Brain network dynamics reflect psychiatric illness status and transdiagnostic symptom profiles across health and disease

network dynamics, transdiagnostic, psychiatric, symptom profiles, brain-based prediction 24 25 26 27 28 29 30 31 32 33 34 35 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 2

36 The network organization of the human brain dynamically reconfigures in response to changing 37 environmental demands, an adaptive process that may be disrupted in a symptom -relevant manner 38 across psychiatric illnesses. Here, in a transdiagnostic sample of participants with (n=134) and without 39 (n=85) psychiatric diagnoses, functional connectomes from intrinsic (resting-state) and task-evoked fMRI 40 were decomposed to identify constraints on brain network dynamics across six cognitive states . 41 Hierarchical clustering of 110 clinical, behavioral, and cognitive measures identified participant -specific 42 symptom profiles, revealing four core dimensions of functioning: internalizing, externalizing, cognitive, 43 and social/reward. Brain network dynamics were flattened across cognitive states in individuals with 44 psychiatric illness and could be used to accurately separate dimensional symptom profiles more robustly 45 than both case-control status and primary diagno stic grouping. A key role of inhibitory cognitive control 46 and frontoparietal network interactions was uncovered through systematic model comparison. We 47 provide novel evidence that brain network d ynamics can accurately differentiate the extent that 48 psychiatrically-relevant dimensions of functioning are exhibited across health and disease. 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 .CC-BY-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 3

71 The human brain is a complex dynamical system 1,2 that enables rapid, context -relevant information 72 processing to occur in a time-varying manner3,4. This fundamental property of brain functioning enables 73 flexible and adaptive responses within brain regions and across large -scale functional networks 5-15. 74 Substantial progress has been made in characterizing topological properties that allow large-scale brain 75 networks to maintain a balance between functional flexibility and the stability required for central nervous 76 system integrity7,16,17. For example, intrinsic (i.e., resting-state)18 functional connectivity has been 77 characterized in terms of its small worldness19,20, hub structure21,22, hierarchical organization23-25, multiple 78 network assignments26-28, and probabilistic distributions 29. This literature provides converging evidence 79 for the key role of network-based mechanisms in the facilitation of flexible information processing across 80 the cortical sheet 7,16,30. While resting -state connectivity patterns reliably approximate the properties of 81 functional specialization within and between large-scale brain systems 31-33, task -linked changes in 82 connectivity patterns are thought to enable dynamic shifts in context -dependent processing over 83 time16,34,35. Moreover, there is mounting evidence that time-varying, flexible shifts in connectivity patterns 84 (here termed reconfigurations30,36-39) more optimally account for individual differences in both cognition 85 and behavior than traditional, “static” approaches for studying brain functions7,40-44. Importantly, given that 86 flexible processing is key for adaptively responding to changi ng environmental or cognitive demands , 87 alterations in brain network dynamics are theorized to contribute to the onset and maintenance of 88 psychiatric illness45-49. However, despite the importance of network reconfigurations in accounting for 89 individual differences in behavior across health and disease, the extent that brain network dynamics link 90 with psychiatric symptom structure remains unclear. 91 92 Disrupted brain dynamics may constitute a key mechanism that accounts for symptom profiles across 93 patient populations. This is evident, for example, in patients diagnosed with schizophreni a, where 94 electrophysiological studies have shown disrupted oscillatory activity and transient synchronizations , 95 particularly in low-frequency bands4,50,51. Recent work has uncovered dynamic connectivity patterns in 96 patients diagnosed with bipolar disorder and schizophrenia that reflect symptoms of mania and 97 psychosis52-56. Suboptimal network dynamics have also been reported in obsessive compulsive 98 disorder57, attentional disorders 58, and major depressive disorder 59,60. Importantly, with respect to 99 predicting diagnostic status, network dynamics (Fig. 1A , e.g., inter- or across -state changes in 100 connectivity, as in Fig. 1B ) tend to outperform measures that assume stable, temporally-invariant, 101 patterns of connectivity53,58,61. However, despite growing evidence for altered network dynamics in patient 102 populations, the extent that shifts in connectivity patterns across cognitive states accounts for individual-103 specific symptom structure remains to be established. Here, we examine brain-based features that 104 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 4 constrain across-state network reconfigurations, and how such dynamics may be impaired in a symptom-105 relevant manner across psychiatric illnesses (Fig. 1B). 106 107 One longstanding challenge in establishing reliable links between brain processes and clinical 108 presentation is the observation that psychiatric symptoms tend to co -occur across discrete diagnostic 109 categories, while the expression of symptoms within a given diagnosis group can also be highly variable62-110 70 (Fig. 1A). For instance, patients diagnosed with either major depressive disorder or schizophrenia can 111 present with internalizing symptoms, such as anhedonia, avolition, fatigue, and flat affect71,72. There may 112 also be overlapping expressions of cognitive deficits across diagnoses, such as a reduced ability to plan, 113 impaired working memory, and a tendency to ruminate73,74. Moreover, individuals not meeting criteria for 114 diagnosis can also express subclinical or transient functional defi cits. For example, a non -diagnosed 115 individual may temporarily experience low levels of mood, motivation, or reward sensitivity. Observations 116 such as this have prompted some theorists to reframe psychopathology as a hierarchy of deficits across 117 broad dimensions (also termed domains) of cognitive and behavioral functioning, rather than a collection 118 of distinct diagnostic constructs 75-80, emphasizing the importance of individualized, precision medicine 119 approaches81,82. 120 121 Here, we adopt a transdiagnostic framework by analyzing data from a sample of individuals with and 122 without psychiatric diagnoses. We used cognitive and behavioral measures to uncover individual ly-123 profiled symptom structures across core dimensions of psychiatrically-relevant functioning62,68,69,83-85 (Fig. 124 1), which we term symptom fingerprints. Given the importance of brain network dynamics to adaptive 125 information processing, we hypothesized that reconfigur ing connectivity patterns across a varied set of 126 cognitive states would accurately separate individuals based on dimensionally-organized symptom 127 structure3,53,58,61 (Fig. 1), as well as by their case-control status and primary diagnostic labels. We used 128 non-negative matrix factorization (NMF) to model brain network reconfiguration dynamics, which can 129 reveal underlying constraints upon connectivity patterns as individuals transition across cognitive states86-130 92. We discovered that network reconfiguration dynamics were flattened (i.e., less differentiated) across 131 cognitive states in patients, and these dynamics could be used to accurately predict case-control status, 132 primary diagnoses, and participant-specific symptom fingerprints. The link between network dynamics 133 and dimensionally-organized symptom fingerprints was prominently driven by shifts from intrinsic to 134 cognitive control93,94 (i.e., Stroop95,96) task states. Further, control (frontoparietal), thalamic, and task -135 linked (i.e., visual and somato/motor) network regions were particularly important for classifying symptom 136 fingerprints. 137 138 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 5 139 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 6 Figure 1. Theoretical framework. (A) Schematic distinguishing diagnostic (top) and dimensional 140 phenotypes (bottom), with respect to establishing brain -behavior relationships. Color saturations: 141 symptom expression levels in two example dimensions of functioning: internalizing and cogniti on. 142 Uniform diagnoses: example individuals with the same diagnosis but variable symptom expression. 143 Varied diagnoses: different diagnoses but overlapping symptoms. Without diagnoses: subclinical or 144 transient expression. For simplicity, individuals with zero symptoms are not shown. Dimensional 145 phenotyping groups individuals based on symptom expression patterns ( fingerprints), which are thus 146 highly consistent in each group. We propose that, generally, neurobiological features better account for 147 the dimensionally -based phenotypes (right circle plots), given increased consistency in underlying 148 symptom structure. Further, it is likely that some brain features link with behavior better than others; with 149 brain network dynamics outperforming static network features, uni variate brain activity, and chance. 150 Noise ceiling: theoretical cap on explainable variance with current neuroimaging, although this may 151 increase with future methods. (B) Example schematic of how clinically-relevant brain network dynamics 152 may be exhibited a cross cognitive states , for example, shifting from rest to task states . In health, 153 connectivity patterns shift in a small but functionally -relevant manner to optimally meet task demands. 154 Here, there is a differentiable change in connectivity that flexibly and adaptively supports the online 155 recruitment of processing resources. In less healthy systems, one possibility is that rest-to-task shifts in 156 connectivity are suboptimal such that little-to-no reconfiguration (flattened, or less differentiated) leads to 157 failures in recruiting the resources needed to account for task demands or behavioral goals. 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 7

175 Brain network dynamics are flattened in psychiatric illness 176 We applied non-negative matrix factorization (NMF) to functional connectivity (FC) estimates from each 177 participant and cognitive state to characterize across -state brain network dynamics (Fig. 2A-C). NMF 178 has previously identified transition dynamics between states with varied cognitive demand in healthy 179 adults86. NMF decomposes positively-thresholded inputs (here: across-state FC patterns) into two lower 180 rank matrices ( Fig. 2C ). These matrices can be multiplied to approximately reconstruct the input 181 connectivity patterns, and each extract distinct, but related, attributes of input connectivity patterns. One 182 is a features matrix, which is similar to a basis set – here termed subgraphs. When applied to an across-183 state connectivity input, subgraph values quantify features constraining brain network dynamics86,97. The 184 next matrix contains coefficients, analogous to an encoding matrix. Coefficients quantify the expression 185 of each subgraph over each observation, which here were cognitive states and participants. Consistent 186 with other dimensionality reduction techniques, NMF extracts a hidden structure that best accounts for a 187 given input dataset. Compared to other techniques, such as PCA, NMF has the benefit of being parts -188 based, allowing for straightforward inferences regarding reconstruction accuracy. NMF is also less 189 constrained with respect to orthogonality. Non -orthogonality is neurobiologically relevant for multi -state 190 or multi-modal data, as features underlying time -varying neural data are likely expressed with some 191 overlap. For example, while we explicitly examined reconfigurations across intrinsic and task-evoked FC 192 patterns, these states share a structural connectivity backbone 98-100 and their underlying dynamics are 193 not strictly orthogonal. NMF -uncovered subgraph features, and corresponding expressions of those 194 subgraphs, allowed us to infer key constraints upon the dynamics that underlie across-state brain network 195 reconfigurations. Cross-validation of the discovery dataset tuned NMF parameters as follows: alpha 196 (regularization)=0.2, beta (loss)=1.2, and k (number of subgraphs)=5 (Supplemental Fig. S1). 197 198 While we tested hypotheses about dimensionally-organized symptom profiles downstream (see following 199 Results), we first tested the hypothesis that network dynamics are relatively flattened in participants with 200 psychiatric diagnoses (Fig. 1). The across-state expression patterns of subgraph coefficients ( Fig. 2C) 201 were less differentiated (i.e., less variable, or more flattened, across states) in those with psychiatric 202 diagnoses relative to those without diagnoses (Fig. 2D). Given that across-state subgraph coefficient 203 patterns were expressed consistently in each of the k=5 subgraphs for both groups ( Fig. 2D), we first 204 restructured each subgraphs ’ features (Fig. 2C, middle matrix; Supplemental Fig. S1 ) into standard 205 region-by-region networks ( Fig. 2A-B) and then applied a network efficiency metric 101-103. Subgraph 1 206 exhibited the greatest network efficiency ( Fig. 2E) both locally (brain regions) and globally (functional 207 brain networks), suggesting that information flows across this subgraph’s functional archit ecture more 208 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 8 readily than subgraphs 2 through 5. Thus, for ease of interpretation, main-text Results refer to subgraph 209 1 and subgraphs 2 through 5 are reported in the Supplement. 