c/p\u003e\n\u003cp\u003eWe then re-trained predictive SVR models restricted to activations in (1) the bilateral dmPFC, (2) the bilateral vmPFC, (3) a meta-analytic \u0026ldquo;moral\u0026rdquo; map\u003csup\u003e54\u003c/sup\u003e (Extended Data Fig. 6), (4) a consciousness network, (5) a subcortical network, and (6) each of seven large-scale cerebral networks (Methods). Our findings showed that the dmPFC (prediction\u0026ndash;outcome correlation \u003cem\u003er\u003c/em\u003e = 0.18 and 0.38 for discovery (cross-validation)), the vmPFC (prediction\u0026ndash;outcome correlation \u003cem\u003er\u0026nbsp;\u003c/em\u003e= 0.27 for discovery (cross-validation)), and other brain networks (Fig. 5c\u0026ndash;e) could, to some extent, predict graded moral wrongness judgments. Nonetheless, although statistically significant (\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001), the effect sizes in terms of prediction\u0026ndash;outcome correlations (including searchlight- and parcellation-based predictions) were substantially smaller than those obtained from the MJS, which used features spanning multiple brain systems.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo control for potential effects of the number of features/voxels in prediction analyses (that is, the whole-brain model contains many more features), we randomly selected voxels (repeated 1,000 times) from a uniform distribution spanning the entire brain (black; Fig. 5e), consciousness network (light orange), subcortical (brown) or individual large-scale cerebral networks (averaged over 1,000 iterations)\u003csup\u003e27-29\u003c/sup\u003e. The asymptotic prediction when sampling from all brain systems as we did with the MJS (black line in Fig. 5e) was substantially higher than the asymptotic prediction within individual networks (coloured lines in Fig. 5e). Notably, only the visual system showed an initially higher prediction-outcome correlation compared to the whole-brain model when sampling fewer than 1000 voxels, which further underlines the relative importance of the occipital lobe for representing moral\u003csup\u003e13,45-46\u003c/sup\u003e and emotional\u003csup\u003e55\u003c/sup\u003e concepts. Nevertheless, this analysis thus demonstrated that whole-brain models have much larger effect sizes than those using features from a single network. Moreover, model performance was optimized (that is, reaching asymptote) when approximately 10,000 voxels were randomly sampled across the whole brain, as long as voxels were drawn from multiple brain systems, further confirming that information about moral wrongness is contained in patterns of activity that span multiple systems. Together, converging lines of evidence from the above systematic analyses point to the fact that subjective, graded moral wrongness judgments are encoded in distributed neural patterns that span multiple systems, adding to increasing evidence that morality is represented in distributed brain systems rather than single brain regions or networks.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eSeparable signatures of moral wrongness, negative affect, and disgust\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eSentimentalist theories of moral cognition assert that moral judgments are fundamentally rooted in emotional responses\u003csup\u003e31,39-42\u003c/sup\u003e, but the extent to which neural representations for computing graded moral wrongness overlap with brain representations of graded (negative) emotional experience remains unclear. Accordingly, we determined whether (i) neural biomarkers of general negative emotion experience (PINES)\u003csup\u003e25\u003c/sup\u003e and specific socio-moral emotions such as disgust (VIDS)\u003csup\u003e29\u003c/sup\u003e can predict graded moral wrongness judgments and (ii) whether the MJS is sensitive and specific to graded \u003cem\u003emoral\u003c/em\u003e wrongness judgments. To address these questions, we performed a series of analyses. First, we investigated spatial similarities between stable decoding maps and a set of regions of interest (ROIs) commonly involved in moral cognition and emotion as well as networks\u003csup\u003e10\u003c/sup\u003e. Speaking to the distributed nature of moral wrongness computations, the MJS was the only model that contained stable predictive voxels in all regions (Fig. 6a). Amygdala, dmPFC, anterior Insula, and precuneus contained stable predictive voxels in all three models, but the visually induced disgust signature (VIDS) showed larger contributions to the amygdala, anterior Insula, and precuneus than MJS and picture-induced negative emotion signature (PINES). In contrast, the MJS exhibited the largest contributions in dlPFC and pSTS, while dlPFC, vmPFC, thalamus, and pSTS exhibited a degree of specificity for MJS and VIDS.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn line with previous work\u003csup\u003e29\u003c/sup\u003e, all networks showed stable predictive voxels across the three models, but the relative contributions of each network to each model varied, such that the visual network contributed more strongly to the MJS than to VIDS and PINES; the ventral attention and limbic networks contributed more strongly to VIDS than to MJS and PINES; and the somatomotor network strongly contributed to PINES (Fig. 6b).\u003c/p\u003e\n\u003cp\u003eSecond, we examined functional similarities between the MJS, PINES and VIDS, respectively. The results showed that the MJS was more sensitive and specific to predict high versus low moral wrongness as compared with PINES or VIDS, as reflected by larger effect sizes for MJS in (cross-validated) discovery (2.18\u0026ndash;6.95 times larger), validation (3.76\u0026ndash;10.44 times larger), replication (2.73\u0026ndash;3.74 times larger), and generalization (2.02\u0026ndash;3.34 times larger) cohorts (Fig. 6c). PINES was more sensitive to predict high versus low negative emotion with effect sizes 4.17\u0026ndash;48.0 times higher than those for MJS or VIDS (Fig. 6d), and VIDS more accurately predicted high versus low disgust with effect sizes 3.13\u0026ndash;27.32 higher than those for PINES or MJS (Fig. 6e). The above findings were further substantiated by the comparisons of the overall and within-subject prediction\u0026ndash;outcome correlations of the three decoders across four datasets (Supplementary Table 8).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFinally, we employed multi-level mediation models, which examined whether the covariance between two variables (\u003cem\u003eX\u003c/em\u003e and \u003cem\u003eY\u003c/em\u003e) can be explained by a third variable (\u003cem\u003eM\u003c/em\u003e), to determine the neurofunctional relationship between the representations of moral wrongness encoded in MJS, PINES, and VIDS. While PINES could to some degree track moral wrongness (discovery cohort: \u003cem\u003er\u003c/em\u003e = 0.46; validation cohort: \u003cem\u003er\u003c/em\u003e = 0.16; replication cohort \u003cem\u003er\u003c/em\u003e = 0.32; generalization cohort \u003cem\u003er\u003c/em\u003e = 0.22), the MJS response partially mediated the effect of the PINES response on moral wrongness ratings in the discovery cohort, and in the validation cohort the MJS response fully mediated the effect of the PINES response on moral wrongness ratings (Fig. 6f). The MJS response also partially mediated the effect of the PINES response on moral wrongness ratings in the replication cohort, whereas in the generalization cohort the MJS response did not mediate the effect of the PINES response on moral wrongness ratings (Extended Data Fig. 6a). In contrast, the PINES response failed to mediate the effect of the MJS response on wrongness ratings in discovery and validation cohorts (Fig. 6g) as well as in replication and generalization cohorts (Extended Data Fig. 6b). VIDS did not reliably predict graded moral wrongness ratings (discovery cohort: \u003cem\u003er\u003c/em\u003e = 0.15; validation cohort: \u003cem\u003er\u003c/em\u003e = 0.05; replication cohort \u003cem\u003er\u003c/em\u003e = 0.13; generalization cohort \u003cem\u003er\u003c/em\u003e = 0.12). Together, these findings underscore that neural representations of graded moral wrongness engage shared yet distinct neural representations underlying subjective experiences of negative emotions.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eSensitivity and specificity of the MJS for intentional versus accidental harms\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eHumans\u0026rsquo; moral wrongness judgments are reliably informed by inferring the beliefs and motives of observed agents\u003csup\u003e56\u003c/sup\u003e. Accordingly, across different types of moral transgressions, intentional (versus accidental) harms are robustly judged as more morally wrong\u003csup\u003e9,20\u003c/sup\u003e. Notably, in neurotypical populations (NT), the right TPJ exerts distinct neural activation patterns when judging intentional versus accidental harms, which has been shown to predict individual differences in subjects\u0026rsquo; moral judgments\u003csup\u003e9\u003c/sup\u003e. However, clinical populations characterized by difficulties with social interactions (e.g., Autism Spectrum Disorder, ASD) do not show this neural dissociation between intentional and accidental harms, underlining a disproportionate impairment on moral judgment tasks that require recursive mentalizing\u003csup\u003e9,57-58\u003c/sup\u003e. We explored whether the MJS responds more strongly to intentional as opposed to accidental harms, and whether this effect is diminished in individuals with ASD compared to NT populations. To this end, we performed predictions by applying the MJS pattern along with the PINES and VIDS to another independent fMRI dataset\u003csup\u003e9\u003c/sup\u003e where NT and ASD subjects judged the moral wrongness of intentional versus accidental harm scenarios (study 7, n = 39; Supplementary Table 1). We found that only the MJS could significantly predict intentional versus accidental harms in NT individuals (n=25; accuracy 76% (\u0026plusmn;15% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.43, sensitivity 73%, specificity 73%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.01). In contrast, PINES (accuracy 60% (\u0026plusmn;12% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.09, sensitivity 54%, specificity 54%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.42) as well as VIDS (accuracy 36% (\u0026plusmn;7% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.05, sensitivity 53%, specificity 53%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.23) could not significantly predict intentional versus accidental harms. Compellingly, neither the MJS (n=14; accuracy 50% (\u0026plusmn;13% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.09, sensitivity 57%, specificity 57%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 1.00), nor the PINES (accuracy 57% (\u0026plusmn;15% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.13, sensitivity 61%, specificity 61%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.79) or VIDS (accuracy 57% (\u0026plusmn;15% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.17, sensitivity 59%, specificity 59%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.79) could distinguish between intentional versus accidental harms in individuals with ASD, providing further evidence that clinical populations characterized by difficulties with social interactions rely on fundamentally different neural systems when evaluating moral transgressions as compared to NT individuals.