Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Mar 27.
Published in final edited form as: Biol Psychiatry. 2022 May 18;92(12):973–983. doi: 10.1016/j.biopsych.2022.05.014

Linking Individual Differences in Personalized Functional Network Topography to Psychopathology in Youth

Zaixu Cui 1, Adam R Pines 1, Bart Larsen 1, Valerie J Sydnor 1, Hongming Li 1, Azeez Adebimpe 1, Aaron F Alexander-Bloch 1, Dani S Bassett 1, Max Bertolero 1, Monica E Calkins 1, Christos Davatzikos 1, Damien A Fair 1, Ruben C Gur 1, Raquel E Gur 1, Tyler M Moore 1, Sheila Shanmugan 1, Russell T Shinohara 1, Jacob W Vogel 1, Cedric H Xia 1, Yong Fan 1, Theodore D Satterthwaite 1
PMCID: PMC10040299  NIHMSID: NIHMS1880507  PMID: 35927072

Abstract

BACKGROUND:

The spatial layout of large-scale functional brain networks differs between individuals and is particularly variable in the association cortex, implicated in a broad range of psychiatric disorders. However, it remains unknown whether this variation in functional topography is related to major dimensions of psychopathology in youth.

METHODS:

The authors studied 790 youths ages 8 to 23 years who had 27 minutes of high-quality functional magnetic resonance imaging data as part of the Philadelphia Neurodevelopmental Cohort. Four correlated dimensions were estimated using a confirmatory correlated traits factor analysis on 112 item-level clinical symptoms, and one overall psychopathology factor with 4 orthogonal dimensions were extracted using a confirmatory factor analysis. Spatially regularized nonnegative matrix factorization was used to identify 17 individual-specific functional networks for each participant. Partial least square regression with split-half cross-validation was conducted to evaluate to what extent the topography of personalized functional networks encodes major dimensions of psychopathology.

RESULTS:

Personalized functional network topography significantly predicted unseen individuals’ major dimensions of psychopathology, including fear, psychosis, externalizing, and anxious-misery. Reduced representation of association networks was among the most important features for the prediction of all 4 dimensions. Further analysis revealed that personalized functional network topography predicted overall psychopathology (r = 0.16, permutation testing p < .001), which drove prediction of the 4 correlated dimensions.

CONCLUSIONS:

These results suggest that individual differences in functional network topography in association networks is related to overall psychopathology in youth. Such results underscore the importance of considering functional neuroanatomy for personalized diagnostics and therapeutics in psychiatry.

The human cerebral cortex is organized into spatially distributed large-scale functional networks that support diverse perceptual, executive, and socioemotional functions (1,2). Recent evidence from multiple independent studies has demonstrated that the spatial layout of these functional networks—their functional topography—varies substantially across individuals, even after accurate alignment to a common structural template (3–11). This interindividual variability in functional topography is maximal in the association cortex, higher-order, phylogenetically expanded areas of the cortex that support integrative, abstract, and advanced mental functions (5–10). The functional topography of the association cortex is refined during development and has been previously associated with individual differences in executive function during childhood, adolescence, and young adulthood (10). However, it remains unknown if individual differences in the functional topography of the association cortex are linked to major dimensions of psychopathology in youth.

While psychiatric illness is typically described according to the DSM-5 (12), categorical clinical diagnoses fail to capture variability in disease severity, have notable heterogeneity within diagnoses, and are marked by a high degree of comorbidity (13,14). Accordingly, efforts such as the Research Domain Criteria initiative and the Hierarchical Taxonomy of Psychopathology frameworks have proposed dimensional models of psychopathology (13,15–17). Dimensional taxonomies describe psychopathology as hierarchically organized, correlated dimensions of symptoms, wherein an individual receives a continuous score on each dimension (13,17). Previous studies conducted in youth samples have often identified major dimensions of psychopathology, including fear, psychosis, externalizing, and anxious-misery dimensions (13,18,19).