210 211 Next, we quantified the extent that network dynamics were flattened for subgraph 1 (subgraphs 2-5: 212 Supplemental Fig s. S1-S2) by applying cosine similarity to the vectors of participants’ expression 213 coefficients for each state -to-state pair. Thus, a higher cosine similarity score indicates that subgraph 214 features underlying FC reconfiguration dynamics were expressed similarly across cognitive states. We 215 subtracted without-diagnosis similarity scores from with-diagnosis similarity scores for each pair of states 216 (Fig. 2F). Thus, a positive difference in cosine similarity indicated relative flattening of dynamics, which 217 we expected for non-diagnosed participants relative to diagnosed participants (Fig. 1B). In the majority 218 of state -to-state pairings, the transdiagnostic psychiatric sample exhibited more flattened subgraph 219 expression coefficients. Across state -to-state pairs, the average cosine similarity of those without 220 diagnosis was 0.91 (standard deviation (SD)=0.01) and 0.95 with diagnoses (SD=0.03; cosine similarity 221 can range from -1 to 1). The difference of with -diagnosis minus without -diagnosis cosine similarities 222 across all state-to-state pairs was significantly greater than zero (t(14)=5.08, p=8.38 x 10-5), suggesting 223 that the dynamics constraining across-state brain network reconfigurations were comparatively flattened 224 in those with psychiatric diagnoses versus those without diagnoses. 225 226 Next, we explored the extent that primary diagnostic groups104 exhibited flattened brain network dynamics 227 (Fig. 2G). The most flattened dynamics were exhibited by those diagnosed with panic disorder, eating 228 disorder, obsessive compulsive disorder, and attention-deficit/hyperactivity disorder (and “other”, which 229 had n=2 therefore we caution against over-interpretation). A moderate amount (in this sample) of flattened 230 brain network dynamics was exhibited by those diagnosed with an anxiety disorder, bipolar disorder, a 231 depressive disorder , schizophrenia/schizoaffective disorder, posttraumatic stress disorder, and 232 substance use disorder. Dynamics were flattened (given by cosine similarity, Fig. 2G) to a similar extent 233 in varied diagnostic groups. This is consistent with the literature suggesting that neurobiological estimates 234 may be noisy across select diagnostic categories ( Fig. 1A) 67. To further account for this, we examined 235 the relationship between brain network reconfiguration dynamics and dimensions of functioning 236 underlying psychiatric illness in remaining sections. 237 238 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 9 239 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 10 Figure 2. Across -state brain network dynamics in health and psychiatric illness. (A) The 240 organization of the human cerebral cortex is revealed through patterns of intrinsic functional connectivity 241 and anatomical segmentation. Left: network organization of the human cortex was based on the 17 -242 network solution from Yeo et al.105 across the 400-parcel atlas from Yan et al.106 (black borders around 243 cortical regions). Control network = frontoparietal cognitive control network. Right: non-cortical regions 244 were based on Tian et al.107 (subcortex) and Buckner et al.108 (cerebellum). Network color-coding is used 245 in all subsequent figures. (B) FC estimates (across-participant average r, Fisher-z transformed) for each 246 of the six fMRI scans (cognitive states). Regions (x, y axes) sorted per network assignment. Grey arrow: 247 FC was used as input to NMF in panel C. “Region” refers to functional parcellation of the cortex or 248 anatomical subdivisions of non-cortex, and the term “network” refers to a collection of such regions with 249 underlying organization, for simplicity (see Methods for details). (C) NMF pipeline. Input across -state 250 network configuration matrix (left): FC estimates from upper triangle of B (positively thresholded) ; 251 flattened and stacked vertically, by state and participant (black lines). NMF decomposes this into 252 subgraph features (middle) and expression coefficients (right). (D) Across-participant average expression 253 coefficients (C, right; normalized relative percent) for each subgraph and state. Here, participants were 254 grouped by case-control status. (E) Local efficiency of subgraph features (C, middle), projected back into 255 region-by-region networks. Dots: brain regions (averaged across participants; grand average via black 256 line) color-matched for networks. Subgraph 1 exhibited the most efficiency, suggesting information flows 257 across these features more readily. We highlight subgraph 1 based on this (and NMF orthogonality) in all 258 downstream analyses (2-5 subgraph results in Supplement). ( F) Across-participant cosine similarity of 259 expression coefficients (D) were computed for each group and pair of states. Non -diagnosis-group 260 similarities were subtracted from diagnosis -group similarities. Network dynamics were more flattened 261 (less differentiated) across participants with psychiatric diagnoses in most state -to-state comparisons. 262 (G) Same as F, but participants grouped by primary diagnosis (rows). NO DX: no diagnosis; ANX: anxiety; 263 BP: bipolar disorder; DEP: depression; ED: eating disorder; PD: panic disorder or phobia; SCZ/SZA: 264 schizophrenia/schizoaffective; SUD: substance use disorder. Rightmost column: average cosi ne 265 similarity (here, an index flattened dynamics) of each diagnostic group (i.e., averaging the lower triangle 266 in F, without group-based subtraction). 267 268 269 270 271 272 273 274 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 11 Symptom fingerprints: core dimensions of functioning are exhibited across health and psychiatric illness 275 Known limitations in identifying robust links between psychiatrically -relevant behavior and the brain 276 include heterogeneous symptomatology within diagnostic categories as well as overlapping symptom 277 structures across diagnostic groups 67,68,84 (Fig. 1A). One meta -analysis reported that resting -state 278 networks exhibited alterations common to more than six distinct diagnostic categories109. This suggests 279 that prior methods were either poorly constrained with respect to differentiating diagnostic groups (and 280 likely more applicable to transdiagnostic symptom expression), or that static-network-based biomarkers 281 may lack sensitivity to a single diagnostic group (or both). Here, we accounted for two key constructs for 282 psychiatric phenotyping. First, by using the TCP dataset110, which includes participants with and without 283 varied mental health concerns and diagnoses. Second, we incorporated behavioral data with wide 284 coverage across dimensions of functioning (Supplemental Table S1). This strategy aims to improve 285 validity by optimizing the balance between precise, individualized phenotyping and broad, group -level 286 diagnostic bins. 287 288 After curating, transforming, and imputing 110 behavioral measures that covered a wide breadth of 289 cognitive, behavioral, and psychiatric domains (henceforth referred to collectively as behavioral 290 measures for simplicity ) (Supplemental Table S1 and Figs. S 3, S4), we performed hierarchical 291 clustering80,111 on the resulting matrix of individual differences correlations (Fig. 3A). The optimal number 292 of clusters according to 18 performance indices was four (Fig. 3A black boxes), which is consistent with 293 prior work examining the latent structure of various behaviors80,112,113. To limit researcher bias in naming 294 these clusters, we pre -labeled each measure with possible dimensions of functioning ( full mapping: 295 Supplemental Table S1, Fig. S4). Following a separate PCA on the observed scores of each of the four 296 clusters of measures, we identified which dimensions of functioning was dominant in each cluster by the 297 largest percentage of labels, which were weighted by factor loadings on the first PC. Accordingly, the four 298 clusters were labelled: internalizing (approximately 40% of measures in the cluster had this as the primary 299 label), externalizing (31% of measures), cognition (27%), and social/reward (20%). Canonical examples 300 of measures in the internalizing cluster included (Fig. 3B) the Depression, Anxiety, and Stress Scales114 301 and the neuroticism factor of the NEO Five -Factor Inventory-3 (NEO-FFI-3115). Example externalizing 302 measures included the extraversion factor of the NEO -FFI-3 and the Young Mania Rating Scale 116 303 irritability subscale. Example cognition measures included the entire TestMyBrain suite 117 and Shipley 304 intelligence measures 118. Example s ocial/reward measures included the Experiences in Close 305 Relationships119 scales and the Temperament and Character Inventory120 novelty seeking subscale. 306 307 To quantify phenotypes that represented multi -dimensional information (i.e., all cluster expressions 308 considered simultaneously) and were still differentiable, we developed a method termed symptom 309 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 12 fingerprinting. With four clusters ( dimensions of functioning), there were 16 possible combinations of 310 binarized expressions (i.e., expressed cluster versus did not express cluster; Fig. 3C), which can also be 311 thought of as symptom profiles. Here, expression was quantified via each participant’s score on the first 312 PC of a given cluster (log likelihood; see Methods). One-sample t-testing was applied to cluster ized 313 expression scores and compared each participant’s score versus all other participants’ scores (multiple-314 comparisons corrected). Thus, binarization specifically refers to significant cluster expression versus non-315 significant expression ( Fig. 3C ). For quality assurance, we examined the d istribution of participants’ 316 primary psychiatric diagnoses and case-control statuses across the 16 symptom fingerprints (Fig. 3D). 317 As expected, the fingerprint consisting solely of expressing the internalizing cluster included a dominant 318 percentage of individuals with depression diagnoses. The fingerprint with all four clusters expressed had 319 a diverse and more evenly spread distribution of individuals with and without psychiatric diagnoses. 320 Perhaps reflecting the multiple domains of functioning impacted by PTSD121,122, individuals with that 321 primary diagnosis were st rongly represented in the internalizing -cognition fingerprint as well as the 322 internalizing-externalizing-social fingerprint. Altogether we validated an empirically-driven approach to 323 individually profile dimensional symptom structures that were exhibited by a transdiagnostic cohort of 324 participants with and without psychiatric diagnosis. 325 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 13 326 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 14 Figure 3. Uncovering dimensional symptom profiles. (A) Individual differences correlations ( the 327 extent that behavioral measure scores covary across individuals; rho) between all pairs of 110 behavioral 328 measures (see Methods; Supplemental Fig. S3 ). Agglomerative hierarchical clustering was applied, 329 and 18 metrics indicated a 4 -cluster solution was optimal. Dendrogram: Ward’s distance. Black boxes 330 outline clusters, named : internalizing, externalizing, cognition, and social/reward (see: Supplemental 331 Table S1). (B) After PCA of measures in each cluster (independently assessed), the top measures (factor 332 loadings) in PC 1 are shown ( Supplemental Fig. S4); this guided naming c onventions for the four 333 dimensions of functioning in A and downstream. (C) Developing symptom fingerprints: each participant’s 334 expression of the first PC of each cluster ( see Methods ). Top: all possible expression patterns (16 335 fingerprints), given by combinations of significant and non-significant expressions of each cluster. Middle: 336 raw expression scores (log-likelihood) for all participants. Bottom: after testing whether each expression 337 score in the middle panel was significant, they were binarized into o ne of 16 possible patterns per 338 participant. ( D) The percent -based distributions of case -control (top) and primary diagnosis (bottom) 339 groups exhibiting each of the symptom profiles (or fingerprints) in C. Here, y-axes capture 100% of the 340 total for each fingerprint individually. Abbreviations: int.: internalizing; ext.: externalizing; cog.: cognition; 341 social: social/reward. Expected patterns relevant to quality assurance: participants diagnosed with 342 depression were the largest proportion of the internalizing-only fingerprint; there was a diverse spread in 343 the all-cluster (rightmost) fingerprint; and participants with anxiety diagnoses were in the externalizing 344 and social fingerprint. 