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eValidating the moral judgment signature in visual scenes and social contexts\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFinally, we examined whether the moral judgment signature can predict graded moral wrongness judgments in visual scenes and social contexts. Humans frequently encounter morally relevant situations that they directly perceive\u003csup\u003e59\u003c/sup\u003e, even to an extent that moral stimuli \u0026ldquo;pop-out\u0026rdquo; in early visual perception\u003csup\u003e60\u003c/sup\u003e. Accordingly, we tested whether the MJS, which was developed solely on text-based moral vignettes, also captures graded moral judgments of real-world visual scenes. To this end, we performed predictions by applying the MJS pattern along with PINES and VIDS to two independent fMRI datasets (study 8, n = 30; study 9, n = 30) acquired while subjects morally judged photographic images sampled representatively from the socio-moral image database\u003csup\u003e35,43\u003c/sup\u003e (Supplementary Methods and Supplementary Table 1). Contrary to the vignette paradigms (studies 1\u0026ndash;4) in which stimulus presentation was not decoupled from graded moral judgment ratings to capture subjects\u0026rsquo; intuitive moral response, both visual scene tasks presented stimuli separately from the moral judgment period to better decouple the motor and moral judgment response (Extended Data Fig. 8). Furthermore, we deliberately included images depicting morally good actions and modified the graded moral judgment rating scale to range from very moral (1) to neutral (3) to very immoral (5), allowing us to further probe whether the MJS, which was only trained on sociomoral transgressions, can also discriminate between morally right and wrong stimuli.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs shown in Fig. 8a,b, the MJS could accurately predict graded moral image ratings in both samples: for individual subjects the average within-subject correlation between predicted and actual morality ratings (5 pairs of scalar values per subject) was \u003cem\u003er\u003c/em\u003e = 0.36 \u0026plusmn; 0.11 (standard error (SE), n= 30, study 8) and \u003cem\u003er\u003c/em\u003e = 0.41 \u0026plusmn; 11 (standard error (SE), n=30, study 9) and the overall (between- and within-subjects) prediction outcome was 0.14 (study 8, \u003cem\u003ep\u0026nbsp;\u003c/em\u003e= .08) and 0.31 (study 9, \u003cem\u003ep\u0026nbsp;\u003c/em\u003e= .0001). In addition, the MJS could predict moral versus immoral image ratings with high accuracy in both samples: study 8, accuracy 83% (\u0026plusmn;15% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.33, sensitivity 78%, specificity 78%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001; study 9, accuracy 83% (\u0026plusmn;15% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.68, sensitivity 86%, specificity 86%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001). In contrast, PINES (study 8, accuracy 43% (\u0026plusmn;8% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = \u0026ndash;0.07, sensitivity 45%, specificity 45%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.58; study 9, accuracy 73% (13% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.4, sensitivity 78%, specificity 78%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.02) and VIDS (study 8, accuracy 57% (\u0026plusmn;10% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.07, sensitivity 55%, specificity 55%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.58; study 9, accuracy 57% (10% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.15, sensitivity 61%, specificity 61%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.58) could not reliably predict moral versus immoral image ratings across both studies. These results indicated that the MJS captures neural representations of graded moral wrongness that generalize from judgments of text-based vignettes to visual photographic scenes and can discriminate between morally right and wrong stimuli.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo further test whether the MJS extends beyond third-party contexts captured by the moral vignettes and images and is also sensitive to second-party transgressions in social contexts, we applied the MJS to another independent fMRI dataset\u003csup\u003e29\u003c/sup\u003e that acquired neural responses during a series of unfair offers in a social exchange (ultimatum game) task (study 10, N = 43; Supplementary Table 1). As shown in Fig. 8b, the MJS did not predict high versus low unfairness (n=43, accuracy 35% (\u0026plusmn;5% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = \u0026ndash;0.17, sensitivity 44%, specificity 44%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.07). However, replicating previous work\u003csup\u003e29\u003c/sup\u003e, PINES could predict high versus low unfairness (accuracy 70% (\u0026plusmn;11% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.73, sensitivity 71%, specificity 71%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e = 0.01), as could VIDS (accuracy 74% (\u0026plusmn;11% s.e.m.), Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e = 0.81, sensitivity 74%, specificity 74%, two-sided binomial test \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001). Combined, these findings suggest that the MJS is not sensitive to unfair offers in a social context, yet it is relatively specific for moral judgments of visual scenes.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe relative severity of moral transgressions provides a fundamental compass for regulating human behavior, from everyday life\u003csup\u003e59,61-62\u003c/sup\u003e to the courtroom\u003csup\u003e1\u003c/sup\u003e, yet how the human brain represents and computes gradations of moral wrongness remains a longstanding neuroscientific question\u003csup\u003e2,5,50,63-65\u003c/sup\u003e. Evolutionary frameworks postulate an innate neural circuitry which evolved to detect and punish moral transgressors\u003csup\u003e1,3,66-67\u003c/sup\u003e. On the other hand, decades of research demonstrate that the subjective experience of graded moral wrongness for distinct moral violations varies considerably across individuals\u003csup\u003e13,68-69\u003c/sup\u003e, neurodiverse populations\u003csup\u003e9,58\u003c/sup\u003e, and cultures\u003csup\u003e70-71\u003c/sup\u003e. Thus, it remains unclear whether brain representations of graded moral wrongness are instantiated via a shared or idiographic neural code, and which neural systems reliably contribute to graded computations of moral wrongness.\u003c/p\u003e\n\u003cp\u003eHere we combined fMRI with predictive modeling approaches designed to uncover whether neural representations of graded moral wrongness are (a) generalizable across individuals, MRI systems, two different cultures, and neurodiverse populations, (b) common or domain-specific across different kinds of moral norm transgressions, (c) sensitive and specific to moral wrongness versus general negative affect or subjective disgust, and (d) robust across multimodal experimental paradigm variations. In study 1, we applied SVR to identify and evaluate a distributed, whole-brain neural signature (MJS) for predicting graded moral wrongness judgments, with reliable contributions from cortical (e.g., dmPFC, vmPFC, and insula) and subcortical (e.g., amygdala, thalamus, and caudate) regions previously linked to human moral cognition\u003csup\u003e5,10-12\u003c/sup\u003e. By jointly estimating common (general) and domain-specific representations of moral wrongness via PLS-R across eight theory-derived categories of (socio)moral transgressions\u003csup\u003e34\u003c/sup\u003e, we found that computations of graded moral wrongness are encoded in a combination of general (common) and domain-specific representations. Studies 2\u0026ndash;4 provided further evidence that brain representations of moral wrongness severity are generalizable across different kinds of moral norm violations, as well as individuals, cohorts, MRI systems, and two different cultures.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMoreover, our findings contribute to ongoing debates concerning the role of affect in human moral cognition\u003csup\u003e31,39-42,72\u003c/sup\u003e, demonstrating that neural representations of moral wrongness and general negative emotion (study 5) and subjective disgust (study 6) exhibit shared yet separable representations, with the MJS response mediating the association between the PINES response and graded moral wrongness judgments. Attesting to the potential clinical relevance of our moral judgment biomarker for future translational applications, we found that the MJS could accurately discriminate between moral judgments of intentional versus accidental harms in neurotypical adults, but not in individuals diagnosed with autism spectrum disorder (study 7). Finally, from a biomarker perspective, it is imperative that a neuroaffective signature captures the respective mental process across variations of experimental contexts\u003csup\u003e73\u003c/sup\u003e. Accordingly, we showed that brain representations of graded moral wrongness derived from text-based moral vignettes generalize to graded moral judgment of visual scenes (studies 8\u0026ndash;9), albeit not to unfair offers in second-party social interactions probed via the ultimatum game (UG; study 10). It remains a matter of debate to what extent the UG probes \u003cem\u003emoral\u003c/em\u003e preferences and decisions as it has been argued that the essence of a moral transgression is an intentional agent causing harm to a suffering moral patient\u003csup\u003e74-75\u003c/sup\u003e; features that are salient across the vignette scenarios used for developing the MJS. Although the UG likely captures intentional decisions, what is intended is gain to the player and not necessarily intended harm to the recipient. Thus, whether this task induces suffering and is construed as moral is debatable: Given that the worst possible outcome for a recipient in an ultimatum game is to receive nothing, and even putatively \u0026ldquo;unfair\u0026rdquo; transfers in the ultimatum game (i.e., \u0026lt; 50%) yield benefits for the recipient, it may be inappropriate to construe the ultimatum game as a \u003cem\u003emoral\u003c/em\u003e task paradigm\u003csup\u003e76\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConcerning the neural basis of moral cognition, it is now commonly understood that moral judgment recruits multiple domain-general systems that are distributed throughout the brain\u003csup\u003e50,64\u003c/sup\u003e. Aligned with this distributed account, in a series of analyses combining both univariate and multivariate analyses, our findings confirm that graded computations of moral wrongness require concerted engagement of brain-wide distributed representations with comparably strong contributions of cortical systems engaged in mentalizing, valuation, and mental imagery such as dorsomedial and ventromedial prefrontal cortex, as well as precuneus and visual cortex\u003csup\u003e8,9,13,77\u003c/sup\u003e, and subcortical regions involved in rapid threat detection, punishment, and avoidance responses (amygdala, thalamus and caudate)\u003csup\u003e3,78-79\u003c/sup\u003e. While no single network reached the predictive performance of the whole-brain MJS model for predicting graded computations of moral wrongness, the visual network exhibited comparably strong contributions, underlining the importance of visual imagery for moral judgment\u003csup\u003e45-46\u003c/sup\u003e and representations of affective stimuli\u003csup\u003e55\u003c/sup\u003e. Consistent with previous predictive models for subjective emotional experiences\u003csup\u003e27-29\u003c/sup\u003e, our analyses revealed that approximately 10,000 voxels that were randomly sampled across the whole brain could lead to high predictive performance for moral wrongness predictions. Together, our results complement a growing body of research from both univariate\u003csup\u003e6,8,19\u003c/sup\u003e and multivariate perspectives\u003csup\u003e13-16\u003c/sup\u003e demonstrating that capturing moral cognition requires integration across multiple distributed neural systems. Analogously, the distributed representation perspective aligns with modular\u003csup\u003e80-81\u0026nbsp;\u003c/sup\u003eand constructionist\u003csup\u003e82\u003c/sup\u003e theories of morality that propose that shared but also distinct distributed functional assemblies integrate to facilitate subjective moral judgment.