Emerging evidence additionally points to the importance of characterizing clinical and neural correlates of a dimensional overall psychopathology factor (also called the p factor) (20,21). The p factor quantifies an individual’s shared vulnerability to a broad range of transdiagnostic psychiatric symptoms and thus accounts for comorbidity among psychiatric disorders (20–22). Higher overall psychopathology scores, above and beyond specific psychopathology dimensions, are linked to earlier onset of psychiatric disorders and greater life impairment (21). The burden of overall psychopathology has been linked to abnormal patterns of functional connectivity between large-scale cortical networks (23,24). Consistent with these findings, our prior work demonstrated that the loss of segregation between the default mode network and the executive network is a common feature across transdiagnostic dimensions (25). However, all of these studies measured individuals’ internetwork functional connectivity using standard group-level atlases (9,26). Convergent evidence has demonstrated that use of group-level atlases for each person ignores the individual variation of functional topography, which may lead differences in spatial distribution to be aliased into interregional functional connectivity, potentially biasing both inference and interpretation (9,26).

At present, it remains unknown whether individual differences in spatial topography of functional networks are associated with major dimensions of psychopathology in youth. To address this gap, we capitalized on recently developed machine learning techniques and a large sample of youths who underwent clinical phenotyping and functional neuroimaging as part of the Philadelphia Neurodevelopmental Cohort (27). We hypothesized that major dimensions of psychopathology would be linked to individual differences in functional topography. Furthermore, we expected that associations with major dimensions of psychopathology would be largely driven by shared deficits in the association cortex linked to transdiagnostic overall psychopathology.

METHODS AND MATERIALS

Participants

A total of 1601 participants were studied as part of the Philadelphia Neurodevelopmental Cohort (27). However, 154 participants were excluded due to medical disorders, and 657 participants were excluded for low quality in T1-weighted or functional images (see Supplemental Methods and Figure S1 in Supplement 1). The final sample included 790 youths aged 8 to 23 years (Table S1 in Supplement 1). All study procedures were approved by the Institutional Review Boards of both the University of Pennsylvania and the Children’s Hospital of Philadelphia.

Clinical Assessments

As described previously (18,19,28), psychopathology symptoms were evaluated using a structured screening interview (GOASSESS) that included 112 items (see Supplemental Data in Supplement 2 and Table S1 in Supplement 1). An exploratory factor analysis (19,28–30) of 112 item-level symptoms identified 4 correlated dimensions of psychopathology: fear, psychosis, externalizing, and anxious-misery. Based on the acquired factor configuration, we applied a confirmatory correlated traits model to generate each participant’s factor scores on the 4 correlated dimensions. Given the high correlation among these 4 dimensions, a confirmatory bifactor analysis (19,28–30) was conducted to generate 5 orthogonal dimensions of psychopathology: fear, psychosis, externalizing, anxious-misery, and overall psychopathology (p factor), the latter of which describes a shared vulnerability to a broad range of symptoms across mental disorders. See Supplemental Data in Supplement 2 for factor loadings of both confirmatory models (i.e., correlated traits and bifactor).

Image Acquisition and Processing

All magnetic resonance imaging (MRI) scans were acquired using the same 3T scanner, and each participant had 1 resting-state and 2 task-based (i.e., n-back and emotion recognition) functional MRI (fMRI) scans (27). The structural images were processed using FreeSurfer (https://surfer.nmr.mgh.harvard.edu/), and the functional images were processed using a top-performing preprocessing pipeline implemented via the XCP Engine (31). The task activation model was regressed out from n-back or emotion identification fMRI data (32–34), and we next concatenated the three fMRI acquisitions, yielding 27 minutes and 45 seconds of functional data for each participant. See Supplemental Methods in Supplement 1 for details.