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 15 Brain network dynamics classify dimensional symptom fingerprints across health and disease 362 We used support vector classification (SVC 123) to test the hypothesis that brain network dynamics are 363 linked with dimensional phenotypes across health and psychiatric illness. Subgraph expression 364 coefficients exhibited across six cognitive states (Fig. 2) were used to classify 16 symptom fingerprint 365 labels ( Fig. 3 ). To train the model, we randomly subsampled 80% of the discovery dataset (1000 366 permutations) and tested classification accuracy on the held -out validation dataset (see Methods). We 367 built null models by randomly shuffling the fingerprint labels 1000 times to empirically obtain chance-level 368 classification accuracy. Empirically -based chance accuracy was highly similar to theoretical chance, 369 which was 1/16 or 6.25% (empirical chance=6.6%). When classifying fingerprints, there were two types 370 of accuracy: strict and fuzzy. Strict accuracy was based on an all-or-nothing decision boundary of correct 371 (100% accurate) or incorrect (0%). Given that symptom fingerprints were individualized profiles based on 372 binarized patterns of cluster expression, we also considered a “fuzzier” decision boundary that accounted 373 for the extent of overlap between predicted and actual fingerprint labels. Fuzzy accuracy was based on 374 the percent o f four possible cluster expressions that were correctly classified, and included possible 375 values of 0%, 25%, 50%, 75%, and 100% on each permutation. An example fuzzy accuracy score of 376 50% is a predicted fingerprint label of internalizing-externalizing (1, 1, 0, 0) and an actual fingerprint label 377 of internalizing-cognition (1, 0, 1, 0). This example model is half correct in identifying internalizing cluster 378 expression, but half incorrect in mislabeling cognition -cluster expression as externalizing -cluster 379 expression. 380 381 Symptom fingerprints were classified significantly above chance (strict: t(42)=7.2, p=3.8x10-9, Cohen’s 382 D=17.8; fuzzy: t(42)=24.8, p=5.4x10-27, Cohen’s D=61.2; Fig. 4A). We implemented two comparison 383 models with the same SVC pipeline except with the outcome labels of: (1) case-control status (t(42)=10.7, 384 p=6.4x10-14, Cohen’s D=3.3), and (2) primary diagnosis group ( t(42)=5.2, p=2.9x10-4, Cohen’s D=9.6; 385 Fig. 4A). While brain network dynamics were able to significantly classify all approaches to g rouping 386 individuals, dimensional symptom profiles were classified with the largest effect size via both strict and 387 fuzzy accuracy. This suggests that the boundaries between groups of individuals exhibiting symptom 388 fingerprints are separated more precisely by brain network dynamics than case-control status or primary 389 diagnosis categories. 390 391 Next, we examined the extent that each large-scale functional network (as in Fig. 2A) contributed to the 392 accurate classification of individual symptom profiles. For each functional network, we iteratively applied 393 simulated lesioning of FC estimates, within- and between-network. Then, we re-implemented NMF (Fig. 394 2) to uncover reconfiguration dynamics without brain regions from that network considered. Next, we 395 classified symptom fingerprints (as in Fig. 4A, the “full model”) and quantified changes in accuracy from 396 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 16 the full model. A statistically significant (nonparametric permutation testing; see Methods) decrease in 397 accuracy indicated which connectivity patterns were i nfluential in linking brain network dynamics with 398 symptom profiles. These networks included: control (frontoparietal) A, limbic A/B, somato/motor A/B, 399 visual B, and thalamus (Fig. 4B-C). This suggests that brain systems known to enable cognitive control 400 processing (frontoparietal and thalamus) and networks linked with task -relevant resource processing 401 (visual and somato /motor) exhibit dynamics that are prominently linked with dimensionally -organized 402 cognitive and behavioral functioning across health and psychiatric disease. 403 404 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 17 405 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 18 Figure 4. Brain network dynamics classified dimensional symptom fingerprints better than case-406 control status and primary diagnoses. (A) Classification accuracy of case-control status, primary 407 diagnosis grouping, and symptom fingerprints (left-to-right) (dots: participants; colors, all plots: primary 408 diagnosis) of the validation dataset. In each model, predictors included NMF -uncovered subgraph 409 expression coefficients across all six cognitive states (three resting - and three task-states). Effect sizes 410 (Cohen’s D) suggest that while all models were significant, classification accuracies of fingerprints (strict 411 and fuzzy) were the most robust. t-statistics against empirical chance (dotted gray lines) based on 1000 412 permuted null models reported in-text. Theoretical and empirical chance (left, right) on dotted lines. ( B) 413 Classification of fingerprints was iteratively performed with each large-scale functional network withheld 414 (simulated lesioning) before NMF. In each iteration, features underlying brain network dynamics did not 415 include the lesioned networks’ FC estimates. A significant reduction in model performance suggests 416 dynamics conferred by changing connectivity patterns in a given network were particularly important for 417 classifying symptom fingerprints. Error bars: standard error of the mean across validation-set participants. 418 (C) The same results as in B but projected onto a cortical surface (non-cortex indicated in B). Darker 419 blue: functional brain networks whose across -state dynamics were particularly important for classifying 420 dimensional symptom profiles. 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 19 Cognitive control processing influences the link between brain network dynamics and dimensional 441 symptom profiles 442 To test how influential each cognitive state was to the link between brain network dynamics and symptom 443 fingerprints, we implemented NMF with varied combinations of input connectivity patterns. This allowed 444 us to differentiate context-specific shifts in connectivity from general rest-to-task shifts in connectivity (as 445 in Fig. 4). The seven comparison models included these inputs to NMF: (1) all states ( Fig. 4; a general 446

model); (2) resting-state FC only, which likely contains small shifts in intrinsic18,31,105,124 network 447 organization (not task-linked); (3) task-state FC only, consisting of Stroop96 and Emotional Faces125 FC; 448 (4) resting- and Stroop-task-state FC, likely containing shifts in intrinsic connectivity relevant to cognitive 449 control processing; (5) resting - and Emotional Faces -task-state FC, with shifts relevant to processing 450 fearful faces; (6) Stroop FC only, containing static FC patterns relevant for cognitive control; and (7) 451 Emotional Faces FC only, containing static FC patterns relevant for fear processing. 452 453 Across all models, symptom fingerprints were classified with greater effect size than case-control status 454 and primary diagnostic group (Fig. 5A-D). As expected, there was a broadly consistent pattern of model 455 performances between strict-accuracy and fuzzy -accuracy (Fig. 5C-D). The all-states model exhibited 456 the lowest performance (strict accuracy: D=17.25; fuzzy accuracy: D=54.73), suggesting that network 457 dynamics across wider -spanning task contexts may yield suboptimal decision boundaries separating 458 clinical symptom structures expressed across individuals. The rest-and-Stroop model performed the best 459 (strict accuracy: D=18.85; fuzzy accuracy: D=62.91), suggesting that connectivity shifts between an 460 intrinsic state and a cognitive control state accounted for individual clinical variability particularly well. 461 Interestingly, case-control and primary diagnosis were classified with a similar pattern across m odels. 462 The main difference was that all -states performed worst and best in classifying primary diagnosis and 463 case-control statuses, respectively. These data suggest that separability of case -control status may be 464 sensitive to the number of cognitive states (or, amount of data) used to uncover network dynamics. 465 Primary diagnoses may be classified best by dynamics across similar cognitive contexts. Conversely, 466 dimensional symptom profiles may be separated best by a specific, foundational shift in cognitive context. 467 468 Further, we found that control (frontoparietal) network regions of the first NMF-uncovered subgraph (Fig. 469 5E) in the rest -and-Stroop model strongly exhibited a functional hub property based on participation 470 coefficient126 (Fig. 5F; Supplemental Fig. S5). Additionally, striatal, thalamic, and default C networks 471 included strong hub regions and the default A/B, somato/motor A/B, dorsal attention B, and 472 salience/ventral attention A networks exhibited the fewest hubs. For the majority of networks, hubs either 473 stayed consistent or slightly decreased in the rest -and-Stroop model versus the rest -only model , 474 suggesting that the increased prediction accuracy in the rest -and-Stroop model ( Fig. 5C-D) was not 475 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 20 solely driven by increased diversity in network interaction patterns while engaging with a cognitively 476 demanding task. We propose that accurate separation of symptom fingerprints is influenced by the 477 dynamic shift in cognitive context from an intrinsic state to a more cognitively demanding state. Altogether, 478 and considering Fig. 4 results uncovering the importance of networks linked with cognitive control, we 479 propose that the flexible recruitme nt of cognitive resources to meet increased task demands is a key 480 determinant of dimensional symptom structures exhibited across health and psychiatric disease. 481 482 483 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 21 Figure 5. Reconfiguring connectivity patterns from rest -to-Stroop task states prominently 484 influence the classification of dimensional symptom profiles . (A) Classification models with 485 connectivity patterns from different cognitive states inputted to NMF (model inputs indicated on x-axes; 486 see Methods and text for details). Headers indicate model comparison results for classifying case/control 487 status, primary diagnosis grouping, symptom fingerprints with strict accuracy, and fingerprints with fuzzy 488 accuracy, respectively (effect size of cross-participant model accuracies given by Coh en’s D). Bars are 489 color-coded from blue-to-orange to indicate worst-to-best model performance. In the models assessing 490 how well brain network dynamics classify dimensionally -organized symptom fingerprints, 491 reconfigurations across rest-to-Stroop performed best, suggesting that information processing shifts from 492 an intrinsic state to a cognitive control state account best for the expression of symptom profiles. (B) 493 NMF-uncovered subgraph 1, reformatted into whole -brain region-by-region networks. This was used in 494 (C): regional participation coefficient (normalized) of the best performing model (rest & Stroop) for 495 classifying fingerprints (Supplemental Fig. S5 for other models). 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 22

519 We provide evidence that brain network dynamics are linked with transdiagnostic symptom profiles 520 across health and disease (Fig. 1). The application of NMF to whole-brain functional connectivity patterns 521 across varied cognitive states ( Fig. 2) allowed us to examine dynamic constraints upon across -state 522 shifts in functional network interactions86. Individuals with psychiatric diagnoses exhibited flattened (i.e., 523 less differentiated) network dynamics, particularly across resting - and Stroop-task states. These data 524 suggest that cognitive control/executive functioning is disrupted across psychiatric illnesses45-49, and that 525 suboptimal recruitment of processing resources may emerge via flattened network dynamics. 526 Hierarchically-informed methods 75-80 were used to characterize symptom fingerprints, or individual 527 symptom structure profiles that mapped to core dimensions of functioning (Fig. 3). Four well-established 528 dimensions of functioning were evident across participants: internalizing, externalizing, cognitive, and 529 social/reward (Supplemental Table S1 110). Symptom fingerprints – which profile the extent that an 530 individual expressed each of these dimensions – were accurately classified by brain network dynamics. 531 When compared to primary diagnostic group and case-control status, symptom fingerprints exhibited the 532 largest effect size ( Fig. 4 ). Simulated lesioning of functional brain networks revealed that cognitive 533 control-linked systems (frontoparietal and thalamus) and task -relevant resource systems (visual and 534 somato/motor) were most important to the link between brain network dynamics and symptom 535 fingerprints. Reconfiguration dynamics across resting -to-Stroop states separated fingerprints with the 536 highest accuracy ( Fig. 5), further corroborating the clinical relevance of cognitive control processing. 537 Altogether, these results reveal that individualized symptom profiles exhibited across four core 538 dimensions of functioning can be accurately accounted for by brain network dynamics. 539 540 Resting-state dynamic FC has been show n to outperform static FC and composite static -dynamic FC 541 models in classifying schizophrenia, bipolar disorder, and healthy controls 61. In the present study, we 542 examined a transdiagnostic sample that consisted of 11 primary psychiatric diagnosis groups and 543 investigated the extent that whole -brain network dynamics exhibited across rest- and task -based 544 cognitive states can account for varied models of health and disease. We demonstrated that brain 545 network dynamics can discriminate between dimensionally-organized symptom profiles, case -control 546 status, and primary diagnoses with high accuracy. This supports the proposition that brain dynamics 547 contain more process -relevant information than static network features and thus confer an enhanced 548 separation of the boundaries between diagnostic groups (Fig. 1). This is consistent with recent evidence 549 suggesting that ADHD diagnoses are accounted for by increased range but decreased fluidity in dynamic 550 FC, which was not discoverable via static FC 58. Similarly, individuals with schizophrenia exhibit: altered 551 dwell times across varied states 53; transient connect ivity profiles predicting active psychosis 52; shorter 552 dwell times in states engaging the frontoparietal network54; alterations to dynamic shifting between brain 553 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 23 states55; and less complex trajectories through connectivity state -space56. Dysfunctions in information 554 processing across brain systems have also been linked with less dynamic connectivity repertoires127,128, 555 deviations from healthy reconfiguration statistics 38, and altered microstate topography in select EEG 556 frequency bands129. 