Neural correlates of emotional responses to self-selected music: evidence from multivariate pattern analysis

for a complete description of the fMRIPrep methods. The following sections describe the imaging data analyses performed using custom Python scripts based on the package Nilearn v0.11.1 (Abraham et al., 2014; Nilearn contributors et al., 2025). .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 2.5 fMRI Data Analysis 7 2.5.1 GLM Analysis First-level analyses were performed using a single General Linear Model (GLM) estimated separately for each participant. The design matrix included one predictor for each of the nine emotion categories, based on the participant’s self-labeled musical stimuli, as well as nuisance regressors to account for head motion (six motion parameters, their first-order derivatives, and squared terms, totaling 24 confound regressors). Prior to model estimation, we applied a high-pass temporal filter with a cut-off frequency of 0.007 Hz and spatial smoothing using a Gaussian kernel with a full-width at half-maximum (FWHM) of 4 mm. A second-order autoregressive model (AR(2)) was used to account for temporal autocorrelations. For the second-level analysis, we computed a one-way ANOVA with ’emotion’ as the within-subjects factor. This analysis used the first-level contrast maps for each emotion (emotion > white noise) to identify brain regions where the magnitude of responses varied as a function of emotion. We restricted our analysis to grey-matter voxels, defined using the MNI152NLin2009cAsym segmentation probability map with a threshold of 0.15. To correct for multiple comparisons, we applied Bonferroni correction at the voxel level (p < 0.05). Clusters were defined using a minimum cluster size of 25 contiguous voxels. Anatomical labels for significant clusters were assigned based on the AAL3 maps (Rolls et al., 2020) and cross-validated with functional annotations from the Neurosynth database (Yarkoni et al., 2011). 2.5.2 Multivoxel Pattern Analysis We performed feature selection to focus the analysis on a subset of voxels most likely to provide discrim- inative information. Specifically, we used a voxel-wise stability metric as described by (Just et al., 2010). This approach identifies voxels whose activation profiles across the nine emotions were consistent over repeated presentations of the musical excerpts, i.e., indicating stable relative variation between emotional responses. The rationale is that stable voxels are more likely to reflect meaningful, stimulus-driven neural responses. The voxel stability was estimated as the average pairwise Pearson correlation between its nine-emotion activation profiles, a vector of nine responses of a voxel to the emotions during a single presentation, across repeated presentations of the nine emotion categories. Voxels with an average sta- bility correlation of at least 0.1 were retained, resulting in individual stability masks for each participant. This threshold, while defining a large feature input space, was chosen as a compromise to provide a clear criterion for stability while ensuring a comprehensive sampling of the whole brain and imposing minimal restrictions. For each 24-second music excerpt, we extracted denoised BOLD signals after regressing out nuisance variables related to head motion (six motion parameters, their first-order derivatives, and squared terms, .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 8 2 METHODS totaling 24 confound regressors), applying a high-pass temporal filter with a cut-off frequency of 0.007 Hz, detrending, and normalizing the time series to unit variance. Each excerpt was divided into four consecutive, non-overlapping 6-second segments. The BOLD signal within each segment was averaged, resulting in four time-resolved feature values per voxel per block. Feature extraction was restricted to gray matter voxels, defined using the MNI152NLin2009cAsym gray matter probability map thresholded at 0.15. Multivariate pattern analysis was conducted using a supervised learning method as the estimator model - a Support Vector Machine (SVM) with a linear kernel and an L1 regularization factor. The multiclass problem was addressed using a one vs. all method. We implemented cross-validation to split the data into different sets. We then fitted the estimator on the training dataset and measured an unbiased error on the testing set. To avoid temporal dependencies between training and testing samples (bias due to temporal proximity of samples, particularly relevant in functional data), we implemented a leave-one- run-out (LORO) strategy, where the classifier was trained on three of the four runs and tested on the remaining one. Leaving out blocks of correlated observations, rather than individual observations, is crucial for non-biased estimates (Varoquaux et al., 2017). .