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMoreover, sentimentalist and rationalist perspectives disagree whether affect serves as input, output, or constituent element of moral cognition\u003csup\u003e6,31,39,41-42,72\u003c/sup\u003e. Our results show that computations of moral wrongness exhibit shared yet separable neural representations with established predictive models for the subjective experience of non-specific negative affect or disgust, such that all models engaged subcortical regions and regions implicated in emotional awareness and appraisal, whereas the neural signature of moral wrongness robustly mediated the response of the PINES on subjective moral wrongness ratings but not vice versa. Although the three neural signatures showed a certain extent of similarity in the range of intense emotional experiences (for a similar observation, see also\u003csup\u003e29\u003c/sup\u003e), the MJS (as well as PINES and VIDS) were more sensitive to the target construct in the low versus high intensity range. However, the MJS outperformed other brain markers in predicting moral judgments in response to both intentional versus accidental harm scenarios and socio-moral images, while it did not track responses to unfair offers.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Applications of our moral judgment signature to new studies include (a) characterizing individual differences in brain representations of moral judgement severity related to sociodemographics, hormonal fluctuations, disorders, and subgroups, (b) predicting or monitoring the development and progression of moral wrongness representations over time, (c) establishing the sensitivity and specificity of the MJS with regard to alternative moral judgment types, such as norm and blame judgments\u003csup\u003e4\u003c/sup\u003e as well as additional mental processes known to modulate moral judgment, including empathic care and distress\u003csup\u003e83\u003c/sup\u003e, and (d) decoding spontaneous (implicit) moral condemnation in naturalistic settings\u003csup\u003e84-85\u003c/sup\u003e. Further validation will help refine the use cases and boundary conditions for such applications, with a particular focus on their implications for neuroethics\u003csup\u003e86-88\u003c/sup\u003e .\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur study has several limitations. First, the primary datasets used for developing and validating the MJS (studies 1\u0026shy;4) only included subjects from the United States and the Netherlands, raising the question whether the MJS can also predict subjective moral judgments in individuals from cultures that are not Western, Educated, Industrialized, Rich, and Democratic (WEIRD)\u003csup\u003e70\u003c/sup\u003e. Second, although the experimental vignettes used for developing the MJS cover a broad range of different moral scenarios, they feature decontextualized scenarios of \u0026ldquo;raceless, genderless strangers\u0026rdquo;\u003csup\u003e89\u003c/sup\u003e from a third-party perspective and hence do not exhaust humans\u0026rsquo; rich everyday moral experience. Future studies must consider an array of additional contextual factors that may modulate brain representations of graded moral wrongness, including the (social) identities of agents and victims as well as their intentions, motives, and character\u003csup\u003e56\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn conclusion, we show that graded computations of moral severity are neurally encoded in a sensitive, specific, and distributed neural signature, which robustly generalizes across individuals, cohorts, experimental variations, and MRI systems. Our resulting brain markers for graded moral wrongness judgments provide robust measures that can serve to understand pathological moral cognition, track the development of brain representations of moral severity over time and across two Western cultures, and advance the precision and generalizability of moral neuroscience. They also lead to further basic and translational research questions. One area for future development concerns the identification of cerebral sources of individual differences in graded moral wrongness representations\u003csup\u003e90-91\u003c/sup\u003e, which may provide potential targets for personalized assessment of aberrant moral cognition\u003csup\u003e92\u003c/sup\u003e and serve as interventions for morally rooted neural polarization\u003csup\u003e93\u003c/sup\u003e. Likewise, illuminating how graded moral wrongness is represented and integrated across modalities, from written scenarios and photographic images to auditory stories and audiovisual movies\u003csup\u003e34,84-85,94\u0026nbsp;\u003c/sup\u003emay reveal how degrees of moral wrongness are hierarchically computed across sensory modalities, helping to explain, for example, where in the brain pathologies interfere with moral cognition\u003csup\u003e95-96\u003c/sup\u003e. Considering that condemnation of what are seen as moral transgressions is a human universal\u003csup\u003e1,3\u0026nbsp;\u003c/sup\u003ethis systematic evaluation provides answers to fundamental questions in the cognitive neurosciences, namely that there exists a sensitive and specific neural code for computing graded moral wrongness, which is distributed throughout the brain and robustly tracks moral severity across varied scenarios, individuals, and cultures.\u0026nbsp;\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003e\u003cstrong\u003eEthics\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe present study includes 10 datasets (Supplementary Table 1). Among them, study 2 (validation cohort), study 4 (generalization cohort), study 9 (visual scenes 1), and study 10 (visual scenes 2) were new and original experiments designed and implemented by the authors. Informed consent was obtained before each of these experiments. Corresponding experimental protocols were approved by the University of California at Santa Barbara (study 2 and study 8 with the approval numbers 21-17-0123 and 32-25-0079) and the University of Amsterdam (study 4 and study 9 with the approval number FMG-3172). Subjects in study 2 and study 8 were compensated with $50 USD, those in study 4 and study 9 with 25\u0026euro;. The current work also includes secondary analysis of previously acquired experiments based on anonymized data (that is, the remaining six studies; Supplementary Table 1). All subjects in these studies provided informed consent in line with local ethics and institutional review boards. Detailed descriptions of the ethics approval and information on subject compensation are available through the corresponding references.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eSubjects\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eDiscovery cohort\u003c/em\u003e.\u0026nbsp;\u003c/strong\u003eDetails of the discovery cohort were reported in a previous study\u003csup\u003e13\u003c/sup\u003e. Briefly, 64 native English speakers were recruited from the University of California Santa Barbara (UCSB) Department of Communication subject pools and from the local Santa Barbara community. Exclusion criteria included a history of systemic or neurological disorders, psychiatric disorders, psychoactive medication or drug use and pregnancy (33 males; mean \u0026plusmn; s.d. age 20.78 \u0026plusmn; 2.45 years).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eValidation cohort\u003c/em\u003e.\u003c/strong\u003e 31 healthy volunteers were recruited from the University of California, Santa Barbara. Exclusion criteria included a history of systemic or neurological disorders, psychiatric disorders, psychoactive medication or drug use and pregnancy. One subject was excluded due to excessive head motion, resulting in a final sample of 30 (11 males; mean \u0026plusmn; s.d. age 20.46 \u0026plusmn; 2.33 years).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eReplication cohort\u003c/em\u003e.\u003c/strong\u003e Details of the replication cohort were reported in a previous study\u003csup\u003e14\u003c/sup\u003e. Briefly, 30 native English speakers were recruited from the Durham, North Carolina area. Exclusion criteria included a history of psychiatric or neurological disorders. Data from the first three subjects was excluded from analysis because of an error in the experiment script, resulting in a final sample of 27 (14 male, mean \u0026plusmn; s.d. age 24.65 \u0026plusmn; 4.21 years).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eGeneralization cohort\u003c/em\u003e.\u003c/strong\u003e Healthy volunteers were recruited from the University of Amsterdam\u0026rsquo;s subject pool and the local Amsterdam community. Exclusion criteria included a history of systemic or neurological disorders, psychiatric disorders, psychoactive medication or drug use and pregnancy. For the fMRI study, we recruited 31 Dutch subjects, and data from 1 subject was excluded due to errors with stimulus presentation, resulting in a final sample of 30 (12 males; mean \u0026plusmn; s.d. age 26.4 \u0026plusmn; 9.17 years).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eStimuli and paradigm used in discovery, validation, replication, and generalization cohorts\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eSubjects in the discovery and replication cohorts were presented with the original set of Moral Foundations Vignettes (MFV)\u003csup\u003e34\u003c/sup\u003e while undergoing fMRI, whereas subjects in the generalization cohort were presented with the validated Dutch version\u003csup\u003e35\u003c/sup\u003e of the MFV (Extended Data Fig. 1). The MFV span 120, one sentence descriptions (14\u0026ndash;17 words) detailing the violation of one (and only one) of seven moral foundations: physical care, emotional care, fairness, liberty, loyalty, authority and sanctity. The vignettes also contain a non-moral, social norm transgression category. Each of the eight conditions features 15 vignettes. In the validation cohort, subjects rated a total of 72 vignettes, of which 42 were drawn from the MFV (6 per category). The remaining 30 vignettes were newly created by the authors to depict violations of six moral domains drawn from the Morality as Cooperation (MAC)\u003csup\u003e97\u003c/sup\u003e framework, including family, reciprocity, bravery, property, respect, and group values (5 per category).\u003c/p\u003e\n\u003cp\u003eVignettes were organized in an event-related design, randomly distributed over three ~8 min functional runs. Subjects viewed one vignette at a time and were instructed to vividly imagine the described scene. While the vignette was on screen, subjects were asked to make a judgment of how morally wrong the action described in the vignette was using an MRI-safe button box (not morally wrong (1) to extremely morally wrong (4)). Subjects had ~8s to read and make judgments of each vignette. Detailed paradigm variations across datasets are reported in Extended Data Fig. 1.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eMRI data acquisition and preprocessing\u003c/strong\u003e\u003c/h2\u003e\n\u003ch2\u003e\u003cstrong\u003eDiscovery and validation cohort\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003efMRI scanning was performed on a 3T Siemens Magnetom Prisma with a Siemens head coil, at the Brain Imaging Centre of the University of California, Santa Barbara. Functional images were taken using a multiband echo-planar gradient sequence (TR, 720 ms; echo time, 37 ms; flip angle, 52\u0026deg;; field of view, 208 mm; acceleration factor, 8). Volumes consisted of 72 interleaved slices (2 mm isotropic) acquired with an angle of ~20\u0026deg; relative to the AC\u0026ndash;PC plane, so that the slices are acquired more dorsally near the eyes relative to the back of the brain (in that fashion we were able to acquire the entire brain volume including the cerebellum for every subject). High-resolution T1-weighted whole-brain acquisitions were collected before functional image acquisition (TR, 2,500 ms; echo time, 2.22 ms; flip angle, 7\u0026deg;; field of view, 241 mm; 0.9 mm, isotropic resolution). All data was preprocessed using fMRIprep\u003csup\u003e98\u003c/sup\u003e version 24.1.1 (Supplementary Methods).