557 558 The present study provides key insights on the extent that across -state network reconfigurations track 559 dimensions of cognitive and behavioral functioning. However, why brain network dynamics account for 560 behavior better than static network features remains an open question. We provide evidence that brain 561 connectivity patterns reconfiguring in response to shifts in task context are strongly linked with core 562 domains of functioning expressed across all participants (with and without diagnoses). Building on this 563 (as well as key findings in the literature51,52,61,70,127,130), it is likely that multi-scale, integrated mechanisms 564 of information processing99,131 are embedded in, and revealed through, brain network dynamics, which 565 are in turn approximated by static network properties 3,132. One framework that is particularly relevant to 566 psychiatry is that, while the brain is a dynamical system, so too is symptom expression 4,133, as many 567 illnesses involve cycling or phases over time. In both neural systems and psychiatric symptomatology, 568 research has characterized dynamical regimes with patterns of dominating (attractor) and less stable 569 (repeller) states. For example, obsessive compulsive disorder has been linked with over -stability of 570 attractor states, recapitulated across brain systems at multiple scales 57. Moreover, dynamical systems 571 theory has begun to infer the neurobiological relevance of transitioning between such states17,134, as well 572 as the energetic costs linked with the processing resources needed for differe nt functions135,136. In this 573 account, network dynamics are more closely aligned to dimensionally-organized axes of psychiatry than 574 static network properties (Fig. 1). This is a key hypothesis that future work should address by building on 575 the present methods with longitudinal data collected synchronous to symptom expression. A related 576 framework posits an intrinsic (baseline) set of statistics that characterize brain network dynamics that 577 give rise to task-induced modifications 137-140. Here, the controllability of select brain regions is 578 quantified138, and shifts in network dynamics may be associated with altered information entropy and 579 metabolic costs141. This framing is not directly competitive with dynamical systems theory but does make 580 nonequivalent predictions. For example, a key hypothesis would be the extent that network dynamics 581 account for symptom structure is determined by a breakdown in the efficiency of information processing 582 across varied task contexts. 583 584 There are important considerations when interpreting standard dynamic FC estimates, such as sliding 585 window correlations, particularly with respect to non -stationarity132. Time-invariant FC assumes that the 586 statistics underlying network interaction patterns are static or fixed over time. Therefore, the careful 587 implementation of dynamic FC may improve neurobiological plausibility by accounting for non-stationary 588 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 24 statistical structure in neural processes. However, a critical confound in standard approaches to dynamic 589 connectivity is that such statistics are also influenced by non -neural, nuisance sources 142. This is 590 particularly problematic with psychiatric participants, given increased motion artifacts 143,144, respiratory 591 effects145,146, and altered arousal processing147,148. We sought to address these concerns in three ways. 592 First, the present data were processed with careful consideration of motion correction, denoising, and 593 quality control benchmarking110. Second, we used a train-test-validation data splitting scheme to account 594 for data leakage and reduce spurious effects149,150 (Supplemental Fig. S6). Third, we used a method for 595 uncovering constraints upon brain network dynamics that depended chiefly on the structure of the input 596 data, and contained no free parameters linked with events or windows of time. To make non-stationarity 597 inferences, we instead opted for a model comparison approach (Fig. 5), where connectivity patterns 598 relevant to varied cognitive states were considered to compare reconfiguring network interactions across 599 resting-state-alone, rest -to-Stroop, task -alone, and so forth. The present study advances clinical 600 neuroscience by bypassing undue influence from non -neural sources of variance, however , we 601 encourage future work that systematically compares how well other brain network dynamics methods 602 can discriminate the structure of psychiatric symptoms. 603 604 Finally, cognitive control processing was prominent in establishing a robust link between network 605 dynamics and symptom fingerprints ( Figs. 2, 4 -5). This is consistent with evidence that the flexible 606 recruitment of processing resources is negatively impacted across psychiatric diagnoses 16,45,49,151. 607 However, the extent that network dynamics are altered across cognitive control systems in psychiatric 608 patients – as well as the direction, stability, and generalizability of this effect – remains an i mportant 609 empirical question for future research. The current approach can be readily expanded on. For example, 610 we used state -general estimates of FC – that is, connectivity estimates for the entire duration of the 611 Stroop task, Emotional Faces task, and resting-state. This could be refined in future work by estimating 612 FC based on task context, condition, or other temporally-linked factors. Relatedly, connectivity can be 613 estimated by alternatives to product -moment correlations, such as the improved validity g iven by 614 regularized regression100. NMF has also been successful with multivariate data inputs, such as structural 615 and functional connectivity88, which may be an important extension of the present work given evidence 616 that properties of the structural connectome constrain the spread of psychopathological brain 617 processes130. Lastly, we are enthusiastic about future work that expands upon transdiagnostic symptom 618 fingerprinting, for example, probing if the extent of differentiation in brain network dynamics scales with 619 the extent that dimensions of functioning are expressed in individual symptom profiles. 620 621 Here, we studied a transdiagnostic sample of participants with and without psychiatric diagnoses and 622 demonstrated that whole-brain network reconfiguration dynamics can discriminate symptom fingerprints 623 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 25 with high accuracy. Cognitive control processes and associated brain networks were prominently 624 implicated in these data, suggesting that executive dysfunction is key to developing a brain-process-level 625 account of transdiagnostic psychopathology. Lastly, this work benchmarks a r eadily adaptable pipeline 626 for investigating brain -behavior links with improved neurobiological plausibility ( Fig. 1 ) as well as 627 accounting for known limitations in using discrete diagnostic boundaries in clinical neuroscience. 628 629 Online Methods 630 Data collection 631 Participants 632 Data were collected as part of the Transdiagnostic Connectome Project (TCP) 110 (Supplemental Fig. 633 S6) at the FAS Brain Imaging Center at Yale University in New Haven, Connecticut, US, and the Brain 634 Imaging Center at McLean Hospital in Belmont, Massachusetts, US. All participants were given written 635 informed consent documentation following the protocols of each center’s Institutional Review Board. Full 636 details of the TCP dataset can be found elsewhere110, but in brief, N=241 participants were 18 through 637 70 years old (mean=36.52, SD=13.09 years) and were recruited from the community as well as patient 638 referrals from collaborating psychiatric clinicians. Participants were screened to assess if they met the 639 following inclusion criteria: that they were eligible for MRI scanning, were not colorblind, and were not 640 previously diagnosed with a neurological disorder or abnormality. To maximize data available for 641 functional network analyses in the present study, N=219 TCP participants were included given that these 642 219 participants had neuroimaging data for six of the seven fMRI runs. Of this N=219 in the present study 643 (mean age=36.2, SD=12.9 years; n=93 (42.5%) identified as male, n=123 (56.2%) as female, and n=3 644 (1.4%) as nonbin ary), there were n=134 (61.2%) with psychiatric diagnoses, n=85 (38.8%) without a 645 history of psychiatric diagnoses or treatment (Supplemental Fig. S6C; and see sections below for further 646 details on MRI data acquisition and brain network dynamics analyses). 647 648 Study outline 649 After initial screening, there were three study sessions in the TCP data acquisition pipeline 650 (Supplemental Fig. S6A). First, an in-person session that consisted of demographic and health surveys, 651 the Structured Clinical Interview for DSM -5 (SCID-V-RV104), a battery of clinician -administered scales, 652 and self -report scales. Second, an in -person session that included MRI scanning ( see MRI data 653 acquisition section below for further neuroimaging details) as well as further self-report scales. Third, an 654 online session of supplemental self-report scales as well as the TestMyBrain suite117. Scales and surveys 655 included a variety of cognitive, behavioral, and clinically -relevant measures and will be collectively 656 referred to as “ behavioral measures” in the present study ( see Dimensional symptom analysis section 657 below for further details). 658 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 26 659 Based on the SCID-V-RV, eleven primary psychiatric diagnosis groups were present (Supplemental Fig. 660 S6) including attention-deficit/hyperactivity disorder (ADHD), anxiety, bipolar disorder (BPD), depression, 661 eating disorder, obsessive compulsive disorder (OCD), panic disorder/phobia, posttraumatic stress 662 disorder (PTSD), schizophrenia/schizoaffective, substance use disorder, and other . Select diagnostic 663 groups were concatenated based on central symptomatology, as follows: anxiety refers to any anxiety -664 central diagnosis such as generalized anxiety disorder, social anxiety, and anxiety not otherwise specified 665 (NOS); BPD included BPD I, BPD II, and cyclothymia; depression included major depressive disorder, 666 dysthymia, and depression NOS; SUD included addiction to any substance and/or alcohol (e.g., cocaine 667 use disorder and alcohol use disorder combined into one SUD category); and other refers to rare 668 diagnoses such as premenstrual dysphoric disorder and psychosis NOS. Note that in downstream 669 analyses, there were 12 labels for primary diagnosis; the 12th label was “no diagnosis”, referring to the 670 healthy comparison participants without psychiatric diagnoses based on the SCID-V-RV. 671 672 Data splitting scheme 673 It was possible that in the final analysis testing the extent that functional brain network dynamics can 674 classify symptom structure (described in following secti ons) there would be leakage of information 675 between training and test participants 149,150. This was because prior analytic steps required either 676 supervised learning as well (as in random forest imputation; see Dimensional symptom analysis section 677 below) or hyperparameter optimization with cross validation (as in nonnegative matrix factorization; see 678 Brain network dynamics analysis section below). This potential for leakage between participants has 679 been found to artificially inflate predicti on results , which can be problematic for interpreting 680 neurobiological links with behavior. Note that data collection site, familial relations, and age were not 681 found to be problematic with respect to data leakage impacting modeling results. 682 683 To mitigate leakage across participants, we implemented a data splitting scheme at the outset of the 684 present study (Supplemental Fig. S6B-C). Of the N=219 total participants, approximately 80% (n=176) 685 were allocated to a discovery dataset and approximately 20% (n=43) t o a held -out validation dataset, 686 which was left untouched until the final analyses. The discovery set was randomly split further for various 687 analysis steps prior to classification into training and test sets. To ensure that the underlying structure of 688 the total dataset was captured by each of the discovery and validation sets, two percentage distributions 689 were constrained: (1) proportion of participants in each primary diagnosis category, and (2) proportion of 690 participants from each collection site ( Supplemental Fig. S6C); otherwise, participants were allocated 691 randomly. 