Defining Personalized Functional Networks

As in our previous work (10), we used spatially regularized nonnegative matrix factorization (NMF) to derive individual-specific (personalized) large-scale functional networks (10,35) (Figure S2 and Supplemental Methods in Supplement 1). Briefly, we randomly selected 100 participants and created 1 group network atlas with 17 networks using NMF. We decomposed the cerebral cortex into 17 networks, which allowed for a direct comparison to the other methods used in prior work (1,5,6). This procedure was repeated 50 times, each time with a different subset of participants; this procedure yielded 50 different group network atlases. By clustering these group atlases using spectral clustering, we acquired 1 consensus group atlas. Next, we derived an individual’s specific network atlas using the group consensus atlas as an initialization; each network’s time series was used to minimize the objective function of the regularized NMF model using the multiplicative update strategy (35). The updates of individualized networks and their representative time series were performed alternatively and iteratively until convergence. Finally, for each person, we created a probabilistic (soft) network atlas, where each vertex had 17 loadings quantifying the extent to which this vertex belongs to each network.

Prediction of Psychopathology Factors From Functional Network Topography Using Partial Least Square Regression

We evaluated whether the multivariate spatial pattern of functional network topography encodes major dimensions of psychopathology (36,37). To do this, we combined the loading maps of 17 networks into a feature vector to represent an individual’s unique multivariate pattern of functional topography. Partial least square regression (PLS-R) was used to test whether functional network topography could predict unseen individuals’ 4 correlated dimensional scores (i.e., fear, psychosis, externalizing, and anxious-misery) from the correlated traits model. Subsequently, the same methods were used to predict each dimension (including overall psychopathology factor) from the bifactor model. We used nested twofold cross-validation (2F-CV), with an outer loop evaluating the generalization of the model to unseen individuals and an inner-loop selecting optimal parameters (number of components, L) (Figure S3 in Supplement 1). We evaluated each prediction by the Pearson correlation coefficient (r) between the actual and predicted scores, as well as by the mean absolute error (MAE). Because the split into two halves was random, we repeated the above 2F-CV procedure 101 times and summarized the prediction accuracy using the median of the distribution. We used 101 repetitions rather than 100 to facilitate the selection of a median value. Within each fold of cross-validation, we regressed out age, sex, and motion from each feature in the training data using a linear model. The acquired coefficients estimated from the training data were used to regress these covariates from each feature of the testing data. This procedure prevents leakage of information between training and testing data, which can lead to overfitting. See Supplemental Methods for the details of prediction analysis.

The significance of the prediction was evaluated using permutation testing, which estimates the empirical distribution of the prediction accuracy under a null hypothesis that there is no association between the pattern of functional topography and major dimensions of psychopathology. The null distribution was generated by executing the whole prediction procedure 1000 times, each time permuting the psychopathology factors across training samples without replacement. Bonferroni correction was applied to permutation testing p values to account for multiple predictions (i.e., 4 dimensions).

In each multivariate model, every vertex received a feature weight for each network (i.e., 17 values per vertex). The absolute value of the weight quantifies the importance of the feature in the predictive model, while the sign indicates a negative or positive association between network loading and the dimensions of psychopathology. Across the 101 split-half runs, we evaluated the median weight for each feature to summarize the contribution weight. To derive a network-level summary measure, we summed all feature weights within each network. A positive network-level weight would indicate that a network has greater representation in individuals with more symptoms, while a negative network-level weight would indicate that a network has a reduced representation in individuals with more symptoms (see Supplemental Methods for more details). We also examined the spatial location of important model features on the cortex by summarizing the overall contribution of each cortical location (vertex) using the sum of the absolute weights of this vertex across all 17 networks.

Linking Functional Network Topography and Pattern of Psychopathology Items Using PLS Correlation

We further validated the association between topography and overall psychopathology by relating functional topography to all 112 item-level psychopathology data using PLS correlation. PLS correlation aims to find pairs of latent components with maximal covariance; we focused on the first pair of latent components, which captured the highest and the most stable covariance (Figure S10 in Supplement 1). As in the PLS-R analysis, repeated 2F-CVs were used to evaluate the out-of-sample correlation of the first pair of components. As in prior work (38,39), we evaluated the contribution of each psychopathology item and topography features to the first pair of latent components. See Supplemental Methods for details.