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eReplication cohort\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eScanning was conducted on a research-dedicated 3T GE MR750 scanner with an 8-channel head coil. The scanning session began with a high-resolution T1-weighted structural scan followed by a five-minute resting state scan and three runs of the moral judgment task. Functional scans were collected using a whole-brain spiral-in sequence (TR = 2s, TE = 30 ms, flip angle = 70\u003csup\u003eo\u003c/sup\u003e). Slices were acquired in an interleaved fashion, and subjects\u0026apos; heads were kept in place with cushions to limit head motion. Each run began and ended with 10 seconds (5 TRs) of fixation that were dropped from analysis. The scanning session concluded with five minutes of a localizer task for emotional faces. Data from the resting state and localizer task are not reported here. All data was preprocessed using fMRIprep\u003csup\u003e98\u003c/sup\u003e version 24.1.1 (Supplementary Methods).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eGeneralization cohort\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eData for this sample was collected on a 3T Philips Achieva scanner, with dStream architecture and 32-channel head coil at the Spinoza center located at the University of Amsterdam\u0026rsquo;s Roeterseiland Campus. High-resolution T1-weighted whole-brain acquisitions were collected before functional image acquisition (TR, 8.2 ms; echo time, 3.7 ms; flip angle, 8\u0026deg;; field of view, 240 mm). Functional MRI data were acquired using a multiband echo-planar gradient sequence (TR, 720 ms; echo time, 30 ms; flip angle, 55\u0026deg;; field of view, 216 mm; acceleration factor, 4) and all volumes consisted of 44 axial (\u0026ldquo;ascending\u0026rdquo;) slices. The task was projected into the scanner and viewed by subjects with a mirror placed above the head coil. The experimental task was programmed using Psychopy (version 2022.1.2)\u003csup\u003e99\u003c/sup\u003e, which was also used to collect behavioral responses on an MRI-safe 4-button box. All data was preprocessed using fMRIprep\u003csup\u003e98\u003c/sup\u003e version 24.1.1 (Supplementary Methods).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eFirst-level fMRI analysis used in the discovery, validation, replication, and generalization cohorts\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eWhole-brain univariate GLM (general linear model) analyses were conducted. Each run started with a tail of 11 repetition times (TRs) which were discarded. Thereafter, preprocessed images were spatially smoothed using a Gaussian filter (full-width half-maximum, 8 mm kernel). We conducted two separate subject-level GLM analyses in which the three runs were modeled by separate regressors in the same GLM. To account for residual variance, the temporal derivative of each condition regressor was added in both GLMs as well as a constant regressor for each entire run. The first GLM model was used to obtain beta images for the prediction analysis. In this model we included four separate boxcar regressors time-logged to the presentations of vignettes (7.92s) in each rating (i.e., 1\u0026ndash;4), which allowed us to model brain activity in response to each moral judgment level separately. The second GLM modeled the vignette viewing period and the design matrix also included moral wrongness ratings (1\u0026minus;4) reported for each vignette as a parametric modulator for the vignette viewing period.\u003c/p\u003e\n\u003cp\u003eAll task regressors were convolved with the canonical haemodynamic response function and a standard high-pass filter (90s cutoff) was applied to exclude low-frequency drifts. Regressors of non-interest (nuisance variables) included (1) six head movement parameters and their squares, their derivatives and squared derivatives (leading to 24 motion-related nuisance regressors in total) and (2) vectors indicating motion outlier timepoints.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eMultivariate pattern analysis.\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eWe applied whole-brain multivariate machine-learning pattern analysis to obtain a pattern of brain activity that best predicted subjects\u0026rsquo; self-reported graded moral wrongness ratings. We employed support vector regression (linear kernel with C=1)\u003csup\u003e27-29\u0026nbsp;\u003c/sup\u003eimplemented in the NLTools Python package (v.0.4.5)\u003csup\u003e100\u003c/sup\u003e with individual beta maps (one per rating for each subject) as features to predict subjects\u0026rsquo; continuous moral wrongness ratings of the grouped vignettes they viewed while undergoing fMRI. We only used data from the discovery cohort to develop the MJS. To evaluate the performance of our algorithm, we used a leave-one-subject-out (LOSO) cross-validation procedure, ensuring that every subject served as both training and testing data. This allowed us to evaluate how a model trained on 63 subjects could predict the rating level associated with each of the four beta maps from the left-out subject\u003csup\u003e22,25\u003c/sup\u003e. To facilitate a robust determination of the predictive accuracy of the neurofunctional signature we employed various metrics including correlation, RMSE, and forced-choice classification accuracy. Specifically, we used overall (between- and within-subjects; 256 pairs in total) and within-subject (4 pairs per subject) Pearson correlations between the cross-validated predictions and the actual ratings to indicate the effect sizes and the RMSE to illustrate overall prediction error. In addition, we assessed classification accuracy of the MJS using a forced-choice test, where signature responses were compared for two conditions tested within the same individual, and the higher was chosen as more morally wrong. We also applied the moral judgment-predictive pattern (trained on the whole discovery cohort) to the validation, replication, and generalization cohorts to obtain a signature response for each map (that is, the dot-product of the MJS weight map and the test image plus the intercept) to assess the prediction performance of the MJS.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eWithin-subject trial-wise prediction\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eHere we tested whether the MJS could predict individual trial-by-trial moral wrongness judgments. To this end we performed a single-trial analysis, which was achieved by specifying a GLM design matrix with separate regressors for each stimulus (vignette). Each task regressor was convolved with the canonical hemodynamic response function. Nuisance regressors and high-pass filters were identical to the above GLM analyses. Next, we calculated the MJS pattern expressions of these single-trial beta maps (i.e., the dot-product of vectorized activation images with the MJS weights), which were finally correlated with the true ratings for each subject separately. For subjects in the discovery cohort we again used the LOSO cross-validation procedure to obtain the MJS response of each single-trial beta map for each subject.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePLS-R for common and domain-specific moral wrongness model development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe developed common and domain-specific predictive models of moral wrongness ratings from brain activity across the eight vignette conditions using PLS-R\u003csup\u003e36-37\u003c/sup\u003e. PLS-R estimates a set of latent brain components (voxel-wise spatial maps) and a set of latent moral wrongness rating factors that are optimized to be maximally intercorrelated (that is, maximal variance in ratings explained by brain patterns). Compared to standard predictive brain models that typically characterize a single outcome at a time, PLS-R jointly estimates multiple solutions (that is, separate brain patterns for common and domain-specific outcomes) simultaneously, which is why it is capable of predicting multiple (correlated) stimulus types, as is the case with our data\u003csup\u003e24\u003c/sup\u003e. The predictors (brain activity) are stored in the input matrix \u003cstrong\u003eX\u003c/strong\u003e and the outcome variables (ratings) are stored in the matrix \u003cstrong\u003eY\u003c/strong\u003e. By an iterative application of a singular value decomposition algorithm, which factorizes (decomposes) the cross-product matrix of the two input matrices, PLS-R finds latent variables, also called component scores, that model \u003cstrong\u003eX\u003c/strong\u003e (for example, brain activity) and simultaneously predict \u003cstrong\u003eY\u0026nbsp;\u003c/strong\u003e(for example, ratings). Each run of the singular value decomposition algorithm produces orthogonal latent variables and corresponding regression weights for predictions. By estimating different latent sources, PLS-R can provide improved estimates of common and specific patterns (versus single-outcome models such as SVR), but these are not necessarily fully independent of each other.\u003c/p\u003e\n\u003cp\u003ePLS-R was conducted using \u0026lsquo;PLSRegression\u0026rsquo; in sklearn (version 1.6.1)\u003csup\u003e101\u003c/sup\u003e. Predictors (\u003cstrong\u003eX\u003c/strong\u003e) constituted 1,468 whole-brain activation maps associated with subjects (64) x moral wrongness level (~4) x vignette condition (8), aggregated into an images x voxels matrix (stacked across subjects), and split into training and test sets using leave-one-subject-out cross-validation. The activation maps were derived via univariate GLM analysis in which we modeled the wrongness ratings of each vignette condition against all other event regressors and averaged across trials of each rating level of each vignette condition to obtain ~32 contrast images for each subject. Because not every subject used every wrongness rating option for every vignette condition, the final number of whole-brain activation maps (1,468) was smaller than the theoretical maximum (2,048). We constructed the outcome (\u003cstrong\u003eY\u003c/strong\u003e) matrix to include moral wrongness ratings across all vignette conditions (\u003cstrong\u003eY\u003csub\u003e1\u003c/sub\u003e\u003c/strong\u003e) as well as domain-specific wrongness (\u003cstrong\u003eY\u003csub\u003e2\u003c/sub\u003e\u0026ndash; Y\u003csub\u003e9\u003c/sub\u003e\u003c/strong\u003e, ratings for each vignette condition separately, with values of 0 for other stimulus types). By setting the \u003cstrong\u003eY\u003c/strong\u003e value of other stimulus types to 0 we constrained each pattern to be domain specific. The linear combination of latent brain factors that explains \u003cstrong\u003eY\u003csub\u003e1\u003c/sub\u003e\u003c/strong\u003e reflects a common model of moral wrongness across vignette conditions. Likewise, brain patterns predictive of \u003cstrong\u003eY\u003csub\u003e2\u003c/sub\u003e\u0026ndash; Y\u003csub\u003e9\u003c/sub\u003e\u003c/strong\u003e are models optimized to be selective to violations of physical care, emotional care, fairness, liberty, loyalty, authority, sanctity, and social norms. Each model was then projected into a single predictive spatial map (see also Extended Data Fig. 4).\u003c/p\u003e\n\u003cp\u003eTo evaluate the models\u0026rsquo; performance, we compared the average model response (predicted wrongness rating) of its target outcome (that is, the four average ratings per vignette condition) with the average model response of off-target outcomes via \u003cem\u003et\u003c/em\u003e tests. In addition, we estimated in each subject the Pearson correlation (\u003cem\u003er\u003c/em\u003e) between observed and cross-validated predicted wrongness ratings and tested whether the output of one model predicted the specific type of wrongness it was trained to predict (sensitivity) and not other types (specificity).