692 693 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 27 MRI data acquisition 694 All MRI data were collected at either the Yale University FAS Brain Imaging Center or the McLean Hospital 695 Brain Imaging Center. The best practices put forth by the Human Connectome Project (HCP)152,153 were 696 followed as closely as possible in both MRI acquisition protocols and data processing. Both scanners 697 were Siemens Magnetom 3T Prisma models with 64 -channel head coils. Whole -brain and multi -echo 698 MPRAGE sequences were used to collect T1w anatomical data with the following specifications: 699 repetition time (TR)=2.2 s; echo times (TE)=1.5, 3.4, 5.2, and 7.0 ms (root mean square of each echo 700 used to compute a single image); flip angle=7°; inversion time=1 .1 s; sagittal orientation for phase 701 encoding=anterior (A) to posterior (P); slice thickness=1.2 mm; total slices acquired=144; resolution=1.2 702 mm3. Specifications unique to T2w anatomical images were: TR=2.8 s; TE=326 ms. Whole -brain, 703 multiband, and echopl anar (EPI) functional MRI data were acquired with the following specifications: 704 TR=0.8 s; TE=37 ms; flip angle=52°; voxel size resolution=2 mm3; multiband acceleration factor=8. There 705 were three cognitive states assessed during scanning plus variants based on AP/PA phase encoding 706 direction, totaling 7 fMRI runs altogether: (1) 2 resting state fMRI runs in both AP and PA (4 total); (2) 707 Stroop inhibitory control task 95,96 fMRI in both AP and PA ( Supplemental Fig. S 6D) (2 total); and (3) 708 Emotional Faces task 125 fMRI in the AP direction ( Supplemental Fig. S 6E). There were 488 volumes 709 (TRs) collected for each resting-state fMRI run; 510 TRs for each Stroop task fMRI run; and 493 TRs for 710 Emotional Faces task fMRI. See [110] for further details on TCP MRI data acquisition protocols as well as 711 full details on paradigms used during task fMRI ( Supplemental Fig. S6D-E). As previously noted, only 712 six of the seven TCP fMRI runs were included in the present study to maximize available data for N=219 713 participants: resting-state run two, PA, was excluded, leaving a balanced total of three resting- and three 714 task-state fMRI runs for downstream analyses. 715 716 MRI data processing 717 The HCP minimal processing pipelines152, version 4.7.0 were applied to neuroimaging data. HCP best 718 practices, technical specif ications, and quality assurance analyses are detailed extensively 719 elsewhere110,152, but briefly we implemented: anatomic reconstruction and segmentation; motion 720 correction; EPI reconstruction, segmentation, and spatial normalization to a standard template; intensity 721 normalization; multimodal surface matching (“MSM -All”) registration 154; single -run ICA -FIX 722 denoising155,156; and de-drift and resampling. Resulting data were in CIFTI 91k-vertex grayordinate space, 723 which were then parcellated (i.e., average time -series of enclosed vertices) into 400 cortical regions 106, 724 32 subcortical regions107, and two cerebellar regions108, totaling 434 whole-brain regions. Note that in all 725 figures and text, the term region refers to either a functionally parcellated area of cortex or an anatomical 726 subdivision of subcortex or cerebellum. The term parcel or node is sometimes used for cortical brain 727 areas, but we adopted the general term region for subdivisions of both cortex and non -cortex for 728 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 28 simplicity. In both cortex and non-cortex, a brain network refers to a functionally- or anatomically-linked 729 collection of regions (see Fig. 2 in Results for parcellated regions projected onto a standard brain surface 730 and volume for cortex and non-cortex, respectively; color-coded by network assignment). 731 732 Functional connectivity estimation and functional network partition 733 Functional connectivity (FC) was estimated for each participant and functional scan via product-moment 734 correlation between each pair of regions’ minimally processed and denoised timeseries. The se 434 735 region by 434 region FC matrices were further partitioned into large-scale functional networks based on 736 a widely -used 17 cortical network solution 106,124, three subcortical networks 107, and one cerebellar 737 network108; totaling 21 whole-brain functional networks (see Fig. 5 in Results). Fisher’s Z-transformation 738 was used whenever FC estimates were averaged. 739 740 Brain network dynamics analysis 741 Following prior work86,88,90-92, we implemented an unsupervised machine learning tool called nonnegative 742 matrix factorization (NMF) to uncover hidden constraints upon functional brain network dynamics. NMF 743 is known for being a parts-based decomposition of multi-dimensional input data (matrices or tensors) into 744 two lower ranking matrices; one with features that approximate a basis set (also termed motifs, weights, 745 or subgraphs), and another being an encoding matrix o f coefficients (also termed expressions or 746 expression coefficients). Standardly, all matrices and/or tensors are non -negative. NMF -uncovered 747 features are optimized for a linear approximation of the original input data and are not constrained by 748 independence or orthogonality (as in principal components analysis), which is likely more 749 neurobiologically appropriate because flexibly co -occurring and/or overlapping network structures are 750 permitted. 751 752 In the present application, FC estimates (upper triangle and p ositively-thresholded values) for each 753 participant and fMRI scan were stacked vertically (as in [86]), and this network configuration matrix was 754 used as the input to NMF. In this usage, across -state reconfiguration dynamics are captured by the 755 presence of both resting-state and task-state FC estimates. Here, NMF is thought to uncover features of 756 functional brain network interactions that constrain dynamic transitions across varied cognitive states. 757 While unsupervised, NMF requires three parameters to be optimized: k, or the number of feature 758 matrices; alpha, or the multiplicative regularization term for factorized matrices; and beta, which is a loss 759 parameter based on minimizing divergence. We cross -validated the discovery dataset by splitting 760 participants randomly and evenly into training and test sets over 100 folds (note that this was limited by 761 computational feasibility) to tune these parameters, which were then applied to NMF of the full discovery 762 dataset as well as the validation dataset. Possible values of k ranged from two through 20 (steps of one); 763 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 29 alpha ranged from 0.1 to 2.0 (steps of 0.1); and beta ranged from 0.2 to 2.0 (steps of 0.2). Accuracy of 764 the cross validation was assessed via reconstruction error 86,157 as well as the nonparametric Mantel 765 correlation158 with the actu al FC configuration matrix. To implement NMF, we used the Python toolkit 766 provided by scikit -learn (version 1.6.1) with the multiplicative update solver, and all other parameters 767 were set to their defaults. Accuracy was stable with the following parameter v alues: k = 5, alpha = 0.2, 768 beta = 1.2 (see Results Fig. 2 for NMF pipeline schematic; Supplemental Fig. S1). 769 770 Dimensional symptom analysis 771 The TCP dataset consists of a rich variety of 163 behavioral , cognitive, and clinical measures (here, 772 collectively referred to as behavioral measures for simplicity; see Supplemental Table S1 table for full 773 list of all clinical, self -report surveys, and cognitive batteries used herein), which we used to uncover 774 dimensionally-organized symptom structures75-81,111, that we termed symptom fingerprints or profiles. 775 776 First, we curated the data to exclude redundancies. One type of redundancy was the presence of a 777 summary score and subscale scores; here we prioritized subscale scores to cover the most amount of 778 behavioral variance unless the distribution of subscale scor es were highly inflated at a single point or 779 otherwise highly non-normal. Another type of redundancy was the presence of both a raw score and a t-780 stat; here we followed prior literature for content -specific best practices. Then, select variables were 781 excluded for theoretical reasons, such as the childhood trauma questionnaire159, which is an assessment 782 of predisposing psychiatric risk factors based on traumatic experiences in childhood, while all other 783 measures were re ferential to current state or recent past (i.e., adulthood). The final set of measures 784 totaled 110 clinical, behavioral, and cognitive variables. 785 786 After curation, the distribution of each behavioral measure was assessed for Gaussianity with four 787 metrics: D’Agostino’s K-squared test, the Shapiro-Wilk test, the Anderson-Darling test, and Kolmogorov-788 Smirnov test. If a given measure exhibited a non -normal distribution based on more than two of these 789 metrics, it was submitted to the bestNormalize package (version 1.9.1) in R160, which samples a suite of 790 transformations and suggests the most consistent and accurate function for normalizing the given data 791 (see Supplemental Fig. S3 for representative examples ). After this initial transformation, the four 792 Gaussianity metrics were implemented again to identify measures that were still highly non-normal. Here, 793 the remaining variables were all zero-inflated, which we opted to binarize for interpretability. 794 795 Following normalizing and binarizing trans formations, select behavioral measures were unavailable in 796 some participants. To account for this, we used random forest imputation (also termed “miss forest”) 161 797 based on prior evidence that it is robustly applicable to nonparametric and/or mixed data types. Random 798 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 30 forest imputation is a supervised learning algorithm that we performed on the discovery dataset; where 799 80% of participants were randomly allocated to the training set and 20% to the testing set. In order to 800 ensure that the variety of native score ranges did not impact downstream analyses, resulting data (i.e., 801 behavioral measures that were transformed and imputed) were scaled to the same space with min-max 802 normalization between the values of zero and one. This approach is widely used in machine learning and 803 has the benefit of recapitulating the original shape of the distribution, just shifted between a fixed minimum 804 and maximum value. 805 806 Following prior work, we applied agglomerative hierarchical clustering with Ward’s linkage distance162 to 807 individual differences correlations (here, Spearman’s rho) of all pairs of behavioral measures. An 808 individual differences correlation estimates the extent that scores on a given pair of behavioral measures 809 covaries across individuals, or how well individual differences are matched in each measure-to-measure 810 pair. If individual differences are highly correlated for a given pair of measures, then they are capturing a 811 similar dimension of functioning and will be allocated to the same cluster. Throughout the present study, 812 the term dimensionally-organized refers to this clusterization. The optimal number of clusters (here: 4 813 clusters) was found with the R package NbClust (version 3.0.1)163, which uses 30 performance metrics 814 to determine the best clustering solution. Then, we implemented principal components analysis (PCA) 815 on each of the four clusters of variables (i.e., the original scores, not correlation v alues), and used the 816 first PC for downstream analyses. This approach has the benefit of similar amounts of variance explained 817 being represented for the first PC of each cluster (see Supplemental Fig. S4 and Table S1), as well as 818 including each behavioral measure in only one cluster to prevent overfitting to any given measure (which 819 may happen if PCA were performed directly without the hierarchical clustering step). Per participant, we 820 used an output from scikit-learn called score_samples to estimate the extent (given by log-likelihood) that 821 each participant expressed a given cluster. Following prior work 80,112,113, clusters were named for 822 interpretability, however we caution against overly interpreting this naming system (as well as priorly used 823 naming systems). To partially account for potential researcher bias in naming clusters, one researcher 824 pre-labeled each of the 110 behavioral measures with broad dimensions of cognitive or behavioral 825 functioning (referred to as dimensions of functioning in the main text) , which was corroborated by two 826 other researchers (Supplemental Table S1). Then, following clustering and PCA, the percentages of 827 each label in each cluster – weighted by normalized factor loadings – were summarized programmatically. 828 This guided naming the four clusters as follows: internalizing, externalizing, cognition, and social/reward. 829 830 Lastly, individual cluster expression scores were converted into categorical patterns or profiles, termed 831 symptom fingerprints. For each participant, we performed one sample t -tests to test the null hypothesis 832 that a given cluster expression score was equal to all other participants’ scores (alternative: participant’s 833 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 31 score was greater than all other scores). We corrected for multiple comparisons with the false discovery 834 rate164. A pattern of zeros and ones were assigned for significant and nonsignificant cluster expression 835 (per participant), respectively. This binarization pattern allowed for 16 possible combinations (o r 836 fingerprints) of cluster expressions ( see Results Fig. 3) which were used as 16 possible symptom 837 fingerprint labels in downstream analyses. 