RESULTS

Personalized networks were generated using NMF; this probabilistic (soft) parcellation can be converted into a discrete (hard) parcellation by labeling each vertex according to its highest loading using NMF (Figure 1). As in previous work (10), we named each personalized network according to its overlap with the canonical 17-network solution defined by Yeo et al. (1). The spatial layout of these networks was largely similar to our prior work (10), with only subtle distinctions observed in this sample, which includes persons with more severe psychopathology who were excluded from the previous study. As expected, across-subject variability of personalized functional network topography was highest in the association cortex (Figure S4A, B in Supplement 1).

Figure 1.

Figure 1.

Open in a new tab

Group atlas used to initialize personalized functional networks. A group atlas was constructed to ensure correspondence across individuals; this group atlas was tailored to each individual to yield personalized networks. The networks in the group atlas include visual (networks 5, 7, and 15), somatomotor (networks 1, 2, 11, and 16), auditory (network 12), dorsal attention (network 13), ventral attention (networks 4 and 10), frontoparietal control (networks 9, 14, and 17), and default mode (networks 3, 6, and 8) networks. In this atlas, there are 17 loadings for each vertex that quantify the extent to which the vertex belongs to each network. For each loading map, brighter colors indicate greater loadings. Vertices can be assigned to the network with the highest loading, yielding a discrete network parcellation (center).

Functional Topography Predicts Correlated Dimensions of Psychopathology

We next related the topography of personalized networks to major dimensions of psychopathology, including fear, psychosis, externalizing, and anxious-misery from the correlated traits factor model (Figure 2A). We observed high interfactor correlations among the 4 dimensions (mean r = 0.71) (Figure S5 in Supplement 1) and substantial similarities in the presence of each factor across screening diagnoses (Figure 2A). Using PLS-R, we found that the complex pattern of network topography could significantly predict unseen individuals’ dimensional scores of psychopathology. Specifically, we found that functional topography could predict symptoms of fear (r = 0.20, Bonferroni-corrected permutation testing p value [pBonf] < .001, MAE = 0.86) (Figure 2B), psychosis (r = 0.16, pBonf < .001, MAE = 0.89) (Figure 2C), externalizing (r = 0.14, pBonf < .001, MAE = 0.84) (Figure 2D), and anxious-misery (r = 0.11, pBonf = .008, MAE = 0.89) (Figure 2E). These Bonferroni-corrected permutation tests indicate that the correlation between actual and predicted scores was significantly higher than expected by chance for each of the 4 dimensions (Figure 2F).

Figure 2.

Figure 2.

Open in a new tab

Functional topography predicts individual differences in major dimensions of psychopathology. (A) A confirmatory correlated traits analysis of 112 psychopathology symptoms previously used to calculate scores on the 4 correlated dimensions of psychopathology, including fear, psychosis, externalizing, and anxious-misery. Mean and standard error of the mean (SEM) of factor scores are displayed for each diagnostic category. As expected, dimensional symptom profiles are substantially similar across screening diagnostic categories, as revealed by mean factor scores across participants of each diagnostic category. See Table S1 in Supplement 1 for the number of participants in each category. (B–E) Functional topography predicts unseen individuals’ dimensions of psychopathology, including fear (B), psychosis (C), externalizing (D), and anxious-misery (E). The data points represent the predicted scores (y-axis) of participants in a model trained on independent data using twofold cross-validation, which was implemented by splitting all participants into two subsets. In each panel, the dark and light colors represent participants of the two subsets, respectively. The p values derived from permutation testing with Bonferroni correction indicated that the actual prediction accuracy (i.e., mean correlation r between two folds) was significantly higher than that expected by chance for all 4 dimensions. (F) Distribution of prediction accuracy values (i.e., correlation r) from permutation testing (small dots and histogram/boxplot) and the actual prediction accuracy (large dot). ADHD, attention-deficit/hyperactivity disorder; MDD, major depressive disorder; OCD, obsessive-compulsive disorder; ODD, oppositional defiant disorder; PS, psychosis spectrum; PTSD, posttraumatic stress disorder; Sep Anxiety, separation anxiety; Soc Phobia, social phobia; Spec Phobia, specific phobia; TD, typically developing.