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eDetermining brain regions associated with and predictive of moral wrongness judgments\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTo determine which brain areas made reliable contributions to graded moral wrongness computations and to threshold voxel weights for interpretation and display, we constructed 10,000 bootstrap samples (with replacement) consisting of paired brain and wrongness data from the discovery cohort and performed SVR on each sample. The z-scores at each voxel were estimated based on the mean and standard error of the bootstrap distributions, and the statistical map was thresholded based on the corresponding \u003cem\u003eP\u003c/em\u003e values. The corresponding map was thresholded voxel-wise at \u003cem\u003eq\u0026nbsp;\u003c/em\u003e\u0026lt; 0.05 (FDR-corrected). Next, model encoding (\u0026lsquo;structure coefficient\u0026rsquo;) maps were computed for each subject by regressing the SVR model predictions on voxel-wise fMRI activation maps (four maps per person, corresponding to averages for each moral wrongness level). Structure coefficients identify voxels individually associated with the model\u0026rsquo;s output, mapping individual voxels to the overall multivariate model prediction\u003csup\u003e38\u003c/sup\u003e. The analysis was performed using a standard summary statistics-based mixed-effects GLM, with robust regression at the second level, thresholded at FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 corrected for multiple comparisons. The core system map for graded moral wrongness computations was derived via a conjunction of the model weight map thresholded at FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 created during the bootstrap and the model encoding map thresholded at FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05.\u003c/p\u003e\n\u003cp\u003eNext, we performed a one-sample \u003cem\u003et\u0026nbsp;\u003c/em\u003etest on the first-level univariate parametric modulation beta maps to see which brain regions\u0026rsquo; activation was associated with moral wrongness ratings. In addition to evaluating the population-level model (MJS), we also probed the consistency of each weight for every voxel in the brain across within-subject multivariate classifiers. To this end we performed a prediction analysis (linear SVR with C = 1) for each subject in the discovery cohort separately using their single-trial data (10-fold cross-validated). We then used a one-sample \u003cem\u003et\u003c/em\u003e test (FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05) on the resulting subject-wise weightmaps to identify which voxels reliably contribute to moral wrongness predictions across subjects. We again created model encoding maps (structure coefficients) for the within-subject models by regressing the within-subject SVR model predictions on voxel-wise fMRI activation maps (~120 maps per person, corresponding to each vignette). This analysis was again performed using a standard summary statistics-based mixed-effects GLM, with robust regression at the second level, thresholded at FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 corrected for multiple comparisons. The core system map for within-subject graded moral wrongness computations was derived via a conjunction of the model weight map thresholded at FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 obtained from the one-sample \u003cem\u003et\u003c/em\u003e-test across within-subject weightmaps and the model encoding map thresholded at FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05 obtained from subjects\u0026rsquo; trial-wise beta maps.\u003c/p\u003e\n\u003cp\u003eFurthermore, we asked whether moral wrongness computations could be reducible to activations in a single brain region (e.g., dmPFC) or network (e.g., DMN). To examine this hypothesis, we employed whole-brain searchlight (three-voxel radius spheres)\u0026mdash;and parcellation (279 cortical and subcortical regions)\u003csup\u003e102\u003c/sup\u003e\u0026mdash;based analyses to identify local regions predictive of moral wrongness and compared model performances of local regions with the whole-brain model (i.e., the MJS). In addition, we compared prediction performances of dmPFC (based on a whole-brain parcellation of the coactivation patterns of activations across over 10,000 published studies available in the Neurosynth database; available at https://neurovault.org/images/39711/) and large-scale networks to the whole-brain approach. The networks of interest included seven resting-state functional networks\u003csup\u003e103\u003c/sup\u003e, a subcortical network (including the striatum, thalamus, hippocampus and amygdala) and a \u0026lsquo;consciousness network\u0026rsquo;\u003csup\u003e104\u003c/sup\u003e. To reduce potential biases arising from different atlases, we continued to use the modified 279-region version of the Brainnetome Atlas (which also combined Yeo\u0026rsquo;s seven networks) to extract the nine networks. For these analyses we trained and tested a model for each searchlight sphere, parcellation, brain region or network separately using the discovery data (LOSO cross-validated).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eComparing the performance of MJS with the PINES and VIDS\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003ePrevious studies have developed and evaluated whole-brain emotional decoders for general negative emotion experience (PINES)\u003csup\u003e25\u003c/sup\u003e and subjective disgust experience (VIDS)\u003csup\u003e29\u003c/sup\u003e. To compare the performance of MJS with the PINES and the VIDS, we applied the three decoders to the discovery, validation, replication, generalization,\u003c/p\u003e\n\u003cp\u003ePINES holdout test\u003csup\u003e25\u003c/sup\u003e (study 5, n = 61; see Supplementary Table 1 for details) and VIDS validation\u003csup\u003e29\u003c/sup\u003e (study 6, n = 30; Supplementary Table 1) cohorts and assessed the overall as well as within-subject prediction\u0026ndash;outcome correlations between the pattern expressions and the true ratings. Two-alternative forced-choice classification accuracies between the separate moral wrongness judgment levels (and general negative emotion as well as disgust) based on the pattern expressions were further calculated.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eSpatial similarity between stable decoding maps and a priori ROIs as well as networks of interest\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eRiver plots were created to illustrate spatial similarity between stable decoding maps derived from bootstrap tests and a priori ROIs previously documented as regions linking to moral judgment, negative affect processes, and disgust. We further depicted spatial similarity between stable decoding maps and seven large-scale cerebral networks, the subcortical and consciousness networks. In line with a recent study\u003csup\u003e24\u003c/sup\u003e, spatial similarity was computed as cosine similarity between the ROI or network and the thresholded MJS, PINES, and VIDS (FDR \u003cem\u003eq\u003c/em\u003e \u0026lt; 0.05, retaining positive values) from bootstrap tests.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eMulti-level two-path mediation analysis\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTo explore the relationship between MJS response, moral wrongness rating and PINES response, multi-level two-path mediation analyses were performed using the Mediation Toolbox, available via MediationToolbox\u003csup\u003e105\u003c/sup\u003e(see also \u003csup\u003e27-29,106\u003c/sup\u003e). Briefly, the mediation analysis examines whether the observed covariance between the independent/predictor variable (\u003cem\u003eX\u003c/em\u003e) and the dependent/outcome variable (\u003cem\u003eY\u003c/em\u003e) can be explained by the third variable (\u003cem\u003eM\u003c/em\u003e, also mediator). The predictor\u0026ndash;mediator relation, mediator\u0026ndash;outcome relation and predictor\u0026ndash;outcome relation before and after controlling for the mediator are characterized by paths \u003cem\u003ea\u003c/em\u003e, \u003cem\u003eb\u003c/em\u003e, \u003cem\u003ec\u003c/em\u003e and \u003cem\u003ec\u0026prime;\u003c/em\u003e, respectively. Specifically, the total effect of the predictor on the outcome (path \u003cem\u003ec\u003c/em\u003e) is the sum of direct/non-mediation effect (path \u003cem\u003ec\u0026prime;\u003c/em\u003e) and indirect/mediation effect (the product of the path coefficients of path \u003cem\u003ea\u003c/em\u003e and path \u003cem\u003eb\u003c/em\u003e, that is, \u003cem\u003ea\u003c/em\u003e \u0026times; \u003cem\u003eb\u003c/em\u003e). A significant mediation effect is obtained when \u003cem\u003ea\u003c/em\u003e, \u003cem\u003eb\u003c/em\u003e and \u003cem\u003ea\u003c/em\u003e \u0026times; \u003cem\u003eb\u003c/em\u003e are all significant. Furthermore, when \u003cem\u003ec\u0026prime;\u003c/em\u003e is significant, \u003cem\u003eM\u003c/em\u003e (that is, the mediator) is considered to have a partial mediation effect; otherwise, \u003cem\u003eM\u003c/em\u003e plays a full mediation role. In this study, we constructed two multi-level mediation analyses: (1) the trial-by-trial PINES responses were entered as predictors (\u003cem\u003eX\u003c/em\u003e), moral wrongness ratings were entered as outcomes (\u003cem\u003eY\u003c/em\u003e), and the trial-by-trial MJS responses were entered as mediators (\u003cem\u003eM\u003c/em\u003e); (2) the trial-by-trial MJS responses were entered as predictors (\u003cem\u003eX\u003c/em\u003e), moral wrongness ratings were entered as outcomes (\u003cem\u003eY\u003c/em\u003e), and the trial-by-trial PINES responses were entered as mediators (\u003cem\u003eM\u003c/em\u003e). To do this, the MJS and PINES responses were calculated by the dot product of the single-trial beta maps with the MJS and PINES patterns, respectively, for each subject across discovery (cross-validated), validation, replication, and generalization cohorts. Bootstrap tests with 10,000 iterations were used to assess the statistical significance of mediation effects. If the bootstrapped 95% CI does not include zero, the effect will be considered to be significant (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eValidation in intentional vs. accidental harm scenarios and neurodiverse populations\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eWe examined whether the MJS can discriminate between intentional versus accidental harms in both neurotypical (NT) and individuals diagnosed with autism spectrum disorder (ASD) by applying the MJS, PINES, and VIDS pattern to another independent fMRI dataset\u003csup\u003e9\u003c/sup\u003e during which subjects read 60 vignettes in the second-person point of view that either portrayed intentional or accidental harm violations (n = 39; NT = 25, ASD = 14; Supplementary Methods and Supplementary Table 1). For each brain model (MJS, PINES, and VIDS), we computed the classification accuracy between intentional and accidental harms from receiver operating characteristic curves using forced-choice classification (average of all intentional harms versus average of all accidental harms). \u003cem\u003eP\u003c/em\u003e values were calculated using a two-sided independent binomial test.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eValidation in visual scenes\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTo test whether the MJS\u0026mdash;developed on text-based vignettes\u0026mdash;can generalize to visual scenes, we designed and implemented two new fMRI experiments (study 8, n = 30; and study 9, n = 30; Supplementary Methods and Supplementary Table 1) employing photographic stimuli\u003csup\u003e35,43\u003c/sup\u003e (design based on previous similar studies\u003csup\u003e27-29\u003c/sup\u003e). Next, we applied the MJS\u0026mdash;as well as the PINES and VIDS\u0026mdash;to the visual scenes fMRI data. Specifically, we calculated Pearson correlation coefficients across actual and predicted moral judgments as well as forced-choice classification accuracies between moral and immoral image ratings based on the pattern responses.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eValidation in the social context\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eFirst, to examine whether the MJS\u0026mdash;developed during third-party moral wrongness judgments of concrete moral transgressions\u0026mdash;could be extended into a second-party social context frequently linked to socio-moral disgust (for example, unfairness\u003csup\u003e29,107-108\u003c/sup\u003e), we applied the MJS pattern to another independent fMRI dataset during which subjects were confronted with a series of unfair offers in an ultimatum game task (study 10, n = 43; Supplementary Methods and Supplementary Table 1). Specifically, we calculated forced-choice classification accuracies between high (unfairness level 5) and low (unfairness level 1) unfairness based on the pattern responses.