838 839 Classification analysis 840 In order to test the extent that functional brain network dynamics link with dimensional symptom structures 841 across a transdiagnostic and non -diagnosed sample, we used nonlinear support vector classification 842 (SVC) by implementing the Python toolkit NuSVC provided by scikit-learn123. NMF-uncovered subgraph 843 expression coefficients across each of the six states (per participant) were used as the predictors and 844 symptom fingerprints (16 possible), primary diagnosis (12 possible), and case-control (2 possible) labels 845 were used as to -be-classified labels in three separate models, which were later compared. Over 1000 846 permutations, 80% of discovery set participants were randomly allocated for training, and the held -out 847 validation set was used for testing. Accuracy was based on an average of these permutations. Each 848 model had a different theoretical chance level: fingerprints: 6.25%; primary diagnosis: 8.33%; case -849 control: 50%. These were each empirically corroborated by no nparametric permutation testing, where 850 1000 null models were built on randomly shuffling participant labels165. Note that empirical chance closely 851 matched theoretical chance in each model (see Results for full reporting). Statist ical testing was 852 performed on accuracy values with one sample t -tests against empirical chance, and Cohen’s D effect 853 sizes were used to compare the three models. An added benefit of the fingerprinting approach was that 854 it allowed us to develop a nuanced accuracy metric, which we termed “fuzzy” accuracy (and henceforth 855 referred to the original as “strict” accuracy). Fuzzy accuracy accounted for overlap in fingerprint patterns, 856 for example: label one was based on significant expression of all four clusters (1 , 1, 1, 1) and label two 857 on three of the four clusters (1, 1, 1, 0); a misclassification between labels one and two was considered 858 75% correct in terms of fuzzy accuracy and 0% correct in terms of strict accuracy. 859 860 Model comparison 861 We performed two types of model comparison where connectivity patterns inputted to NMF were 862 modified, NMF was re-implemented, and classification of case/control status, primary diagnosis group, 863 and symptom fingerprints were re -assessed based on the modifie d model of brain network dynamics . 864 Following this, effect size (Cohen’s D) of cross -participant model accuracies (in all models, accuracy 865 given by averaging 1000 permut ations) were compared. First, we used this approach to simulate 866 lesioning of functional network brain regions and compare the relative importance of brain systems to 867 downstream classification. Each set of regions was withheld from NMF one network at a time at a time 868 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 32 (i.e., models were iterated over 22 networks). Model performance before and a fter lesioning was 869 compared and any statistically significant reduction in model accuracy was interpreted as that functional 870 system being important for separating class labels. Second, we used a similar approach to compare the 871 relative importance of connec tivity estimates yielded by different sets of cognitive states (but here, all 872 brain regions were included in all models) . The following models were tested: (1) all states: the original 873 or “full” reference model with FC inputted to NMF from all six cognitive states; (2) rest only: NMF 874 uncovered dynamic shifts in connectivity patterns across resting-state (i.e., intrinsic processing) only; (3) 875 task only: shifts across inhibitory cognitive control (Stroop) to emotional faces task states; (4) rest & 876 Stroop: shifts from intrinsic states to cognitive control state s; (5) rest & emotional faces: shifts from 877 intrinsic states to emotional-faces recognition states; (6) Stroop only: shifts across cognitive control task 878 states only; and (7) emotional faces only: shifts across emotional faces task states only. Models 2, 6, and 879 7 were not expected to include large shifts in information processing or cognitive context (i.e., underlying 880 dynamics are likely stable), whereas models 1, 3, 4, and 5 parameterize d varied sh ifts in cognitive 881 context. 882 883 Data availability 884 All analysis code is openly available here: https://github.com/HolmesLab/ClinicalNetDynamics, which 885 was primarily written in Python version 3.12.1. TCP unprocessed and processed neuroimaging data as 886 well as all behavioral measures are openly available110 here: https://nda.nih.gov/study.html?id=2932 and 887 here: https://openneuro.org/datasets/ds005237. 888 889

890 We would like to sincerely thank all of the participants for their significant contribution to this research . 891 This work was supported by the National Institute of Mental Health (R01MH123245 to AJH and 892 R01MH120080 to AJH). All conclusions, inferences, opinions, findings, suggestions, and 893 recommendations expressed or otherwise presented in this manuscript are those of the authors and do 894 not reflect the views of the funding bodies. The authors acknowledge the Office of Advanced Research 895 Computing at Rutgers University as well as the Yale Center for Research Computing at Yale University 896 for providing access to high p erformance compute clusters and associated resources essential to this 897 research. 898 899 900 901 902 903 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 33

904 1 McKenna, T. M., McMullen, T. A. & Shlesinger, M. F. The brain as a dynamic physical system. 905 Neuroscience 60 (1994). https://doi.org/10.1016/0306-4522(94)90489-8 906 2 Vogels, T. P. et al. NEURAL NETWORK DYNAMICS. Annual Review of Neuroscience 28 907 (2005). https://doi.org/10.1146/annurev.neuro.28.061604.135637 908 3 Calhoun, V. D., Miller, R., Pearlson, G. & Adalı, T. The Chronnectome: Time-Varying 909 Connectivity Networks as the Next Frontier in fMRI Data Discovery. Neuron 84 (2014). 910 https://doi.org/10.1016/j.neuron.2014.10.015 911 4 Durstewitz, D., Huys, Q. J. M. & Koppe, G. Psychiatric Illnesses as Disorders of Network 912 Dynamics. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 6 (2021). 913 https://doi.org/10.1016/j.bpsc.2020.01.001 914 5 Braver, T. S., Paxton, J. L., Locke, H. S. & Barch, D. M. Flexible neural mechanisms of cognitive 915 control within human prefrontal cortex. Proceedings of the National Academy of Sciences 106 916 (2009). https://doi.org/10.1073/pnas.0808187106 917 6 Chiong, W. et al. The salience network causally influences default mode network activity during 918 moral reasoning. Brain 136 (2013). https://doi.org/10.1093/brain/awt066 919 7 Safron, A., Klimaj, V. & Hipólito, I. On the Importance of Being Flexible: Dynamic Brain Networks 920 and Their Potential Functional Significances. Frontiers in Systems Neuroscience 15 (2022). 921 https://doi.org/10.3389/fnsys.2021.688424 922 8 Stokes, Mark G. et al. Dynamic Coding for Cognitive Control in Prefrontal Cortex. Neuron 78 923 (2013). https://doi.org/10.1016/j.neuron.2013.01.039 924 9 Ito, T., Yang, G. R., Laurent, P., Schultz, D. H. & Cole, M. W. Constructing neural network 925 models from brain data reveals representational transformations linked to adaptive behavior. 926 Nature Communications 2022 13:1 13 (2022). https://doi.org/10.1038/s41467-022-28323-7 927 10 Muhle-Karbe, P. S., Jiang, J. & Egner, T. Causal Evidence for Learning-Dependent Frontal Lobe 928 Contributions to Cognitive Control. Journal of Neuroscience 38 (2018). 929 https://doi.org/10.1523/JNEUROSCI.1467-17.2017 930 11 Ma, L., Hyman, J. M., Durstewitz, D., Phillips, A. G. & Seamans, J. K. A Quantitative Analysis of 931 Context-Dependent Remapping of Medial Frontal Cortex Neurons and Ensembles. Journal of 932 Neuroscience 36 (2016). https://doi.org/10.1523/JNEUROSCI.3176-15.2016 933 12 van den Brink, R. L. et al. Flexible sensory-motor mapping rules manifest in correlated variability 934 of stimulus and action codes across the brain. Neuron 111 (2023). 935 https://doi.org/10.1016/j.neuron.2022.11.009 936 13 Chiang, F.-K., Wallis, J. D. & Rich, E. L. Cognitive strategies shift information from single 937 neurons to populations in prefrontal cortex. Neuron 110 (2022). 938 https://doi.org/10.1016/j.neuron.2021.11.021 939 14 Yeo, B. T. T. et al. Functional Specialization and Flexibility in Human Association Cortex. 940 Cerebral Cortex 25 (2015). https://doi.org/10.1093/cercor/bhu217 941 15 Cohen, J. R. & D'Esposito, M. The Segregation and Integration of Distinct Brain Networks and 942 Their Relationship to Cognition. Journal of Neuroscience 36 (2016). 943 https://doi.org/10.1523/JNEUROSCI.2965-15.2016 944 16 Yin, D. & Kaiser, M. Understanding neural flexibility from a multifaceted definition. NeuroImage 945 235 (2021). https://doi.org/10.1016/j.neuroimage.2021.118027 946 17 Ueltzhöffer, K., Armbruster-Genç, D. J. N. & Fiebach, C. J. Stochastic Dynamics Underlying 947 Cognitive Stability and Flexibility. PLOS Computational Biology 11 (2015). 948 https://doi.org/10.1371/journal.pcbi.1004331 949 18 Fox, M. D. et al. The human brain is intrinsically organized into dynamic, anticorrelated 950 functional networks. Proceedings of the National Academy of Sciences 102 (2005). 951 https://doi.org/10.1073/pnas.0504136102 952 19 Bassett, D. S. & Bullmore, E. Small-World Brain Networks. The Neuroscientist 12 (2006). 953 https://doi.org/10.1177/1073858406293182 954 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 34 20 Muldoon, S. F., Bridgeford, E. W. & Bassett, D. S. Small-World Propensity and Weighted Brain 955 Networks. Scientific Reports 6 (2016). https://doi.org/10.1038/srep22057 956 21 Cocuzza, C. V., Ito, T., Schultz, D., Bassett, D. S. & Cole, M. W. Flexible Coordinator and 957 Switcher Hubs for Adaptive Task Control. Journal of Neuroscience 40 (2020). 958 https://doi.org/10.1523/JNEUROSCI.2559-19.2020 959 22 Cole, M. W., Pathak, S. & Schneider, W. Identifying the brain's most globally connected regions. 960 NeuroImage 49 (2010). https://doi.org/10.1016/j.neuroimage.2009.11.001 961 23 Sepulcre, J., Sabuncu, M. R., Yeo, T. B., Liu, H. & Johnson, K. A. Stepwise Connectivity of the 962 Modal Cortex Reveals the Multimodal Organization of the Human Brain. Journal of 963 Neuroscience 32 (2012). https://doi.org/10.1523/JNEUROSCI.0759-12.2012 964 24 Margulies, D. S. et al. Situating the default-mode network along a principal gradient of 965 macroscale cortical organization. Proceedings of the National Academy of Sciences 113 (2016). 966 https://doi.org/10.1073/pnas.1608282113 967 25 Raut, R. V., Snyder, A. Z. & Raichle, M. E. Hierarchical dynamics as a macroscopic organizing 968 principle of the human brain. Proceedings of the National Academy of Sciences 117 (2020). 969 https://doi.org/10.1073/pnas.2003383117 970 26 Allen, E. A. et al. Tracking Whole-Brain Connectivity Dynamics in the Resting State. Cerebral 971 Cortex 24 (2014). https://doi.org/10.1093/cercor/bhs352 972 27 Pedersen, M. et al. Multilayer network switching rate predicts brain performance. Proceedings of 973 the National Academy of Sciences 115 (2018). https://doi.org/10.1073/pnas.1814785115 974 28 Cookson, S. L. & D'Esposito, M. Evaluating the reliability, validity, and utility of overlapping 975 networks: Implications for network theories of cognition. Human Brain Mapping 44 (2023). 976 https://doi.org/10.1002/hbm.26134 977 29 Dworetsky, A. et al. Probabilistic mapping of human functional brain networks identifies regions 978 of high group consensus. NeuroImage 237 (2021). 979 https://doi.org/10.1016/j.neuroimage.2021.118164 980 30 Zhang, W., Tang, F., Zhou, X. & Li, H. Dynamic Reconfiguration of Functional Topology in 981 Human Brain Networks: From Resting to Task States. Neural Plasticity 2020 (2020). 982 https://doi.org/10.1155/2020/8837615 983 31 Cole, Michael W., Bassett, Danielle S., Power, Jonathan D., Braver, Todd S. & Petersen, 984 Steven E. Intrinsic and Task-Evoked Network Architectures of the Human Brain. Neuron 83 985 (2014). https://doi.org/10.1016/j.neuron.2014.05.014 986 32 Krienen, F. M., Yeo, B. T. T. & Buckner, R. L. Reconfigurable task-dependent functional coupling 987 modes cluster around a core functional architecture. Philosophical Transactions of the Royal 988 Society B: Biological Sciences (2014). https://doi.org/10.1098/rstb.2013.0526 989 33 Gratton, C. et al. Functional Brain Networks Are Dominated by Stable Group and Individual 990 Factors, Not Cognitive or Daily Variation. Neuron 98 (2018). 991 https://doi.org/10.1016/j.neuron.2018.03.035 992 34 Arbabshirani, M. R., Havlicek, M., Kiehl, K. A., Pearlson, G. D. & Calhoun, V. D. Functional 993 network connectivity during rest and task conditions: A comparative study. Human Brain 994 Mapping 34 (2013). https://doi.org/10.1002/hbm.22118 995 35 Cole, M. W., Ito, T., Cocuzza, C. V. & Sanchez-Romero, R. The Functional Relevance of Task-996 State Functional Connectivity. Journal of Neuroscience 41 (2021). 997 https://doi.org/10.1523/JNEUROSCI.1713-20.2021 998 36 Bassett, D. S. et al. Dynamic reconfiguration of human brain networks during learning. 999 Proceedings of the National Academy of Sciences 108 (2011). 1000 https://doi.org/10.1073/pnas.1018985108 1001 37 Uddin, L. Q., Supekar, K. S., Ryali, S. & Menon, V. Dynamic Reconfiguration of Structural and 1002 Functional Connectivity Across Core Neurocognitive Brain Networks with Development. Journal 1003 of Neuroscience 31 (2011). https://doi.org/10.1523/JNEUROSCI.4465-11.2011 1004 38 Braun, U. et al. Dynamic reconfiguration of frontal brain networks during executive cognition in 1005 humans. Proceedings of the National Academy of Sciences 112 (2015). 1006 https://doi.org/10.1073/pnas.1422487112 1007 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 35 39 Hearne, L. J., Cocchi, L., Zalesky, A. & Mattingley, J. B. Reconfiguration of Brain Network 1008 Architectures between Resting-State and Complexity-Dependent Cognitive Reasoning. Journal 1009 of Neuroscience 37 (2017). https://doi.org/10.1523/JNEUROSCI.0485-17.2017 1010 40 Gonzalez-Castillo, J. et al. Tracking ongoing cognition in individuals using brief, whole-brain 1011 functional connectivity patterns. Proceedings of the National Academy of Sciences 112 (2015). 