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 9 3 Results Our results are presented in two parts. We first mapped the brain responses to music across emotions (univariate analysis), and then assessed the discriminability of music-evoked emotions using MVPA and explored the spatial distribution of the most discriminative voxels across the brain. 3.1 Brain network of music-evoked emotions Figure 3 displays brain regions where activity varied significantly as a function of emotion (i.e., main effect of emotion), based on whole-brain analysis. Activations were largely bilateral, including highly significant clusters in the superior temporal gyrus, Heschl’s gyrus, the cerebellum, supplementary motor area, putamen, insula, frontal cortex, and nucleus accumbens (Table 1). Figure 3: Brain regions in which BOLD responses varied as a function of music-evoked emotions. The map shows z-values corrected for multiple comparisons with Bonferroni’s method (p = 0.05) and a minimum cluster size of 25 voxels. 3.2 Decoding music-evoked emotions To test whether the nine aesthetic emotions could be reliably discriminated based on neural activation patterns, we applied an MVPA pipeline to whole-brain fMRI data. Feature selection was performed using a stability-based mask, ensuring that only voxels with consistent emotion-related activation profiles contributed to the analysis. The linear SVM classifier was trained and tested using a leave-one-run-out cross-validation scheme. Classification accuracy served as the performance metric, with a chance level .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 10 3 RESULTS Table 1: Cluster table of the activation map of Figure 3. MNI Coordinates of the peak voxel, the z-value of the peak voxel, the cluster size, and the label from the AAL3 atlas. Cluster label X Y Z Z-Value of Peak Voxel Cluster Size (mm 3) Temporal Sup R 54 2 -5 18.45 22224 Temporal Sup L -53 -15 6 17.82 22352 Cerebellum 6 R 30 -61 -25 14.41 8104 Cerebellum 6 L -33 -63 -23 13.27 2728 Precentral R 56 -3 46 12.05 2160 Supp Motor Area L -1 2 68 11.49 3752 Postcentral L -55 -7 48 11.37 1864 Parietal Inf R 60 -53 40 10.71 10408 Cingulate Mid R 4 -33 48 10.69 22528 Frontal Sup 2 R 28 34 50 10.37 15008 Cerebellum 8 L -25 -65 -53 9.76 1568 ACC pre R 4 46 12 9.21 6880 Supp Motor Area L -9 14 70 9.15 928 Temporal Mid R 68 -23 -13 8.80 1960 Cerebellum Crus1 L -49 -65 -27 8.55 344 Cerebellum 8 R 18 -69 -53 8.54 2712 Cerebellum Crus2 L -49 -61 -45 8.51 1712 Cerebellum 7b L -15 -79 -45 8.21 552 Insula R 34 18 -11 8.16 440 Occipital Sup L -19 -67 26 8.08 1152 Temporal Inf R 48 10 -41 8.02 1240 Cerebellum 6 L -11 -67 -13 7.89 824 Putamen R 24 4 6 7.84 544 Temporal Inf R 58 -7 -35 7.72 1072 Angular R 46 -77 40 7.65 3456 Temporal Inf L -53 -9 -31 7.65 992 Insula R 40 -11 -1 7.45 216 Frontal Inf Tri R 54 36 4 7.24 632 N Acc R 14 16 -11 7.18 272 Supp Motor Area R 14 20 62 7.02 400 Broca R 30 30 4 6.85 216 Occipital Mid L -37 -83 38 6.77 296 Postcentral R 54 -7 26 6.77 408 Frontal Mid 2 L -27 18 60 6.75 208 Frontal Mid 2 L -47 48 -5 6.51 240 N Acc L -11 18 -9 6.23 224 Frontal Sup 2 R 26 64 -3 6.08 288 Temporal Pole Mid L -53 10 -29 5.91 256 Angular L -61 -59 22 5.81 216 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 3.2 Decoding music-evoked emotions 11 of 11.1% given the nine emotion classes. This analysis assessed whether distributed patterns of brain activity contained sufficient information to predict the specific emotion class evoked by music. 3.2.1 Voxel selection In Figure 4, we display the sum of the stability masks estimated for each subject. Clusters were found in the temporal superior gyrus, precentral and postcentral gyrus, supramarginal gyrus, and the SMA. The average number of voxels in the stability mask across subjects was 8205.6 ± 2129.6 voxels. Figure 4: Sum of the binary stability masks for the 20 subjects. For this map, we set the threshold to a minimum of 2 participants, with cluster correction of k>5 voxels and spatial smoothing (FWHM = 4 mm). 