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eData availability\u003c/h2\u003e\n\u003cp\u003efMRI data (beta maps) used to train and validate the signature are available via figshare at https://figshare.com/articles/dataset/Discovery_dataset_mjs/29423726?file=55720082 (ref. 109) (study 1); https://figshare.com/articles/dataset/Validation_dataset_mjs/29423789?file=55721972 (ref. 110) (study 2); https://figshare.com/articles/dataset/Replication_dataset_mjs/29423966?file=55724255 (ref. 111) (study 3); and https://figshare.com/articles/dataset/Generalization_dataset_mjs/29423981?file=55724291 (ref. 112) (study 4). The data of study 5 are from a previous study\u003csup\u003e25\u003c/sup\u003e and are available via NeuroVault at https://neurovault.org/collections/1964 (ref. 113). The data of study 6 are from a previous study\u003csup\u003e29\u003c/sup\u003e and are available at figshare https://figshare.com/articles/dataset/validation_dataset_disgust/22841117 (ref. 114). The data of study 7 are from a previous study\u003csup\u003e9\u003c/sup\u003e and are available via OpenNeuro at https://openneuro.org/datasets/ds000212/versions/1.0.0 (ref. 115). The data of study 8 are available via figshare at https://figshare.com/articles/dataset/Visualscenes1_dataset_mjs/29424164?file=55725344 (ref. 116) and the data of study 9 are available via figshare at ataset/Visualscenes2_dataset_mjs/29424176?file=55725368 (ref. 117). The data from the ultimatum game (study 10) were provided by the authors of a previous study (ref. 118). The MJS and the thresholded statistical maps are available via figshare at https://figshare.com/articles/dataset/Brain_models_and_maps/29424206?file=55725515 (ref. 119).\u003c/p\u003e\n\u003ch2\u003eCode availability\u003c/h2\u003e\n\u003cp\u003eThe custom code that supports the findings of this study is available at https://github.com/Moral-Computing-Lab/hopp_mjs. Data were analyzed using the NLTools (v.0.4.5)\u003csup\u003e100\u003c/sup\u003e Python package available at https://github.com/cosanlab/nltools and CanlabCore Tools\u003csup\u003e120\u003c/sup\u003e available at https://github.com/canlab/CanlabCore.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eWe thank Ting Xu for sharing the ultimatum game data (study 10) with us. The acquisition of the discovery dataset was funded by a grant from the Army Research Lab to R.W. (W911NF-15-2-0115); the validation data was funded by a grant from the John Templeton Foundation to R.W. (W911NF-15-2-0115); the replication dataset was partially supported by the Duke Institute for Brain Sciences incubator grant awarded to W.S.A. and additional support from the Duke Institute for Brain Sciences; the generalization dataset received funding from the Amsterdam School of Communication Research awarded to F.R.H. (ASCoR-u-2023-Hopp; ASCoR-u-2024-Hopp). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.\u003c/p\u003e\n\u003ch2\u003eAuthor contributions\u003c/h2\u003e\n\u003cp\u003eF.R.H. and R.W. conceived of this work. F.R.H., R.W., and S.Y. undertook data curation. R.W. conducted the investigation for the discovery dataset, S.Y. and R.W. conducted the investigation for the validation dataset, W.S.A. conducted the investigation for the replication dataset, and F.R.H. conducted the investigation for the generalization dataset. F.R.H. did the formal analysis in consultation with R.W., produced the visualizations, and wrote the original paper. S.Y., W.S.A., and R.W. were responsible for validation. S.Y., W.S.A., and R.W. reviewed and edited the final paper. F.R.H. and R.W. undertook project administration and funding acquisition.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eMikhail, J. (2007). Universal moral grammar: Theory, evidence and the future. \u003cem\u003eTrends in Cognitive Sciences\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(4), 143-152.\u003c/li\u003e\n\u003cli\u003ePowell, D., \u0026amp; Horne, Z. (2017). Moral severity is represented as a domain-general magnitude. \u003cem\u003eExperimental Psychology\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eBuckholtz, J. W., \u0026amp; Marois, R. (2012). The roots of modern justice: cognitive and neural foundations of social norms and their enforcement. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(5), 655-661.\u003c/li\u003e\n\u003cli\u003eMalle, B. F. (2021). Moral judgments. \u003cem\u003eAnnual Review of Psychology\u003c/em\u003e, \u003cem\u003e72\u003c/em\u003e, 293-318.\u003c/li\u003e\n\u003cli\u003eMoll, J., Zahn, R., de Oliveira-Souza, R., Krueger, F., \u0026amp; Grafman, J. (2005). The neural basis of human moral cognition. \u003cem\u003eNature Reviews Neuroscience\u003c/em\u003e, 6(10), 799-809.\u003c/li\u003e\n\u003cli\u003eGreene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M., \u0026amp; Cohen, J. D. (2001). An fMRI investigation of emotional engagement in moral judgment. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e293\u003c/em\u003e(5537), 2105-2\u003c/li\u003e\n\u003cli\u003eCrockett, M. J., Siegel, J. Z., Kurth-Nelson, Z., Dayan, P., \u0026amp; Dolan, R. J. (2017). Moral transgressions corrupt neural representations of value. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(6), 879-885.\u003c/li\u003e\n\u003cli\u003eShenhav, A., \u0026amp; Greene, J. D. (2010). Moral judgments recruit domain-general valuation mechanisms to integrate representations of probability and magnitude. \u003cem\u003eNeuron\u003c/em\u003e, \u003cem\u003e67\u003c/em\u003e(4), 667-677.\u003c/li\u003e\n\u003cli\u003eKoster-Hale, J., Saxe, R., Dungan, J., \u0026amp; Young, L. L. (2013). Decoding moral judgments from neural representations of intentions. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e110\u003c/em\u003e(14), 5648-5653.\u003c/li\u003e\n\u003cli\u003eEres, R., Louis, W. R., \u0026amp; Molenberghs, P. (2018). Common and distinct neural networks involved in fMRI studies investigating morality: an ALE meta-analysis. \u003cem\u003eSocial Neuroscience\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(4), 384-398.\u003c/li\u003e\n\u003cli\u003eHan, H. (2017). Neural correlates of moral sensitivity and moral judgment associated with brain circuitries of selfhood: A meta-analysis. \u003cem\u003eJournal of Moral Education\u003c/em\u003e, \u003cem\u003e46\u003c/em\u003e(2), 97-113.\u003c/li\u003e\n\u003cli\u003eSevinc, G., \u0026amp; Spreng, R. N. (2014). Contextual and perceptual brain processes underlying moral cognition: a quantitative meta-analysis of moral reasoning and moral emotions. \u003cem\u003ePloS one\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(2), e87427.\u003c/li\u003e\n\u003cli\u003eHopp, F. R., Amir, O., Fisher, J., Grafton, S., Sinnott-Armstrong, W., \u0026amp; Weber, R. (2023). Moral foundations elicit shared and dissociable cortical activation modulated by political ideology. \u003cem\u003eNature Human Behavior\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eKhoudary, A., Hanna, E., O\u0026rsquo;Neill, K., Iyengar, V., Clifford, S., Cabeza, R., ... \u0026amp; Sinnott-Armstrong, W. (2022). A functional neuroimaging investigation of Moral Foundations Theory. \u003cem\u003eSocial Neuroscience\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(6), 491-507.\u003c/li\u003e\n\u003cli\u003eTsoi, L., Dungan, J. A., Chakroff, A., \u0026amp; Young, L. L. (2018). Neural substrates for moral judgments of psychological versus physical harm. \u003cem\u003eSocial Cognitive and Affective Neuroscience\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(5), 460-470.\u003c/li\u003e\n\u003cli\u003eWasserman, E. A., Chakroff, A., Saxe, R., \u0026amp; Young, L. (2017). Illuminating the conceptual structure of the space of moral violations with searchlight representational similarity analysis. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e159\u003c/em\u003e, 371-387.\u003c/li\u003e\n\u003cli\u003ePeelen, M., \u0026amp; Downing, P. (2023). Testing cognitive theories using multivariate pattern analysis of neuroimaging data. \u003cem\u003eNature Human Behavior\u003c/em\u003e. \u003c/li\u003e\n\u003cli\u003eMoll, J., Eslinger, P. J., \u0026amp; Oliveira-Souza, R. D. (2001). Frontopolar and anterior temporal cortex activation in a moral judgment task: preliminary functional MRI results in normal subjects. \u003cem\u003eArquivos de neuro-psiquiatria\u003c/em\u003e, \u003cem\u003e59\u003c/em\u003e, 657-664.\u003c/li\u003e\n\u003cli\u003eParkinson, C., Sinnott-Armstrong, W., Koralus, P. E., Mendelovici, A., McGeer, V., \u0026amp; Wheatley, T. (2011). Is morality unified? Evidence that distinct neural systems underlie moral judgments of harm, dishonesty, and disgust. \u003cem\u003eJournal of Cognitive Neuroscience\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(10), 3162-3180.\u003c/li\u003e\n\u003cli\u003eSchaich Borg, J., Hynes, C., Van Horn, J., Grafton, S., \u0026amp; Sinnott-Armstrong, W. (2006). Consequences, action, and intention as factors in moral judgments: An fMRI investigation. \u003cem\u003eJournal of Cognitive Neuroscience\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(5), 803-817.\u003c/li\u003e\n\u003cli\u003ePoldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data?. \u003cem\u003eTrends in Cognitive Sciences\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(2), 59-63.\u003c/li\u003e\n\u003cli\u003eWager, T. D., Atlas, L. Y., Lindquist, M. A., Roy, M., Woo, C. W., \u0026amp; Kross, E. (2013). An fMRI-based neurologic signature of physical pain. \u003cem\u003eNew England Journal of Medicine\u003c/em\u003e, \u003cem\u003e368\u003c/em\u003e(15), 1388-1397.\u003c/li\u003e\n\u003cli\u003eColl, M. P., Slimani, H., Woo, C. W., Wager, T. D., Rainville, P., Vachon-Presseau, \u0026Eacute;., \u0026amp; Roy, M. (2022). The neural signature of the decision value of future pain. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e119\u003c/em\u003e(23), e2119931119.\u003c/li\u003e\n\u003cli\u003eČeko, M., Kragel, P. A., Woo, C. W., L\u0026oacute;pez-Sol\u0026agrave;, M., \u0026amp; Wager, T. D. (2022). Common and stimulus-type-specific brain representations of negative affect. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e(6), 760-770.\u003c/li\u003e\n\u003cli\u003eChang, L. J., Gianaros, P. J., Manuck, S. B., Krishnan, A., \u0026amp; Wager, T. D. (2015). A sensitive and specific neural signature for picture-induced negative affect. \u003cem\u003ePLoS biology\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(6), e1002180.\u003c/li\u003e\n\u003cli\u003eKragel, P. A., Treadway, M. T., Admon, R., Pizzagalli, D. A., \u0026amp; Hahn, E. C. (2023). A mesocorticolimbic signature of pleasure in the human brain. \u003cem\u003eNature Human Behaviour\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(8), 1332-1343.\u003c/li\u003e\n\u003cli\u003eZhou, F., Zhao, W., Qi, Z., Geng, Y., Yao, S., Kendrick, K. M., ... \u0026amp; Becker, B. (2021). A distributed fMRI-based signature for the subjective experience of fear. \u003cem\u003eNature Communications\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(1), 6643.\u003c/li\u003e\n\u003cli\u003eLiu, X., Jiao, G., Zhou, F., Kendrick, K. M., Yao, D., Gong, Q., ... \u0026amp; Becker, B. (2024). A neural signature for the subjective experience of threat anticipation under uncertainty. \u003cem\u003eNature Communications\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(1), 1544.\u003c/li\u003e\n\u003cli\u003eGan, X., Zhou, F., Xu, T., Liu, X., Zhang, R., Zheng, Z., ... \u0026amp; Becker, B. (2024). A neurofunctional signature of subjective disgust generalizes to oral distaste and socio-moral contexts. \u003cem\u003eNature Human Behaviour\u003c/em\u003e, 1-20.\u003c/li\u003e\n\u003cli\u003eKragel, P. A., Koban, L., Barrett, L. F., \u0026amp; Wager, T. D. (2018). Representation, pattern\u003c/li\u003e\n\u003cli\u003einformation, and brain signatures: from neurons to neuroimaging. \u003cem\u003eNeuron\u003c/em\u003e, \u003cem\u003e99\u003c/em\u003e(2),\u003c/li\u003e\n\u003cli\u003e273.\u003c/li\u003e\n\u003cli\u003eHaidt, J. (2001). The emotional dog and its rational tail: a social intuitionist approach to moral judgment. \u003cem\u003ePsychological Review\u003c/em\u003e, \u003cem\u003e108\u003c/em\u003e(4), 814.