1012 https://doi.org/10.1073/pnas.1501242112 1013 41 Ajmera, S., Jain, H., Sundaresan, M. & Sridharan, D. Decoding Task-Specific Cognitive States 1014 with Slow, Directed Functional Networks in the Human Brain. eNeuro 7 (2020). 1015 https://doi.org/10.1523/ENEURO.0512-19.2019 1016 42 Mill, R. D. et al. Network modeling of dynamic brain interactions predicts emergence of neural 1017 information that supports human cognitive behavior. PLOS Biology 20 (2022). 1018 https://doi.org/10.1371/journal.pbio.3001686 1019 43 Ikeda, S., Kawano, K., Watanabe, S., Yamashita, O. & Kawahara, Y. Predicting behavior 1020 through dynamic modes in resting-state fMRI data. NeuroImage 247 (2022). 1021 https://doi.org/10.1016/j.neuroimage.2021.118801 1022 44 MacDowell, C. J., Briones, B. A., Lenzi, M. J., Gustison, M. L. & Buschman, T. J. Differences in 1023 the expression of cortex-wide neural dynamics are related to behavioral phenotype. Current 1024 Biology 34 (2024). https://doi.org/10.1016/j.cub.2024.02.004 1025 45 Cole, M. W., Repovš, G. & Anticevic, A. The Frontoparietal Control System. The Neuroscientist 1026 20 (2014). https://doi.org/10.1177/1073858414525995 1027 46 Goschke, T. Dysfunctions of decision-making and cognitive control as transdiagnostic 1028 mechanisms of mental disorders: advances, gaps, and needs in current research. International 1029 Journal of Methods in Psychiatric Research 23 (2014). https://doi.org/10.1002/mpr.1410 1030 47 McTeague, L. M., Goodkind, M. S. & Etkin, A. Transdiagnostic impairment of cognitive control in 1031 mental illness. Journal of Psychiatric Research C (2016). 1032 https://doi.org/10.1016/j.jpsychires.2016.08.001 1033 48 Uddin, L. Q. Cognitive and behavioural flexibility: neural mechanisms and clinical 1034 considerations. Nature Reviews Neuroscience 2021 22:3 22 (2021). 1035 https://doi.org/10.1038/s41583-021-00428-w 1036 49 Nobukawa, S. & Takahashi, T. Perspectives in brain-network dynamics in computational 1037 psychiatry. Frontiers in Computational Neuroscience 17 (2023). 1038 https://doi.org/10.3389/fncom.2023.1290089 1039 50 Uhlhaas, P. J., Haenschel, C., Nikolić, D. & Singer, W. The Role of Oscillations and Synchrony 1040 in Cortical Networks and Their Putative Relevance for the Pathophysiology of Schizophrenia. 1041 Schizophrenia Bulletin 34 (2008). https://doi.org/10.1093/schbul/sbn062 1042 51 Voytek, B. & Knight, R. T. Dynamic Network Communication as a Unifying Neural Basis for 1043 Cognition, Development, Aging, and Disease. Biological Psychiatry 77 (2015). 1044 https://doi.org/10.1016/j.biopsych.2015.04.016 1045 52 Reinen, J. M. et al. The human cortex possesses a reconfigurable dynamic network architecture 1046 that is disrupted in psychosis. Nature Communications 2018 9:1 9 (2018). 1047 https://doi.org/10.1038/s41467-018-03462-y 1048 53 Damaraju, E. et al. Dynamic functional connectivity analysis reveals transient states of 1049 dysconnectivity in schizophrenia. NeuroImage: Clinical 5 (2014). 1050 https://doi.org/10.1016/j.nicl.2014.07.003 1051 54 Yang, H. et al. Reproducible coactivation patterns of functional brain networks reveal the 1052 aberrant dynamic state transition in schizophrenia. NeuroImage 237 (2021). 1053 https://doi.org/10.1016/j.neuroimage.2021.118193 1054 55 You, W. et al. Impaired dynamic functional brain properties and their relationship to symptoms in 1055 never treated first-episode patients with schizophrenia. Schizophrenia 2022 8:1 8 (2022). 1056 https://doi.org/10.1038/s41537-022-00299-9 1057 56 Blair, D. S. et al. A Dynamic Entropy Approach Reveals Reduced Functional Network 1058 Connectivity Trajectory Complexity in Schizophrenia. Entropy 2024, Vol. 26, Page 545 26 1059 (2024). https://doi.org/10.3390/e26070545 1060 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 36 57 Rolls, E. T., Loh, M. & Deco, G. An attractor hypothesis of obsessive–compulsive disorder. 1061 European Journal of Neuroscience 28 (2008). https://doi.org/10.1111/j.1460-9568.2008.06379.x 1062 58 de Lacy, N. & Calhoun, V. D. Dynamic connectivity and the effects of maturation in youth with 1063 attention deficit hyperactivity disorder. Network Neuroscience 3 (2018). 1064 https://doi.org/10.1162/netn_a_00063 1065 59 Northoff, G. & Hirjak, D. Is depression a global brain disorder with topographic dynamic 1066 reorganization? Translational Psychiatry 2024 14:1 14 (2024). https://doi.org/10.1038/s41398-1067 024-02995-9 1068 60 Schultz, D. H. et al. Global connectivity of the fronto-parietal cognitive control network is related 1069 to depression symptoms in the general population. Network Neuroscience 3 (2018). 1070 https://doi.org/10.1162/netn_a_00056 1071 61 Rashid, B. et al. Classification of schizophrenia and bipolar patients using static and dynamic 1072 resting-state fMRI brain connectivity. NeuroImage 134 (2016). 1073 https://doi.org/10.1016/j.neuroimage.2016.04.051 1074 62 Buckholtz, Joshua W. & Meyer-Lindenberg, A. Psychopathology and the Human Connectome: 1075 Toward a Transdiagnostic Model of Risk For Mental Illness. Neuron 74 (2012). 1076 https://doi.org/10.1016/j.neuron.2012.06.002 1077 63 Clementz, B. A. et al. Identification of Distinct Psychosis Biotypes Using Brain-Based 1078 Biomarkers. American Journal of Psychiatry (2015). 1079 https://doi.org/10.1176/appi.ajp.2015.14091200 1080 64 Holmes, A. J. & Patrick, L. M. The Myth of Optimality in Clinical Neuroscience. Trends in 1081 Cognitive Sciences 22 (2018). https://doi.org/10.1016/j.tics.2017.12.006 1082 65 Xia, C. H. et al. Linked dimensions of psychopathology and connectivity in functional brain 1083 networks. Nature Communications 2018 9:1 9 (2018). https://doi.org/10.1038/s41467-018-1084 05317-y 1085 66 van den Heuvel, M. P. & Sporns, O. A cross-disorder connectome landscape of brain 1086 dysconnectivity. Nature Reviews Neuroscience 2019 20:7 20 (2019). 1087 https://doi.org/10.1038/s41583-019-0177-6 1088 67 Feczko, E. et al. The Heterogeneity Problem: Approaches to Identify Psychiatric Subtypes. 1089 Trends in Cognitive Sciences 23 (2019). https://doi.org/10.1016/j.tics.2019.03.009 1090 68 Parkes, L., Satterthwaite, T. D. & Bassett, D. S. Towards precise resting-state fMRI biomarkers 1091 in psychiatry: synthesizing developments in transdiagnostic research, dimensional models of 1092 psychopathology, and normative neurodevelopment. Current Opinion in Neurobiology 65 1093 (2020). https://doi.org/10.1016/j.conb.2020.10.016 1094 69 Vanes, L. D. & Dolan, R. J. Transdiagnostic neuroimaging markers of psychiatric risk: A 1095 narrative review. NeuroImage: Clinical 30 (2021). https://doi.org/10.1016/j.nicl.2021.102634 1096 70 Brandl, F. et al. Common and specific large-scale brain changes in major depressive disorder, 1097 anxiety disorders, and chronic pain: a transdiagnostic multimodal meta-analysis of structural 1098 and functional MRI studies. Neuropsychopharmacology 2022 47:5 47 (2022). 1099 https://doi.org/10.1038/s41386-022-01271-y 1100 71 Achenbach, T. M., Ivanova, M. Y., Rescorla, L. A., Turner, L. V. & Althoff, R. R. 1101 Internalizing/Externalizing Problems: Review and Recommendations for Clinical and Research 1102 Applications. Journal of the American Academy of Child & Adolescent Psychiatry 55 (2016). 1103 https://doi.org/10.1016/j.jaac.2016.05.012 1104 72 Krynicki, C. R., Upthegrove, R., Deakin, J. F. W. & Barnes, T. R. E. The relationship between 1105 negative symptoms and depression in schizophrenia: a systematic review. Acta Psychiatrica 1106 Scandinavica 137 (2018). https://doi.org/10.1111/acps.12873 1107 73 Smucny, J. et al. Levels of Cognitive Control: A Functional Magnetic Resonance Imaging-Based 1108 Test of an RDoC Domain Across Bipolar Disorder and Schizophrenia. 1109 Neuropsychopharmacology 43 (2017). https://doi.org/10.1038/npp.2017.233 1110 74 Grahek, I., Everaert, J., Krebs, R. M. & Koster, E. H. W. Cognitive Control in Depression: 1111 Toward Clinical Models Informed by Cognitive Neuroscience. Clinical Psychological Science 6 1112 (2018). https://doi.org/10.1177/2167702618758969 1113 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 37 75 Cuthbert, B. N. & Insel, T. R. Toward the future of psychiatric diagnosis: the seven pillars of 1114 RDoC. BMC Medicine 2013 11:1 11 (2013). https://doi.org/10.1186/1741-7015-11-126 1115 76 Kotov, R. et al. The hierarchical taxonomy of psychopathology (HiTOP): A dimensional 1116 alternative to traditional nosologies. Journal of abnormal psychology 126 (2017). 1117 https://doi.org/10.1037/abn0000258 1118 77 Conway, C. C. et al. A Hierarchical Taxonomy of Psychopathology Can Transform Mental Health 1119 Research. Perspectives on Psychological Science 14 (2019). 1120 https://doi.org/10.1177/1745691618810696 1121 78 Ruggero, C. et al. Integrating the Hierarchical Taxonomy of Psychopathology (HiTOP) into 1122 clinical practice. Journal of Consulting and Clinical Psychology 87 (2019). 1123 https://doi.org/10.1037/ccp0000452 1124 79 Kotov, R. et al. The Hierarchical Taxonomy of Psychopathology (HiTOP): A Quantitative 1125 Nosology Based on Consensus of Evidence. Annual Review of Clinical Psychology 17 (2021). 1126 https://doi.org/10.1146/annurev-clinpsy-081219-093304 1127 80 Forbes, M. K. et al. Reconstructing Psychopathology: A Data-Driven Reorganization of the 1128 Symptoms in the Diagnostic and Statistical Manual of Mental Disorders. Clinical Psychological 1129 Science (2024). https://doi.org/10.1177/21677026241268345 1130 81 Insel, T. R. The NIMH Research Domain Criteria (RDoC) Project: Precision Medicine for 1131 Psychiatry. American Journal of Psychiatry 171 (2014). 1132 https://doi.org/10.1176/appi.ajp.2014.14020138 1133 82 Tiego, J. et al. Precision behavioral phenotyping as a strategy for uncovering the biological 1134 correlates of psychopathology. Nature Mental Health 2023 1:5 1 (2023). 1135 https://doi.org/10.1038/s44220-023-00057-5 1136 83 Barch, D. M. The Neural Correlates of Transdiagnostic Dimensions of Psychopathology. The 1137 American Journal of Psychiatry 174 (2017). https://doi.org/10.1176/appi.ajp.2017.17030289 1138 84 Fusar-Poli, P. et al. Transdiagnostic psychiatry: a systematic review. World Psychiatry 18 1139 (2019). https://doi.org/10.1002/wps.20631 1140 85 Chopra, S. et al. Generalizable and replicable brain-based predictions of cognitive functioning 1141 across common psychiatric illness. Science Advances 10 (2024). 1142 https://doi.org/10.1126/sciadv.adn1862 1143 86 Khambhati, A. N., Medaglia, J. D., Karuza, E. A., Thompson-Schill, S. L. & Bassett, D. S. 1144 Subgraphs of functional brain networks identify dynamical constraints of cognitive control. PLOS 1145 Computational Biology 14 (2018). https://doi.org/10.1371/journal.pcbi.1006234 1146 87 Khambhati, A. N., Mattar, M. G., Wymbs, N. F., Grafton, S. T. & Bassett, D. S. Beyond 1147 modularity: Fine-scale mechanisms and rules for brain network reconfiguration. NeuroImage 1148 166 (2018). https://doi.org/10.1016/j.neuroimage.2017.11.015 1149 88 Anderson, A. et al. Non-negative matrix factorization of multimodal MRI, fMRI and phenotypic 1150 data reveals differential changes in default mode subnetworks in ADHD. NeuroImage 102 1151 (2014). https://doi.org/10.1016/j.neuroimage.2013.12.015 1152 89 Lee, D. D. & Seung, H. S. Learning the parts of objects by non-negative matrix factorization. 1153 Nature 1999 401:6755 401 (1999). https://doi.org/10.1038/44565 1154 90 Aonishi, T. et al. Imaging data analysis using non-negative matrix factorization. Neuroscience 1155 Research 179 (2022). https://doi.org/10.1016/j.neures.2021.12.001 1156 91 Cheung, V. C. K., Devarajan, K., Severini, G., Turolla, A. & Bonato, P. Decomposing time series 1157 data by a non-negative matrix factorization algorithm with temporally constrained coefficients | 1158 IEEE Conference Publication | IEEE Xplore. 2015 37th Annual International Conference of the 1159 IEEE Engineering in Medicine and Biology Society (EMBC) (2015). 1160 https://doi.org/10.1109/EMBC.2015.7319146 1161 92 Sotiras, A., Resnick, S. M. & Davatzikos, C. Finding imaging patterns of structural covariance 1162 via Non-Negative Matrix Factorization. NeuroImage 108 (2015). 1163 https://doi.org/10.1016/j.neuroimage.2014.11.045 1164 93 Cole, M. W. Control and Connectivity. (2017). 1165 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 38 94 Gratton, G., Cooper, P., Fabiani, M., Carter, C. S. & Karayanidis, F. Dynamics of cognitive 1166 control: Theoretical bases, paradigms, and a view for the future. Psychophysiology 55 (2018). 1167 https://doi.org/10.1111/psyp.13016 1168 95 MacLeod, C. M. Half a century of research on the Stroop effect: An integrative review. 1169 Psychological Bulletin 109, 163-203 (1991). https://doi.org/https://doi.org/10.1037/0033-1170 2909.109.2.163 1171 96 Stroop, J. R. Studies of interference in serial verbal reactions. Journal of Experimental 1172 Psychology 18, 643-662 (1935). https://doi.org/https://doi.org/10.1037/h0054651 1173 97 Khambhati, A. N., Sizemore, A. E., Betzel, R. F. & Bassett, D. S. Modeling and interpreting 1174 mesoscale network dynamics. NeuroImage 180 (2018). 1175 https://doi.org/10.1016/j.neuroimage.2017.06.029 1176 98 Honey, C. J., Kötter, R., Breakspear, M. & Sporns, O. Network structure of cerebral cortex 1177 shapes functional connectivity on multiple time scales. Proceedings of the National Academy of 1178 Sciences 104 (2007). https://doi.org/10.1073/pnas.0701519104 1179 99 Avena-Koenigsberger, A. et al. Communication dynamics in complex brain networks. Nature 1180 Reviews Neuroscience 2017 19:1 19 (2017). https://doi.org/10.1038/nrn.2017.149 1181 100 Peterson, K. L., Sanchez-Romero, R., Mill, R. D. & Cole, M. W. Regularized partial correlation 1182 provides reliable functional connectivity estimates while correcting for widespread confounding. 