3.2.2 Classification performance Overall, our model achieved 48.5%±6.2% accuracy across all emotions. In Figure 5, we display the confusion matrix with the average percentages attributed to each class, where a clear discrimination between the correct classification (diagonal) is visible, with no major heterogeneity in the incorrect classifications. 3.2.3 Discriminative brain regions From the trained models, we retrieved the classifier’s weights for each voxel and emotion class for each participant. These weights indicate the relative contribution of each voxel to distinguishing a given emotion from the others. To obtain a group-level summary of voxel importance, we summed the absolute .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 12 4 DISCUSSION Figure 5: Group-level confusion matrix showing the results of the testing phase across the 20 participants. The overall balanced accuracy was 48.5± 6.2%, while the chance level sits at 11.1% in this case. weights across all emotions and participants, achieving a predictive power map, displayed in Figure 6. The clusters identified in this map are listed in Table 2. 4 Discussion This study investigated the neural representation of music-evoked emotions using a personalized fMRI paradigm and multivariate decoding. Participants selected familiar musical excerpts evoking the nine emotion factors of the Geneva Emotional Music Scale, allowing us to maximize emotional salience and ecological validity. Univariate analyses revealed emotion-specific neural activity across a broad bilateral network. Multivariate analyses using MVPA then showed that these emotions could be reliably decoded from whole-brain activity using a linear SVM classifier, with performance significantly above chance. Brain regions contributing to classification included auditory, motor, and interoceptive areas, consistent with prior studies linking these systems to music-induced affective processing (Koelsch, 2020; Vuust et al., 2022). The successful decoding of emotional categories reinforces findings from previous MVPA studies that .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 13 Figure 6: Predictive power brain map for all conditions and subjects. The map shows the sum of the absolute weights of the linear SVM classifier across all participants and emotions. The color bar indicates the relative contribution of each voxel to emotion classification. mostly explored different emotions across the valence dimension (e.g., joy vs. sadness) (Koelsch et al., 2021; Putkinen et al., 2021) and supports the view that aesthetic emotions are represented in distributed networks. Here, we explore the full spectrum of the GEMS, which we considered an important advance- ment when studying the correlates of felt emotions. The use of self-selected, familiar music departs from more controlled but less engaging experimental approaches and shifts the focus from perceived to felt emotions. Familiarity has been shown to enhance emotional engagement and activity in reward-related areas (Freitas et al., 2018; C. S. Pereira et al., 2011), and our paradigm leveraged this to enhance the ecological validity and intensity of the emotional experience. Our univariate analysis replicated earlier work (Koelsch, 2020; Trost et al., 2012), showing broad ac- tivation in sensory and limbic regions during music listening compared to noise. These results suggest that participant-specific emotional experiences, despite being highly individualized, are underpinned by consistent neural mechanisms. Importantly, we also explored emotion-specific activation patterns, but found that most of the core clusters, especially in auditory, motor, and reward-related regions, responded similarly across all emotional categories (Supplementary Figure S4). This lack of fine-grained selectivity supports the idea that these areas act not as “emotion-specific modules” but as part of a general affective interpretation network, consistent with the theory of constructed emotion (Barrett, 2016, 2017), which posits that emotions emerge from domain-general systems involved in conceptualization and interocep- tion. Despite the lack of strong univariate selectivity, we were still able to successfully discriminate emotion categories using multivoxel pattern analysis. This reinforces the view that emotional states are encoded not through isolated regional activations but through distributed spatial patterns across multiple brain .