\u003c/li\u003e\n\u003cli\u003eKragel, P. A., Kano, M., Van Oudenhove, L., Ly, H. G., Dupont, P., Rubio, A., ... \u0026amp; Wager, T. D. (2018). Generalizable representations of pain, cognitive control, and negative emotion in medial frontal cortex. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(2), 283-289.\u003c/li\u003e\n\u003cli\u003eWoo, C. W., Chang, L. J., Lindquist, M. A., \u0026amp; Wager, T. D. (2017). Building better biomarkers: brain models in translational neuroimaging. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(3), 365-377.\u003c/li\u003e\n\u003cli\u003eClifford, S., Iyengar, V., Cabeza, R., \u0026amp; Sinnott-Armstrong, W. (2015). Moral foundations vignettes: A standardized stimulus database of scenarios based on moral foundations theory. \u003cem\u003eBehavior Research Methods\u003c/em\u003e, \u003cem\u003e47\u003c/em\u003e(4), 1178\u0026ndash;1198.\u003c/li\u003e\n\u003cli\u003eHopp, F. R., Jargow, B., Kouwen, E., \u0026amp; Bakker, B. N. (2024). The Dutch moral foundations stimulus database: A validation of moral foundations vignettes and sociomoral images. \u003cem\u003eJudgment and Decision Making\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eMcIntosh, A. R., \u0026amp; Lobaugh, N. J. (2004). Partial least squares analysis of neuroimaging data: applications and advances. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e, S250-S263.\u003c/li\u003e\n\u003cli\u003eWold, S., Sj\u0026ouml;str\u0026ouml;m, M., \u0026amp; Eriksson, L. (2001). PLS-regression: a basic tool of chemometrics. \u003cem\u003eChemometrics and intelligent laboratory systems\u003c/em\u003e, \u003cem\u003e58\u003c/em\u003e(2), 109-130.\u003c/li\u003e\n\u003cli\u003eHaufe, S., Meinecke, F., G\u0026ouml;rgen, K., D\u0026auml;hne, S., Haynes, J. D., Blankertz, B., \u0026amp; Bie\u0026szlig;mann, F. (2014). On the interpretation of weight vectors of linear models in multivariate neuroimaging. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e87\u003c/em\u003e, 96-110.\u003c/li\u003e\n\u003cli\u003eDamasio, A.R., 1994. Descartes\u0026rsquo; Error: Emotion, Reason, and The Human Brain. Avon, New York\u003c/li\u003e\n\u003cli\u003eNichols, S. (2004). \u003cem\u003eSentimental rules: On the natural foundations of moral judgment\u003c/em\u003e. Oxford University Press.\u003c/li\u003e\n\u003cli\u003ePrinz, J. (2007). \u003cem\u003eThe emotional construction of morals\u003c/em\u003e. Oxford University Press.\u003c/li\u003e\n\u003cli\u003ePrinz, J. (2016). Sentimentalism and the moral brain. \u003cem\u003eMoral brains: The neuroscience of morality\u003c/em\u003e, 45-73.\u003c/li\u003e\n\u003cli\u003eCrone, D. L., Bode, S., Murawski, C., \u0026amp; Laham, S. M. (2018). The Socio-Moral Image Database (SMID): A novel stimulus set for the study of social, moral and affective processes. \u003cem\u003ePloS one\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(1), e0190954.\u003c/li\u003e\n\u003cli\u003eLancaster, J. L., Woldorff, M. G., Parsons, L. M., Liotti, M., Freitas, C. S., Rainey, L., ... \u0026amp; Fox, P. T. (2000). Automated Talairach atlas labels for functional brain mapping. \u003cem\u003eHuman brain mapping\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(3), 120-131.\u003c/li\u003e\n\u003cli\u003eAmit, E., \u0026amp; Greene, J. D. (2012). You see, the ends don\u0026rsquo;t justify the means: Visual imagery and moral judgment. \u003cem\u003ePsychological Science\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(8), 861-868.\u003c/li\u003e\n\u003cli\u003eDe Freitas, J., \u0026amp; Alvarez, G. A. (2018). Your visual system provides all the information you need to make moral judgments about generic visual events. \u003cem\u003eCognition\u003c/em\u003e, \u003cem\u003e178\u003c/em\u003e, 133-146.\u003c/li\u003e\n\u003cli\u003eSchaich Borg, J., Sinnott-Armstrong, W., Calhoun, V. D., \u0026amp; Kiehl, K. A. (2011). Neural basis of moral verdict and moral deliberation. \u003cem\u003eSocial Neuroscience\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(4), 398-413.\u003c/li\u003e\n\u003cli\u003eDecety, J., \u0026amp; Cacioppo, S. (2012). The speed of morality: a high-density electrical neuroimaging study. \u003cem\u003eJournal of Neurophysiology\u003c/em\u003e, \u003cem\u003e108\u003c/em\u003e(11), 3068-3072.\u003c/li\u003e\n\u003cli\u003eRissman, J., Gazzaley, A., \u0026amp; D\u0026apos;Esposito, M. (2004). Measuring functional connectivity during distinct stages of a cognitive task. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(2), 752-763.\u003c/li\u003e\n\u003cli\u003eYoung, L., \u0026amp; Dungan, J. (2012). Where in the brain is morality? Everywhere and maybe nowhere. \u003cem\u003eSocial Neuroscience\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(1), 1-10.\u003c/li\u003e\n\u003cli\u003eGreene, J. D. (2007). Why are VMPFC patients more utilitarian? A dual-process theory of moral judgment explains. \u003cem\u003eTrends in Cognitive Sciences\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(8), 322-323.\u003c/li\u003e\n\u003cli\u003eChiong, W., Wilson, S. M., D\u0026rsquo;Esposito, M., Kayser, A. S., Grossman, S. N., Poorzand, P., ... \u0026amp; Rankin, K. P. (2013). The salience network causally influences default mode network activity during moral reasoning. \u003cem\u003eBrain\u003c/em\u003e, \u003cem\u003e136\u003c/em\u003e(6), 1929-1941.\u003c/li\u003e\n\u003cli\u003eReniers, R. L., Corcoran, R., V\u0026ouml;llm, B. A., Mashru, A., Howard, R., \u0026amp; Liddle, P. F. (2012). Moral decision-making, ToM, empathy and the default mode network. \u003cem\u003eBiological Psychology\u003c/em\u003e, \u003cem\u003e90\u003c/em\u003e(3), 202-210.\u003c/li\u003e\n\u003cli\u003eYarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., \u0026amp; Wager, T. D. (2011). Large-scale automated synthesis of human functional neuroimaging data. \u003cem\u003eNature methods\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(8), 665-670.\u003c/li\u003e\n\u003cli\u003eKragel, P. A., Reddan, M. C., LaBar, K. S., \u0026amp; Wager, T. D. (2019). Emotion schemas are embedded in the human visual system. \u003cem\u003eScience Advances\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e(7), eaaw4358.\u003c/li\u003e\n\u003cli\u003eCarlson, R. W., Bigman, Y. E., Gray, K., Ferguson, M. J., \u0026amp; Crockett, M. J. (2022). How inferred motives shape moral judgements. \u003cem\u003eNature Reviews Psychology\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e(8), 468-478.\u003c/li\u003e\n\u003cli\u003eGrant C.M., Boucher J., Riggs K.J., Grayson A. (2005). Moral understanding in children with autism. \u003cem\u003eAutism\u003c/em\u003e 9(3):317\u0026ndash;331\u003c/li\u003e\n\u003cli\u003eMoran, J. M., Young, L. L., Saxe, R., Lee, S. M., O\u0026apos;Young, D., Mavros, P. L., \u0026amp; Gabrieli, J. D. (2011). Impaired theory of mind for moral judgment in high-functioning autism. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e108\u003c/em\u003e(7), 2688-2692.\u003c/li\u003e\n\u003cli\u003eHofmann, W., Wisneski, D. C., Brandt, M. J., \u0026amp; Skitka, L. J. (2014). Morality in everyday life. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e345\u003c/em\u003e(6202), 1340-1343.\u003c/li\u003e\n\u003cli\u003eGantman, A. P., \u0026amp; Van Bavel, J. J. (2015). Moral perception. \u003cem\u003eTrends in Cognitive Sciences\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(11), 631-633.\u003c/li\u003e\n\u003cli\u003eElmore, J. D., Lewis, J. A., Himmelberger, Z. M., \u0026amp; Sherwood, J. A. (2024). Everyday moral transgressions (EMTs): Investigating the morality of everyday behaviors. \u003cem\u003eCurrent Psychology\u003c/em\u003e, \u003cem\u003e43\u003c/em\u003e(11), 10484-10493.\u003c/li\u003e\n\u003cli\u003eYudkin, D. A., Goodwin, G. P., Reece, A., Gray, K., \u0026amp; Bhatia, S. (2025). A large-scale investigation of everyday moral dilemmas. \u003cem\u003ePNAS Nexus\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e(5), pgaf119.\u003c/li\u003e\n\u003cli\u003eDolan, R. J. (1999). On the neurology of morals. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(11), 927-929.\u003c/li\u003e\n\u003cli\u003eGreene, J., \u0026amp; Haidt, J. (2002). How (and where) does moral judgment work?. \u003cem\u003eTrends in Cognitive Sciences\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(12), 517-523.\u003c/li\u003e\n\u003cli\u003eGreene, J. D. (2009). The cognitive neuroscience of moral judgment. \u003cem\u003eThe cognitive neurosciences\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e, 1-48.\u003c/li\u003e\n\u003cli\u003eKrebs, D. L. (2008). Morality: An evolutionary account. \u003cem\u003ePerspectives on Psychological Science\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e(3), 149-172.\u003c/li\u003e\n\u003cli\u003eMoll, J., de Oliveira-Souza, R., \u0026amp; Eslinger, P. J. (2003). Morals and the human brain: a working model. \u003cem\u003eNeuroreport\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(3), 299-305.\u003c/li\u003e\n\u003cli\u003eMoore, A. B., Stevens, J., \u0026amp; Conway, A. R. (2011). Individual differences in sensitivity to reward and punishment predict moral judgment. \u003cem\u003ePersonality and Individual Differences\u003c/em\u003e, \u003cem\u003e50\u003c/em\u003e(5), 621-625.\u003c/li\u003e\n\u003cli\u003evan Baar, J. M., Chang, L. J., \u0026amp; Sanfey, A. G. (2019). The computational and neural substrates of moral strategies in social decision-making. \u003cem\u003eNature Communications\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(1), 1483.\u003c/li\u003e\n\u003cli\u003eAtari, M., Haidt, J., Graham, J., Koleva, S., Stevens, S. T., \u0026amp; Dehghani, M. (2023). Morality beyond the WEIRD: How the nomological network of morality varies across cultures. \u003cem\u003eJournal of Personality and Social Psychology\u003c/em\u003e, \u003cem\u003e125\u003c/em\u003e(5), 1157.\u003c/li\u003e\n\u003cli\u003eAwad, E., Dsouza, S., Kim, R., Schulz, J., Henrich, J., Shariff, A., ... \u0026amp; Rahwan, I. (2018). The moral machine experiment. \u003cem\u003eNature\u003c/em\u003e, \u003cem\u003e563\u003c/em\u003e(7729), 59-64.\u003c/li\u003e\n\u003cli\u003eHorberg, E. J., Oveis, C., \u0026amp; Keltner, D. (2011). Emotions as moral amplifiers: An appraisal tendency approach to the influences of distinct emotions upon moral judgment. \u003cem\u003eEmotion Review\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e(3), 237-244.\u003c/li\u003e\n\u003cli\u003eKiar, G., Mumford, J. A., Xu, T., Vogelstein, J. T., Glatard, T., \u0026amp; Milham, M. P. (2024). Why experimental variation in neuroimaging should be embraced. \u003cem\u003eNature Communications\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(1), 9411.\u003c/li\u003e\n\u003cli\u003eGray, K., Waytz, A., \u0026amp; Young, L. (2012). The moral dyad: A fundamental template unifying moral judgment. Psychological Inquiry, 23(2), 206\u0026ndash;215. http://doi.org/10.1080/10478 40X.2012.686247\u003c/li\u003e\n\u003cli\u003eGray, K.,Young, L., \u0026amp; Waytz, A. (2012). Mind perception is the essence of morality. \u003cem\u003ePsychological Inquiry\u003c/em\u003e, 23(2), 101\u0026ndash;124. http://doi.org/10.1080/1047840X.2012.651387\u003c/li\u003e\n\u003cli\u003eCrockett, M. J. (2016). Computational modeling of moral decisions. In \u003cem\u003eThe social psychology of morality\u003c/em\u003e (pp. 71-90). Routledge.\u003c/li\u003e\n\u003cli\u003eZoh, Y., Chang, S. W., \u0026amp; Crockett, M. J. (2022). The prefrontal cortex and (uniquely) human cooperation: a comparative perspective. \u003cem\u003eNeuropsychopharmacology\u003c/em\u003e, \u003cem\u003e47\u003c/em\u003e(1), 119-133.\u003c/li\u003e\n\u003cli\u003eKrueger, F., \u0026amp; Hoffman, M. (2016). The emerging neuroscience of third-party punishment. \u003cem\u003eTrends in Neurosciences\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(8), 499-501.\u003c/li\u003e\n\u003cli\u003eSeymour, B., Singer, T., \u0026amp; Dolan, R. (2007). The neurobiology of punishment. \u003cem\u003eNature Reviews Neuroscience\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(4), 300-311.\u003c/li\u003e\n\u003cli\u003eHaidt, J., \u0026amp; Joseph, C. (2004). Intuitive ethics: How innately prepared intuitions generate culturally variable virtues. \u003cem\u003eDaedalus\u003c/em\u003e, \u003cem\u003e133\u003c/em\u003e(4), 55-66.