Major Dimensions of Psychopathology Are Predicted by Similar Patterns of Functional Topography

We found that the network contribution weights were mainly negative in association networks, suggesting reduced cortical representation in association networks with more severe symptoms (Figure 3). For example, feature weights in frontoparietal networks (networks 9 and 17) were negative in models of each of the 4 correlated dimensions (Figure 3A–D). In addition, we found the largest negative feature weights in the ventral attention network (network 4) for psychosis (Figure 3B) and anxious-misery symptoms (Figure 3D), and the largest negative feature weights in the dorsal attention network (network 13) for externalizing symptoms (Figure 3C). In contrast, positive feature weights were mainly observed for somatomotor and visual networks, indicating a greater relative cortical representation of these networks in those with higher psychopathology symptoms.

Figure 3.

Figure 3.

Open in a new tab

The 4 major dimensions of psychopathology are predicted by similar patterns of functional topography. (A–D) Summing the model weights of all vertices within each network revealed that reduced cortical representation in association networks drove the prediction of fear (A), psychosis (B), externalizing (C), and anxious-misery (D) symptoms. (E-H) At each location on the cortex, the absolute weight of each network was summed, revealing that the prefrontal, parietal, and occipital-temporal cortices contributed the most to the multivariate model in the prediction of fear (E), psychosis (F), externalizing (G), and anxious-misery (H) dimensions. AU, auditory; DA, dorsal attention; DM, default mode; FP, frontoparietal; SM, somatomotor; VA, ventral attention; VS, visual.

Examining the contribution of each cortical location, we observed that vertices in the prefrontal, parietal, and temporo-occipital cortices contributed most to predict the burden of psychopathology. This pattern was notably consistent across all 4 dimensions, including fear (Figure 3E), psychosis (Figure 3F), externalizing (Figure 3G), and anxious-misery (Figure 3H). Using spatial permutation testing (Supplemental Methods), we found that the cortical distribution of contribution weights was similar across all dimensions (all pairwise spin testing p values < .001, mean pairwise r = 0.78) (Figure S6 in Supplement 1).

Overall Psychopathology Underlies the Similar Patterns of Functional Topography That Predict Dimensions of Psychopathology

The above analyses established that functional topography predicted the 4 correlated dimensions of psychopathology and that the patterns of feature weights driving these predictions were similar. These results prompted the hypothesis that these shared associations might be driven by a general psychopathology factor that captures symptoms integral to all 4 dimensions of psychopathology. To test this hypothesis, we used orthogonal dimensions of psychopathology from the bifactor model, which includes overall psychopathology, fear, psychosis, externalizing, and anxious-misery (Figure 4A). Averaging factor scores by screening diagnostic category revealed that overall psychopathology was high across all disorders. After parsing out the effects of overall psychopathology factor, each disorder became more distinct in terms of the presence and severity of fear, psychosis, externalizing, and anxious-misery symptoms (Figure 4B).

Figure 4.

Figure 4.

Open in a new tab

Common and divergent dimensions of psychopathology revealed by a bifactor model of psychopathology. (A) A confirmatory bifactor analysis was conducted on the 112 psychopathology items of the clinical screening interview to extract the orthogonal dimensions of psychopathology. These included 4 specific dimensions (i.e., fear, psychosis, externalizing, and anxious-misery) and one common dimension (i.e., overall psychopathology). Mean and standard error of the mean (SEM) of factor scores are displayed for each diagnostic category. (B) Mean factor scores across participants of each diagnostic category illustrate that each specific psychopathology dimension loads more onto the relevant diagnostic categories, while the overall psychopathology factor loads onto all diagnostic categories. ADHD, attention-deficit/hyperactivity disorder; MDD, major depressive disorder; OCD, obsessive-compulsive disorder; ODD, oppositional defiant disorder; PS, psychosis spectrum; PTSD, posttraumatic stress disorder; Sep Anxiety, separation anxiety; Soc Phobia, social phobia; Spec Phobia, specific phobia; TD, typically developing.