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 14 4 DISCUSSION Table 2: Cluster table of the predictive power map of Figure 6, showing the MNI coordinates of the peaks, cluster size, and AAL3 labeling. Cluster Label X Y Z Cluster Size (mm3) Temporal Sup R 58 -11 4 24456 Temporal Sup L -57 -17 4 38096 Precentral R 56 -5 44 4312 Supp Motor Area L -11 -7 66 6160 Frontal Sup 2 L -19 64 18 664 Frontal Mid 2 R 40 60 -9 1256 Parietal Sup L -19 -67 48 1648 Frontal Inf Orb 2 L -51 28 -9 416 Frontal Inf Tri L -49 14 28 744 Angular R 34 -67 48 1176 Precuneus R 12 -65 34 400 Parietal Inf R 60 -43 46 320 Temporal Inf R 64 -29 -19 304 Cerebellum Crus1 R 44 -65 -33 272 Temporal Mid L -65 -57 -7 568 Amygdala L -25 -5 -19 312 Temporal Mid R 66 -53 10 320 Cingulate Mid R 6 -23 44 224 Frontal Inf Tri R 56 34 4 352 Frontal Mid 2 R 38 46 34 408 Frontal Inf Tri L -57 20 14 224 Temporal Inf R 58 -61 -5 384 Frontal Med Orb L -9 50 -7 192 Frontal Mid 2 L -41 56 -13 184 Cingulate Mid R 4 -37 46 272 Precuneus R 8 -77 56 216 Frontal Sup 2 R 26 62 18 168 systems. This observation also aligns with emerging evidence from connectivity-based studies, which em- phasize the importance of network-level integration in affective processing (Kober et al., 2008; Lindquist et al., 2016; Pessoa, 2017). Emotion regulation and differentiation appear to depend not only on activity within specific nodes but on the functional coupling between regions involved in perception, interocep- tion, memory, and conceptualization. Gaining finer control over this network-level architecture may be essential for both mechanistic understanding and music-based emotion regulation interventions. Our task design demonstrates that it is possible to obtain reliable univariate and multivariate results when using individualized, self-selected music and focusing specifically on felt emotional experiences. This personalized approach not only enhances the ecological validity of studies on musical emotions, a .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 4.1 Limitations and Future Directions 15 recognized need in music research (Tervaniemi, 2023), but also holds promise for applied fields such as music-based rehabilitation and brain-computer interfaces. These interventions could be optimized for effectiveness by tailoring musical stimuli to individual preferences and emotional responses. 4.1 Limitations and Future Directions While our personalized paradigm enhances emotional engagement and ecological validity, it also intro- duces methodological challenges. Among these is the large between-subject heterogeneity in the selected musical stimuli. Participants’ choices varied widely in terms of low-level acoustic features (e.g., tempo, timbre, loudness) and genres. This variability complicates interpretation: although the classifier reliably distinguished between emotion categories, it remains unclear to what extent decoding reflects affective processes as opposed to differences in stimulus properties. It is important to note, however, that the au- ditory cortex does not merely process sensory/acoustic information, but is functionally embedded within broader emotion-related networks. As highlighted by (Koelsch et al., 2021), the auditory cortex has direct anatomical and functional connections to limbic and paralimbic structures such as the amygdala, insula, cingulate cortex, and ventral striatum, and plays a central role in generating feeling representa- tions from sound. Furthermore, its functional connectivity with reward-related regions has been shown to predict the subjective emotional value of music and to vary with individual traits such as musical anhedonia (Mart ´ ınez-Molina et al., 2016). Thus, even though our paradigm did not fully control for acoustic features across emotion categories, it is likely that decoding success in auditory areas reflects, at least in part, their involvement in affective appraisal and emotional experience, rather than simply