\u003c/li\u003e\n\u003cli\u003eSinnott-Armstrong, W., \u0026amp; Wheatley, T. (2012). The disunity of morality and why it matters to philosophy. \u003cem\u003eThe Monist\u003c/em\u003e, \u003cem\u003e95\u003c/em\u003e(3), 355-377.\u003c/li\u003e\n\u003cli\u003eCameron, C. D., Lindquist, K. A., \u0026amp; Gray, K. (2015). A constructionist review of morality and emotions: No evidence for specific links between moral content and discrete emotions. \u003cem\u003ePersonality and Social Psychology Review\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(4), 371-394.\u003c/li\u003e\n\u003cli\u003eAshar, Y. K., Andrews-Hanna, J. R., Dimidjian, S., \u0026amp; Wager, T. D. (2017). Empathic care and distress: predictive brain markers and dissociable brain systems. \u003cem\u003eNeuron\u003c/em\u003e, \u003cem\u003e94\u003c/em\u003e(6), 1263-1273.\u003c/li\u003e\n\u003cli\u003eWeber, R., Hopp, F. R., Eden, A., Fisher, J. T., \u0026amp; Lee, H. E. (2024). Vicarious punishment of moral violations in naturalistic drama narratives predicts cortical synchronization. \u003cem\u003eNeuroImage\u003c/em\u003e, \u003cem\u003e292\u003c/em\u003e, 120613.\u003c/li\u003e\n\u003cli\u003eKaplan, J. T., Gimbel, S. I., Dehghani, M., Immordino-Yang, M. H., Sagae, K., Wong, J. D., ... \u0026amp; Damasio, A. (2017). Processing narratives concerning protected values: A cross-cultural investigation of neural correlates. \u003cem\u003eCerebral Cortex\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(2), 1428-1438.\u003c/li\u003e\n\u003cli\u003eGazzaniga, M. S. (2005). \u003cem\u003eThe ethical brain\u003c/em\u003e. Dana press.\u003c/li\u003e\n\u003cli\u003eIlles, J., \u0026amp; Bird, S. J. (2006). Neuroethics: a modern context for ethics in neuroscience. \u003cem\u003eTrends in Neurosciences\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(9), 511-517.\u003c/li\u003e\n\u003cli\u003eRoskies, A. (2002). Neuroethics for the new millenium. \u003cem\u003eNeuron\u003c/em\u003e, \u003cem\u003e35\u003c/em\u003e(1), 21-23.\u003c/li\u003e\n\u003cli\u003eHester, N., \u0026amp; Gray, K. (2020). The moral psychology of raceless, genderless strangers. \u003cem\u003ePerspectives on Psychological Science\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(2), 216-230.\u003c/li\u003e\n\u003cli\u003eKohoutov\u0026aacute;, L., Atlas, L. Y., B\u0026uuml;chel, C., Buhle, J. T., Geuter, S., Jepma, M., ... \u0026amp; Woo, C. W. (2022). Individual variability in brain representations of pain. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e(6), 749-759.\u003c/li\u003e\n\u003cli\u003eLewis, G. J., Kanai, R., Bates, T. C., \u0026amp; Rees, G. (2012). Moral values are associated with individual differences in regional brain volume. \u003cem\u003eJournal of Cognitive Neuroscience\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(8), 1657-1663.\u003c/li\u003e\n\u003cli\u003eAnderson, S. W., \u0026amp; Bechara, A. \u0026amp; Damasio, A., R.(1999) Impairment of social and moral behavior related to early damage in human prefrontal cortex. \u003cem\u003eNature Neuroscience\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(11), 1032-1037.\u003c/li\u003e\n\u003cli\u003eLeong, Y. C., Chen, J., Willer, R., \u0026amp; Zaki, J. (2020). Conservative and liberal attitudes drive polarized neural responses to political content. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e117\u003c/em\u003e(44), 27731-27739.\u003c/li\u003e\n\u003cli\u003eMcCurrie, C. H., Crone, D. L., Bigelow, F., \u0026amp; Laham, S. M. (2018). Moral and Affective Film Set (MAAFS): A normed moral video database. \u003cem\u003ePloS one\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(11), e0206604.\u003c/li\u003e\n\u003cli\u003eKoenigs, M., Young, L., Adolphs, R., Tranel, D., Cushman, F., Hauser, M., \u0026amp; Damasio, A. (2007). Damage to the prefrontal cortex increases utilitarian moral judgements. \u003cem\u003eNature\u003c/em\u003e, \u003cem\u003e446\u003c/em\u003e(7138), 908-911.\u003c/li\u003e\n\u003cli\u003eYoung, L., Bechara, A., Tranel, D., Damasio, H., Hauser, M., \u0026amp; Damasio, A. (2010). Damage to ventromedial prefrontal cortex impairs judgment of harmful intent. \u003cem\u003eNeuron\u003c/em\u003e, \u003cem\u003e65\u003c/em\u003e(6), 845-851.\u003c/li\u003e\n\u003cli\u003eCurry, O. S., Chesters, M. J., \u0026amp; Van Lissa, C. J. (2019). Mapping morality with a compass: Testing the theory of \u0026lsquo;morality-as-cooperation\u0026rsquo;with a new questionnaire. \u003cem\u003eJournal of Research in Personality\u003c/em\u003e, \u003cem\u003e78\u003c/em\u003e, 106-124.\u003c/li\u003e\n\u003cli\u003eEsteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, A. I., Erramuzpe, A., ... \u0026amp; Gorgolewski, K. J. (2019). fMRIPrep: a robust preprocessing pipeline for functional MRI. \u003cem\u003eNature Methods\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(1), 111-116.\u003c/li\u003e\n\u003cli\u003ePeirce, J. W. (2007). PsychoPy\u0026mdash;psychophysics software in Python. \u003cem\u003eJournal of neuroscience methods\u003c/em\u003e, \u003cem\u003e162\u003c/em\u003e(1-2), 8-13.\u003c/li\u003e\n\u003cli\u003eChang, L., Jolly, E., Cheong, J. H., Burnashev, A. \u0026amp; Chen, A. cosanlab/nltools: 0.3.11 Zenodo https://zenodo.org/ record/2229813(2018).\u003c/li\u003e\n\u003cli\u003ePedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... \u0026amp; Duchesnay, \u0026Eacute;. (2011). Scikit-learn: Machine learning in Python. \u003cem\u003ethe Journal of machine Learning research\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, 2825-2830.\u003c/li\u003e\n\u003cli\u003eFan, L., Li, H., Zhuo, J., Zhang, Y., Wang, J., Chen, L., ... \u0026amp; Jiang, T. (2016). The human brainnetome atlas: a new brain atlas based on connectional architecture. \u003cem\u003eCerebral Cortex\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(8), 3508-3526.\u003c/li\u003e\n\u003cli\u003eYeo, B. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., ... \u0026amp; Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. \u003cem\u003eJournal of Neurophysiology\u003c/em\u003e, \u003cem\u003e106\u003c/em\u003e(3), 1125-1165.\u003c/li\u003e\n\u003cli\u003eLeDoux, J. E., \u0026amp; Pine, D. S. (2016). Using neuroscience to help understand fear and anxiety: a two-system framework. \u003cem\u003eAmerican journal of psychiatry\u003c/em\u003e, \u003cem\u003e173\u003c/em\u003e(11), 1083-1093.\u003c/li\u003e\n\u003cli\u003eMediationToolbox. GitHub https://github.com/canlab/ MediationToolbox (2014). \u003c/li\u003e\n\u003cli\u003eWoo, C. W., Schmidt, L., Krishnan, A., Jepma, M., Roy, M., Lindquist, M. A., ... \u0026amp; Wager, T. D. (2017). Quantifying cerebral contributions to pain beyond nociception. \u003cem\u003eNature Communications\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(1), 14211.\u003c/li\u003e\n\u003cli\u003eChapman, H. A., Kim, D. A., Susskind, J. M., \u0026amp; Anderson, A. K. (2009). In bad taste: Evidence for the oral origins of moral disgust. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e323\u003c/em\u003e(5918), 1222-1226.\u003c/li\u003e\n\u003cli\u003eSanfey, A. G., Rilling, J. K., Aronson, J. A., Nystrom, L. E., \u0026amp; Cohen, J. D. (2003). The neural basis of economic decision-making in the ultimatum game. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e300\u003c/em\u003e(5626), 1755-1758.\u003c/li\u003e\n\u003cli\u003eMoral judgment discovery cohort dataset. figshare https://figshare.com/articles/dataset/Discovery_dataset_mjs/29423726?file=55720082 \u003c/li\u003e\n\u003cli\u003eMoral judgment validation cohort dataset. figshare https://figshare.com/articles/dataset/Validation_dataset_mjs/29423789?file=55721972 \u003c/li\u003e\n\u003cli\u003eMoral judgment replication cohort dataset. figshare https://figshare.com/articles/dataset/Replication_dataset_mjs/29423966?file=55724255 \u003c/li\u003e\n\u003cli\u003eMoral judgment generalization cohort dataset. figshare https://figshare.com/articles/dataset/Generalization_dataset_mjs/29423981?file=55724291 \u003c/li\u003e\n\u003cli\u003ePINES holdout test dataset. NeuroVault https://neurovault.org/collections/1964 (2015). \u003c/li\u003e\n\u003cli\u003eDisgust validation cohort dataset. figshare https://figshare.com/ articles/dataset/validation_dataset_disgust/22841117 (2023).\u003c/li\u003e\n\u003cli\u003eYoung, L., Chakroff, A., Wasserman, E., Saxe, R., Dungan, J., Brown, A., \u0026amp; Koster-Hale, J. (2019). Moral judgments of intentional and accidental moral violations across Harm and Purity domains. \u003cem\u003eOpenNeuro.[Dataset]\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e. \u003c/li\u003e\n\u003cli\u003eMoral judgment visual scenes 1 dataset figshare https://figshare.com/articles/dataset/Visualscenes1_dataset_mjs/29424164?file=55725344 \u003c/li\u003e\n\u003cli\u003eMoral judgment visual scenes 2 dataset figshare https://figshare.com/articles/dataset/Visualscenes2_dataset_mjs/29424176?file=55725368 \u003c/li\u003e\n\u003cli\u003eXu, T., Zhang, L., Zhou, F., Fu, K., Gan, X., Chen, Z., ... \u0026amp; Becker, B. (2025). Distinct neural computations scale the violation of expected reward and emotion in social transgressions. \u003cem\u003eCommunications Biology\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(1), 106.\u003c/li\u003e\n\u003cli\u003eThe MJS and thresholded statistical maps. figshare https://figshare.com/articles/dataset/Brain_models_and_maps/29424206?file=55725515 \u003c/li\u003e\n\u003cli\u003eCognitive and Affective Neuroscience Laboratory, CANLab. GitHub https://github.com/canlab (2014).\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e Recent theoretical advancements separate fairness into equality and proportionality\u003csup\u003e\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e\u003c/sup\u003e, whereas we examined fairness in its traditional, holistic form.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7935407/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7935407/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHumans universally condemn what they see as moral violations, yet the perceived wrongness of distinct moral transgressions varies across individuals, neurodiverse populations, and cultures. We currently do not know how and to what extent the human brain universally computes and translates moral wrongness into subjective, graded moral judgments. Here, we combined fMRI with pattern recognition techniques to identify and evaluate a neural signature predictive of graded moral wrongness judgments. Drawing on to date’s largest database for studying the neural basis of moral judgment, spanning independent, multi-culture fMRI datasets of moral vignettes for discovery (n = 64), validation (n = 30), replication (n = 27), and generalization (n = 30) analyses (n\u003csub\u003etotal\u003c/sub\u003e=151), we demonstrate that accurate prediction of graded moral wrongness relies on a distributed neural circuit, with important contributions from cortical and subcortical areas. We further evaluate common and domain-isolated (e.g., care, fairness, purity) brain systems for graded moral wrongness and demonstrate shared and dissociable neural representations with negative affect and subjective disgust. Together, we find that graded moral wrongness judgments are robustly computed via a shared and distributed neural code and provide a sensitive and specific moral wrongness biomarker for future studies.\u003c/p\u003e","manuscriptTitle":"A Sensitive and Specific Neural Signature Robustly Predicts Graded Computations of Moral Wrongness","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-25 06:15:33","doi":"10.21203/rs.3.rs-7935407/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fefb227e-79a7-47a2-a8d0-66e05dae7f48","owner":[],"postedDate":"November 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":58394779,"name":"Biological sciences/Psychology/Human behaviour"},{"id":58394780,"name":"Biological sciences/Psychology"},{"id":58394781,"name":"Biological sciences/Neuroscience/Cognitive neuroscience"}],"tags":[],"updatedAt":"2026-03-27T16:50:50+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-25 06:15:33","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7935407","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7935407","identity":"rs-7935407","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}