1183 bioRxiv (2023). https://doi.org/10.1101/2023.09.16.558065 1184 101 Latora, V. & Marchiori, M. Efficient Behavior of Small-World Networks. Physical Review Letters 1185 87 (2001). https://doi.org/10.1103/PhysRevLett.87.198701 1186 102 Achard, S. & Bullmore, E. Efficiency and Cost of Economical Brain Functional Networks. PLOS 1187 Computational Biology 3 (2007). https://doi.org/10.1371/journal.pcbi.0030017 1188 103 Wang, L. et al. Altered brain functional networks in people with Internet gaming disorder: 1189 Evidence from resting-state fMRI. Psychiatry Research: Neuroimaging 254 (2016). 1190 https://doi.org/10.1016/j.pscychresns.2016.07.001 1191 104 First, M. B., Williams, J. B. W., Karg, R. S. & Spitzer, R. L. Structured Clinical Interview for DSM-1192 5 Research Version SCID-5-RV. (American Psychiatric Association, 2015). 1193 105 Yeo, B. T. T. et al. The organization of the human cerebral cortex estimated by intrinsic 1194 functional connectivity. Journal of Neurophysiology 106 (2011). 1195 https://doi.org/10.1152/jn.00338.2011 1196 106 Yan, X. et al. Homotopic local-global parcellation of the human cerebral cortex from resting-state 1197 functional connectivity. NeuroImage 273 (2023). 1198 https://doi.org/10.1016/j.neuroimage.2023.120010 1199 107 Tian, Y. et al. Topographic organization of the human subcortex unveiled with functional 1200 connectivity gradients. Nature Neuroscience 23 (2020). https://doi.org/10.1038/s41593-020-1201 00711-6 1202 108 Buckner, R. L., Krienen, F. M., Castellanos, A., Diaz, J. C. & Yeo, B. T. T. The organization of the 1203 human cerebellum estimated by intrinsic functional connectivity. Journal of Neurophysiology 106 1204 (2011). https://doi.org/10.1152/jn.00339.2011 1205 109 Sha, Z. et al. Meta-Connectomic Analysis Reveals Commonly Disrupted Functional 1206 Architectures in Network Modules and Connectors across Brain Disorders - PubMed. Cerebral 1207 cortex (New York, N.Y. : 1991) 28 (2018). https://doi.org/10.1093/cercor/bhx273 1208 110 Chopra, S. et al. The Transdiagnostic Connectome Project: a richly phenotyped open dataset 1209 for advancing the study of brain-behavior relationships in psychiatry. Scientific Data (in press). 1210 111 Goldberg, L. R. Doing it all Bass-Ackwards: The development of hierarchical factor structures 1211 from the top down. Journal of Research in Personality 40 (2006). 1212 https://doi.org/10.1016/j.jrp.2006.01.001 1213 112 Ringwald, W. R., Forbes, M. K. & Wright, A. G. C. Meta-analysis of structural evidence for the 1214 Hierarchical Taxonomy of Psychopathology (HiTOP) model | Psychological Medicine | 1215 Cambridge Core. Psychological Medicine 53 (2023). 1216 https://doi.org/10.1017/S0033291721001902 1217 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 39 113 Wright, A. G. C. & Simms, L. J. A metastructural model of mental disorders and pathological 1218 personality traits. Psychological Medicine 45 (2015). 1219 https://doi.org/10.1017/S0033291715000252 1220 114 Lovibond, P. F. & Lovibond, S. H. The structure of negative emotional states: Comparison of the 1221 Depression Anxiety Stress Scales (DASS) with the Beck Depression and Anxiety Inventories. 1222 Behaviour Research and Therapy 33 (1995). https://doi.org/10.1016/0005-7967(94)00075-U 1223 115 Costa, P. T. & McCrae, R. R. NEO-FFI NEO Five-Factor Invetory Test Booklet-Form S (Adult). 1224 (Psychological Assessment Resources, Inc., 2003). 1225 116 Young, R. C., Biggs, J. T., Ziegler, V. E. & Meyer, D. A. A Rating Scale for Mania: Reliability, 1226 Validity and Sensitivity. The British Journal of Psychiatry 133 (1978). 1227 https://doi.org/10.1192/bjp.133.5.429 1228 117 Passell, E. et al. Digital Cognitive Assessment: Results from the TestMyBrain NIMH Research 1229 Domain Criteria (RDoC) Field Test Battery Report. PsyArXiv (2019). 1230 https://doi.org/10.31234/osf.io/dcszr 1231 118 Shipley, W. C. Shipley Institute of Living Scale Administration Form. (Western Psychological 1232 Services, 1967). 1233 119 Brennan, K. A., Clark, C. L. & Shaver, P. R. in Attachment theory and close relationships (eds 1234 J. A. Simpson & W. S. Rholes) 46-76 (The Guilford Press, 1998). 1235 120 Cloninger, C. R. A Systematic Method for Clinical Description and Classification of Personality 1236 Variants. Archives of General Psychiatry 44 (1987). 1237 https://doi.org/10.1001/archpsyc.1987.01800180093014 1238 121 Bao, W. et al. Alterations in large-scale functional networks in adult posttraumatic stress 1239 disorder: A systematic review and meta-analysis of resting-state functional connectivity studies. 1240 Neuroscience & Biobehavioral Reviews 131 (2021). 1241 https://doi.org/10.1016/j.neubiorev.2021.10.017 1242 122 Galatzer-Levy, I. R. & Bryant, R. A. 636,120 Ways to Have Posttraumatic Stress Disorder. 1243 Perspectives on Psychological Science 8 (2013). https://doi.org/10.1177/1745691613504115 1244 123 Chang, C.-C. & Lin, C.-J. LIBSVM: A library for support vector machines. ACM Transactions on 1245 Intelligent Systems and Technology (TIST) 2 (2011). https://doi.org/10.1145/1961189.1961199 1246 124 Schaefer, A. et al. Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic 1247 Functional Connectivity MRI. Cerebral Cortex 28 (2018). https://doi.org/10.1093/cercor/bhx179 1248 125 Hariri, A. R., Tessitore, A., Mattay, V. S., Fera, F. & Weinberger, D. R. The Amygdala Response 1249 to Emotional Stimuli: A Comparison of Faces and Scenes. NeuroImage 1 (2002). 1250 https://doi.org/10.1006/nimg.2002.1179 1251 126 Guimerà, R. & Amaral, L. A. N. Cartography of complex networks: modules and universal roles. 1252 Journal of Statistical Mechanics: Theory and Experiment 2005 (2005). 1253 https://doi.org/10.1088/1742-5468/2005/02/P02001 1254 127 Sun, Y., Zhang, M. & Saggar, M. Cross-attractor modeling of resting-state functional connectivity 1255 in psychiatric disorders. NeuroImage 279 (2023). 1256 https://doi.org/10.1016/j.neuroimage.2023.120302 1257 128 Kottaram, A. et al. Brain network dynamics in schizophrenia: Reduced dynamism of the default 1258 mode network. Human Brain Mapping 40 (2019). https://doi.org/10.1002/hbm.24519 1259 129 Lehmann, D. et al. EEG microstate duration and syntax in acute, medication-naïve, first-episode 1260 schizophrenia: a multi-center study. Psychiatry Research: Neuroimaging 138 (2005). 1261 https://doi.org/10.1016/j.pscychresns.2004.05.007 1262 130 Chopra, S. et al. Network-Based Spreading of Gray Matter Changes in Psychosis. JAMA 1263 Psychiatry 80 (2023). https://doi.org/10.1001/jamapsychiatry.2023.3293 1264 131 Noble, S., Curtiss, J., Pessoa, L. & Scheinost, D. The tip of the iceberg: A call to embrace anti-1265 localizationism in human neuroscience research. Imaging Neuroscience 2 (2024). 1266 https://doi.org/10.1162/imag_a_00138 1267 132 Laumann, T. O., Snyder, A. Z. & Gratton, C. Challenges in the measurement and interpretation 1268 of dynamic functional connectivity. Imaging Neuroscience 2 (2024). 1269 https://doi.org/10.1162/imag_a_00366 1270 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 40 133 Scheffer, M. et al. A Dynamical Systems View of Psychiatric Disorders—Theory. JAMA 1271 Psychiatry 81 (2024). https://doi.org/10.1001/jamapsychiatry.2024.0215 1272 134 John, Y. J. et al. It’s about time: Linking dynamical systems with human neuroimaging to 1273 understand the brain. Network Neuroscience 6 (2022). https://doi.org/10.1162/netn_a_00230 1274 135 Galadí, J. A. et al. Capturing the non-stationarity of whole-brain dynamics underlying human 1275 brain states. NeuroImage 244 (2021). https://doi.org/10.1016/j.neuroimage.2021.118551 1276 136 Bullmore, E. & Sporns, O. The economy of brain network organization. Nature Reviews 1277 Neuroscience 2012 13:5 13 (2012). https://doi.org/10.1038/nrn3214 1278 137 Pessoa, L. Understanding brain networks and brain organization. Physics of Life Reviews 11 1279 (2014). https://doi.org/10.1016/j.plrev.2014.03.005 1280 138 Gu, S. et al. Controllability of structural brain networks. Nature Communications 2015 6:1 6 1281 (2015). https://doi.org/10.1038/ncomms9414 1282 139 Lin, P. et al. Dynamic Default Mode Network across Different Brain States. Scientific Reports 1283 2017 7:1 7 (2017). https://doi.org/10.1038/srep46088 1284 140 Bansal, K. et al. Cognitive chimera states in human brain networks. Science Advances 5 (2019). 1285 https://doi.org/10.1126/sciadv.aau8535 1286 141 Lynn, C. W., Cornblath, E. J., Papadopoulos, L., Bertolero, M. A. & Bassett, D. S. Broken 1287 detailed balance and entropy production in the human brain. Proceedings of the National 1288 Academy of Sciences 118 (2021). https://doi.org/10.1073/pnas.2109889118 1289 142 Liégeois, R., Laumann, T. O., Snyder, A. Z., Zhou, J. & Yeo, B. T. T. Interpreting temporal 1290 fluctuations in resting-state functional connectivity MRI. NeuroImage 163 (2017). 1291 https://doi.org/10.1016/j.neuroimage.2017.09.012 1292 143 Makowski, C., Lepage, M. & Evans, A. C. Head motion: the dirty little secret of neuroimaging in 1293 psychiatry. Journal of Psychiatry & Neuroscience : JPN 44 (2018). 1294 https://doi.org/10.1503/jpn.180022 1295 144 Parkes, L., Fulcher, B., Yucel, M. & Fonito, A. An evaluation of the efficacy, reliability, and 1296 sensitivity of motion correction strategies for resting-state functional MRI. NeuroImage 171 1297 (2018). https://doi.org/10.1016/j.neuroimage.2017.12.073 1298 145 Birn, R., JB, D., MA, S. & PA, B. Separating respiratory-variation-related fluctuations from 1299 neuronal-activity-related fluctuations in fMRI - PubMed. NeuroImage 31 (2006). 1300 https://doi.org/10.1016/j.neuroimage.2006.02.048 1301 146 Power, J. D. et al. Distinctions among real and apparent respiratory motions in human fMRI 1302 data. NeuroImage 201 (2019). https://doi.org/10.1016/j.neuroimage.2019.116041 1303 147 Power, J. D., Plitt, M., Laumann, T. O. & Martin, A. Sources and implications of whole-brain fMRI 1304 signals in humans. NeuroImage 146 (2017). https://doi.org/10.1016/j.neuroimage.2016.09.038 1305 148 Gu, Y., Han, F. & Liu, X. Arousal Contributions to Resting-State fMRI Connectivity and 1306 Dynamics. Frontiers in Neuroscience 13 (2019). https://doi.org/10.3389/fnins.2019.01190 1307 149 Rosenblatt, M., Tejavibulya, L., Jiang, R., Noble, S. & Scheinost, D. Data leakage inflates 1308 prediction performance in connectome-based machine learning models. Nature 1309 Communications 2024 15:1 15 (2024). https://doi.org/10.1038/s41467-024-46150-w 1310 150 Kapoor, S. & Narayanan, A. Leakage and the reproducibility crisis in machine-learning-based 1311 science. Patterns 4 (2023). https://doi.org/10.1016/j.patter.2023.100804 1312 151 Spielberg, J. M., Miller, G. A., Heller, W. & Banich, M. T. Flexible brain network reconfiguration 1313 supporting inhibitory control. Proceedings of the National Academy of Sciences 112 (2015). 1314 https://doi.org/10.1073/pnas.1500048112 1315 152 Glasser, M. F. et al. The minimal preprocessing pipelines for the Human Connectome Project. 1316 NeuroImage Complete (2013). https://doi.org/10.1016/j.neuroimage.2013.04.127 1317 153 Van Essen, D. C. et al. The WU-Minn Human Connectome Project: An Overview. NeuroImage 1318 80 (2013). https://doi.org/10.1016/j.neuroimage.2013.05.041 1319 154 Glasser, M. F. et al. A multi-modal parcellation of human cerebral cortex. Nature 536 (2016). 1320 https://doi.org/10.1038/nature18933 1321 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint 41 155 Griffanti, L. et al. ICA-based artefact removal and accelerated fMRI acquisition for improved 1322 resting state network imaging. NeuroImage Complete (2014). 1323 https://doi.org/10.1016/j.neuroimage.2014.03.034 1324 156 Salimi-Khorshidi, G. et al. Automatic denoising of functional MRI data: Combining independent 1325 component analysis and hierarchical fusion of classifiers. NeuroImage 90 (2014). 1326 https://doi.org/10.1016/j.neuroimage.2013.11.046 1327 157 Lin, X. & Boutros, P. C. Optimization and expansion of non-negative matrix factorization. BMC 1328 Bioinformatics 2019 21:1 21 (2020). https://doi.org/10.1186/s12859-019-3312-5 1329 158 Mantel, N. The Detection of Disease Clustering and a Generalized Regression Approach. 1330 Cancer Research 27, 209-220 (1967). 1331 159 Pennebaker, J. W. & Susman, J. R. Disclosure of traumas and psychosomatic processes. Social 1332 Science & Medicine 26 (1988). https://doi.org/10.1016/0277-9536(88)90397-8 1333 160 Peterson, R. Finding Optimal Normalizing Transformations via best Normalize. R Journal 13 1334 (2021). 1335 161 Stekhoven, D. J. & Bühlmann, P. MissForest—non-parametric missing value imputation for 1336 mixed-type data. Bioinformatics 28 (2012). https://doi.org/10.1093/bioinformatics/btr597 1337 162 Ward, J. H. Hierarchical Grouping to Optimize an Objective Function. Journal of the American 1338 Statistical Association (1963). https://doi.org/10.1080/01621459.1963.10500845 1339 163 Charrad, M., Ghazzali, N., Boiteau, V. & Niknafs, A. NbClust: An R Package for Determining the 1340 Relevant Number of Clusters in a Data Set. Journal of Statistical Software 61 (2014). 1341 https://doi.org/10.18637/jss.v061.i06 1342 164 Benjamini, Y. & Hochberg, Y. Controlling the False Discovery Rate: A Practical and Powerful 1343 Approach to Multiple Testing. Journal of the Royal Statistical Society: Series B (Methodological) 1344 57 (1995). https://doi.org/10.1111/j.2517-6161.1995.tb02031.x 1345 165 Nichols, T. E. & Holmes, A. P. Nonparametric permutation tests for functional neuroimaging: A 1346 primer with examples. Human Brain Mapping 15 (2002). https://doi.org/10.1002/hbm.1058 1347 1348 .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint .CC-BY-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 May 28, 2025. ; https://doi.org/10.1101/2025.05.23.655864doi: bioRxiv preprint