encoding acoustic differences. The self-selection of stimuli also limits our ability to pinpoint which mechanisms of emotion induction were engaged. According to the BRECVMA model (Juslin, 2013), music can evoke emotions through multiple mechanisms, including brainstem reflexes, rhythmic entrainment, evaluative conditioning, mu- sical expectancy, autobiographical memory, and others. Our personalized paradigm likely triggered a combination of these pathways, particularly those linked to personal associations and memories, yet their individual contributions could not be isolated. Disentangling these pathways will require paradigms specifically designed to manipulate each mechanism independently, possibly through controlled stimulus design or complementary behavioral measures. Finally, the use of a linear support vector machine as our decoding model represents a methodologi- cal trade-off. While linear classifiers offer strong interpretability and are robust to overfitting in high- dimensional fMRI data, they may not capture more complex, non-linear relationships between neural .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 16 4 DISCUSSION activity and emotional states (Hebart & Baker, 2018; F. Pereira et al., 2009). That said, if the fea- ture space is well-structured and informative, as was ensured here through stability-based selection, then classifier complexity may be less critical. Future work could explore the use of non-linear models or deep learning approaches, especially in larger datasets, to determine whether more flexible classifiers can uncover additional structure in the representation of music-evoked emotions. Looking forward, combining personalized paradigms with computational models of acoustic and emotional features may help reconcile ecological validity with experimental control. Additionally, incorporating real- time decoding techniques could pave the way for music-based neurofeedback or affective brain-computer interfaces, with promising applications in clinical contexts where improving emotion regulation is a key therapeutic target. Data and Code Availability All scripts used for the analyses mentioned in this work can be found in this git repository: https: //github.com/CIBIT-UC/brainplayback task02. The anonymized and defaced dataset can be found in BIDS format at https://doi.org/10.57979/P4SAYV. The data management plan regarding this project can be found at Zenodo https://doi.org/10.5281/zenodo.10563831. Author Contributions Conceptualization - AS, IB, BD; Data curation - AS, BD, Formal Analysis - AS, BD, Funding acquisition - MCB, IB, BD; Investigation - AS, BD, Methodology - AS, CL, BD; Project administration - IB, BD; Resources - BD, MCB; Software - AS, BD; Supervision - TS, MCB, BD, Visualization - AS, BD, Writing - original draft - AS, BD; Writing - review & editing - AS, CL, IB, TS, MCB, BD Funding This work was supported by the Portuguese Foundation for Science and Technology (FCT) (EXPL/PSI- GER/0948/2021, UIDB/04950/2020/2025, UIDP/04950/2020/2025, LA/P/0104/2020, UIDB/00326/2020, 2023.04365.BD, CEECINST/00117/2021/CP2784/CT0002, 2022.04701.PTDC, 2021.01469.CEECIND, CEECINSTLA/00026/2022/CP2919/CT0001). AS holds a PhD grant from Siemens Healthineers Por- .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint

17 tugal. Computational support was provided by the Portuguese National Distributed Computing Infras- tructure (INCD). Declaration of Competing Interests The authors declare no competing interests.

We thank all the participants who took part in this study. We also thank the remaining team members of the Brainplayback project - Andr´ e Granjo, Carolina Travassos, Jo˜ ao Pereira, Renato Panda, Rui Pedro Paiva, Daniela Pereira, and Inˆ es Almeida - for the insightful contributions. A last word of gratitude to the team of the MRI Unit of ICNAS - S´ onia Afonso, Tˆ ania Lopes - for their support and assistance during data acquisition.

Abraham, A., Pedregosa, F., Eickenberg, M., Gervais, P., Mueller, A., Kossaifi, J., Gramfort, A., Thirion, B., & Varoquaux, G. (2014). Machine learning for neuroimaging with scikit-learn.Frontiers in Neuroinformatics,8. https://doi.org/10.3389/fninf.2014.00014 Barrett, L. F. (2016). The theory of constructed emotion: An active inference account of interoception and categorization.Social Cognitive and Affective Neuroscience, nsw154. https://doi.org/10. 1093/scan/nsw154 Barrett, L. F. (2017). How emotions are made: The secret life of the brain. Mariner Books. Blood, A. J., & Zatorre, R. J. (2001). Intensely pleasurable responses to music correlate with activity in brain regions implicated in reward and emotion. Proceedings of the National Academy of Sciences, 98(20), 11818–11823. https://doi.org/10.1073/pnas.191355898 Brattico, E., Alluri, V., Bogert, B., Jacobsen, T., Vartiainen, N., Nieminen, S., & Tervaniemi, M. (2011). A Functional MRI Study of Happy and Sad Emotions in Music with and without Lyrics. Frontiers in Psychology, 2. https://doi.org/10.3389/fpsyg.2011.00308 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 18 REFERENCES Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, A. I., Erramuzpe, A., Kent, J. D., Goncalves, M., DuPre, E., Snyder, M., Oya, H., Ghosh, S. S., Wright, J., Durnez, J., Poldrack, R. A., & Gorgolewski, K. J. (2019). fMRIPrep: A robust preprocessing pipeline for functional MRI.Nature Methods,16 (1), 111–116. https://doi.org/10.1038/s41592-018-0235-4 Esteban, O., Markiewicz, C. J., Goncalves, M., Provins, C., Kent, J. D., DuPre, E., Salo, T., Ciric, R., Pinsard, B., Blair, R. W., Poldrack, R. A., & Gorgolewski, K. J. (2023, August). fMRIPrep: A robust preprocessing pipeline for functional MRI. https://doi.org/10.5281/ZENODO.852659 Faro Viana, M., Almeida, P., & Santos, R. C. (2001). Adapta¸ c˜ ao portuguesa da vers˜ ao reduzida do Perfil de Estados de Humor – POMS. An´ alise Psicol´ ogica,19(1), 77–92. https://doi.org/10.14417/ ap.345 Freitas, C., Manzato, E., Burini, A., Taylor, M. J., Lerch, J. P., & Anagnostou, E. (2018). Neural Correlates of Familiarity in Music Listening: A Systematic Review and a Neuroimaging Meta- Analysis. Frontiers in Neuroscience, 12, 686. https://doi.org/10.3389/fnins.2018.00686 Hebart, M. N., & Baker, C. I. (2018). Deconstructing multivariate decoding for the study of brain function. NeuroImage, 180, 4–18. https://doi.org/10.1016/j.neuroimage.2017.08.005 Hoefle, S., Engel, A., Basilio, R., Alluri, V., Toiviainen, P., Cagy, M., & Moll, J. (2018). Identifying musical pieces from fMRI data using encoding and decoding models. Scientific Reports, 8(1), 1–13. https://doi.org/10.1038/s41598-018-20732-3 Juslin, P. N. (2013). From everyday emotions to aesthetic emotions: Towards a unified theory of musical emotions. Physics of Life Reviews, 10(3), 235–266. https://doi.org/10.1016/j.plrev.2013.05.008 Just, M. A., Cherkassky, V. L., Aryal, S., & Mitchell, T. M. (2010). A Neurosemantic Theory of Concrete Noun Representation Based on the Underlying Brain Codes (O. Sporns, Ed.). PLoS ONE, 5(1), e8622. https://doi.org/10.1371/journal.pone.0008622 Kober, H., Barrett, L. F., Joseph, J., Bliss-Moreau, E., Lindquist, K., & Wager, T. D. (2008). Func- tional grouping and cortical–subcortical interactions in emotion: A meta-analysis of neuroimaging studies. NeuroImage, 42(2), 998–1031. https://doi.org/10.1016/j.neuroimage.2008.03.059 Koelsch, S. (2020). A coordinate-based meta-analysis of music-evoked emotions. NeuroImage, 223, 117350. https://doi.org/10.1016/j.neuroimage.2020.117350 Koelsch, S., Cheung, V. K. M., Jentschke, S., & Haynes, J.-D. (2021). Neocortical substrates of feelings evoked with music in the ACC, insula, and somatosensory cortex. Scientific Reports, 11(1), 10119. https://doi.org/10.1038/s41598-021-89405-y Law, L. N. C., & Zentner, M. (2012). Assessing Musical Abilities Objectively: Construction and Validation of the Profile of Music Perception Skills (J. Snyder, Ed.). PLoS ONE , 7(12), e52508. https: //doi.org/10.1371/journal.pone.0052508 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint

19 Lindquist, K. A., Satpute, A. B., Wager, T. D., Weber, J., & Barrett, L. F. (2016). The Brain Basis of Positive and Negative Affect: Evidence from a Meta-Analysis of the Human Neuroimaging Literature.Cerebral Cortex,26 (5), 1910–1922. https://doi.org/10.1093/cercor/bhv001 Mart ´ ınez-Molina, N., Mas-Herrero, E., Rodr ´ ıguez-Fornells, A., Zatorre, R. J., & Marco-Pallar´ es, J. (2016). Neural correlates of specific musical anhedonia. Proceedings of the National Academy of Sciences, 113(46). https://doi.org/10.1073/pnas.1611211113 Nilearn contributors, Chamma, A., Frau-Pascual, A., Rothberg, A., Abadie, A., Abraham, A., Gramfort, A., Savio, A., Cionca, A., Sayal, A., Thual, A., Kodibagkar, A., Kanaan, A., Pinho, A. L., Joshi, A., Idrobo, A. H., Kieslinger, A.-S., Kumari, A., Rokem, A., . . . N´ ajera, ´O. (2025, January). Nilearn. https://doi.org/10.5281/ZENODO.14697221 Pereira, C. S., Teixeira, J., Figueiredo, P., Xavier, J., Castro, S. L., & Brattico, E. (2011). Music and Emotions in the Brain: Familiarity Matters. PLoS ONE, 6(11). https://doi.org/10.1371/journal. pone.0027241 Pereira, F., Mitchell, T., & Botvinick, M. (2009). Machine learning classifiers and fMRI: A tutorial overview. NeuroImage, 45(1), S199–S209. https://doi.org/10.1016/j.neuroimage.2008.11.007 Pessoa, L. (2017). A Network Model of the Emotional Brain. Trends in Cognitive Sciences, 21(5), 357– 371. https://doi.org/10.1016/j.tics.2017.03.002 Putkinen, V., Nazari-Farsani, S., Sepp¨ al¨ a, K., Karjalainen, T., Sun, L., Karlsson, H. K., Hudson, M., Heikkil¨ a, T. T., Hirvonen, J., & Nummenmaa, L. (2021). Decoding Music-Evoked Emotions in the Auditory and Motor Cortex. Cerebral Cortex, 31(5), 2549–2560. https://doi.org/10.1093/ cercor/bhaa373 Rolls, E. T., Huang, C.-C., Lin, C.-P., Feng, J., & Joliot, M. (2020). Automated anatomical labelling atlas 3. NeuroImage, 206, 116189. https://doi.org/10.1016/j.neuroimage.2019.116189 Sachs, M. E., Habibi, A., Damasio, A., & Kaplan, J. T. (2018). Decoding the neural signatures of emo- tions expressed through sound. NeuroImage, 174, 1–10. https://doi.org/10.1016/j.neuroimage. 2018.02.058 Salimpoor, V. N., Zald, D. H., Zatorre, R. J., Dagher, A., & McIntosh, A. R. (2015). Predictions and the brain: How musical sounds become rewarding. Trends in Cognitive Sciences, 19(2), 86–91. https://doi.org/10.1016/j.tics.2014.12.001 Sayal, A., Direito, B., Sousa, T., Singer, N., & Castelo-Branco, M. (2025). Music in the loop: A system- atic review of current neurofeedback methodologies using music. Frontiers in Neuroscience, 19, 1515377. https://doi.org/10.3389/fnins.2025.1515377 Sayal, A., Guedes, A. G., Almeida, I., Pereira, D. J., Lima, C. F., Panda, R., Paiva, R. P., Sousa, T., Castelo-Branco, M., Bernardino, I., & Direito, B. (2024). Decoding Musical Valence and Arousal: .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint 20 REFERENCES Exploring the Neural Correlates of Music-Evoked Emotions and the Role of Expressivity Features. IEEE Transactions on Affective Computing, 1–13. https://doi.org/10.1109/TAFFC.2024.3507192 Schubert, E. (2013). Emotion felt by the listener and expressed by the music: Literature review and theoretical perspectives. Frontiers in Psychology, 4. https://doi.org/10.3389/fpsyg.2013.00837 Spotipy contributors. (2024). Spotipy: A light weight Python library for the Spotify Web API. Tervaniemi, M. (2023). The neuroscience of music – towards ecological validity. Trends in Neurosciences, 46(5), 355–364. https://doi.org/10.1016/j.tins.2023.03.001 Trost, W., Ethofer, T., Zentner, M., & Vuilleumier, P. (2012). Mapping Aesthetic Musical Emotions in the Brain. Cerebral Cortex, 22(12), 2769–2783. https://doi.org/10.1093/cercor/bhr353 Tyszka, M. (2023). Bidskit: Utility functions for working with DICOM and BIDS neuroimaging data. Varoquaux, G., Raamana, P. R., Engemann, D. A., Hoyos-Idrobo, A., Schwartz, Y., & Thirion, B. (2017). Assessing and tuning brain decoders: Cross-validation, caveats, and guidelines. NeuroImage, 145, 166–179. https://doi.org/10.1016/j.neuroimage.2016.10.038 Vuust, P., Heggli, O. A., Friston, K. J., & Kringelbach, M. L. (2022). Music in the brain. Nature Reviews Neuroscience. https://doi.org/10.1038/s41583-022-00578-5 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. https://doi.org/10.1038/nmeth.1635 Zentner, M., Grandjean, D., & Scherer, K. R. (2008). Emotions evoked by the sound of music: Characteri- zation, classification, and measurement. Emotion, 8(4), 494–521. https://doi.org/10.1037/1528- 3542.8.4.494 Zentner, M., & Strauss, H. (2017). Assessing musical ability quickly and objectively: Development and validation of the Short-PROMS and the Mini-PROMS. Annals of the New York Academy of Sciences, 1400(1), 33–45. https://doi.org/10.1111/nyas.13410 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 31, 2025. ; https://doi.org/10.1101/2025.07.28.667190doi: bioRxiv preprint