We found that the multivariate pattern of functional topography significantly predicted overall psychopathology factor in unseen participants (r = 0.16, permutation testing p < .001, MAE = 0.87) (see Figure 5A, B). We found the largest negative contribution weights in multiple association networks, including the ventral attention (network 4), frontoparietal (network 17), and dorsal attention (network 13) networks (Figure 5C), suggesting that the reduced cortical representation of these networks drove the prediction of overall psychopathology. In contrast, the contribution weights were mainly positive in the somatomotor (networks 2 and 16) and visual (network 5) networks (Figure 5C). To further examine these features, we evaluated the features with the highest (top 25%) absolute contribution weights (Figure S7 in Supplement 1). We observed that vertices in the ventral attention (network 4; Figure 5D) and frontoparietal (e.g., network 17; Figure 5E) networks were predominantly assigned negative weights.

Figure 5.

Figure 5.

Open in a new tab

The functional topography of association networks predicts individual differences in the overall psychopathology factor. (A) Functional topography predicted unseen individuals’ overall psychopathology factor scores. Data points represent the predicted scores of the participants in a model trained on independent data using two-fold cross-validation. The p value was derived from permutation testing. (B) Distribution of prediction accuracies (i.e., correlation r) from permutation testing (small dots and histogram/boxplot) and the actual prediction accuracy (large red dot). (C) The frontoparietal (FP), ventral attention (VA), and dorsal attention (DA) networks contained the highest negative contribution weights, indicative of an inverse relationship between the total cortical representation of those networks and overall psychopathology. (D) Model weights of features driving prediction mainly represented negative values in network 4, including the occipital-temporal junction, insula, and inferior frontal areas. The top 25% of vertices in terms of feature importance are displayed. (E) The vertices in network 17 also mainly represented negative contribution weights in prefrontal areas and the occipital-temporal junction. (F) Vertices located at the prefrontal, parietal, and occipital-temporal cortices drive the prediction of overall psychopathology. (G) The vertices that contributed the most were those sitting at the top of the principal gradient of functional connectivity (54). AU, auditory; DM, default mode; SM, somatomotor; VS, visual.

We next summed the 17 absolute contribution weights for each vertex across networks and found that the vertices in the prefrontal cortex and temporo-occipital junction contributed the most to prediction of the overall psychopathology (Figure 5F). This contribution pattern aligned well (mean r = 0.86, all spin testing p values < .001) with the patterns of contribution weights in the prediction models of the 4 correlated dimensions (Figure 3E–H, above). This result suggests that to a large extent, the association between functional topography and the overall psychopathology factor could explain the predictions of the 4 correlated dimensions of psychopathology. Finally, using spatial permutation testing, we evaluated the association between a vertex’s contribution to predicting overall psychopathology (Figure 5F) and its position along a sensorimotor-to-association cortical hierarchy defined by the principal gradient of functional connectivity in an independent dataset (Figure S8). This analysis revealed a significant positive correlation (r = 0.23, spin testing p = .018), indicating that the transmodal association cortex had higher contribution in the model (Figure 5G).

Overall, the above results indicated that the contribution patterns in prediction of overall psychopathology were highly consistent with those in predictions of the 4 correlated dimensions. We further examined whether functional topography could predict other specific subfactors from the bifactor model, which describe specific dimensions of psychopathology while accounting for overall psychopathology. We found that functional topography significantly predicted the fear factor but did not predict other specific dimensions, including psychosis, anxious-misery, or externalizing (see Supplemental Results and Figure S9 in Supplement 1 for details). Notably, all prediction accuracies declined compared with those of the 4 correlated dimensions, suggesting that the association between functional topography and individual symptom dimensions was largely explained by overall psychopathology.

Finally, we examined whether the association between network topography and overall psychopathology differed by age and sex by training age- and sex-stratified prediction models. We found that the association between topography and overall psychopathology did not significantly differ by age (younger vs. older: p = .16) or sex (male vs. female: p = .33). See Supplemental Results for details.

Analysis of Granular Psychopathology Symptoms Provides Convergent Results

Using PLS correlation to link functional topography and item-level psychopathology symptoms (112 items), we found that the out-of-sample correlation between the first pair of latent components was significantly higher than that expected by chance (r = 0.18, permutation testing p < .001) (Figure 6A, B). Of the 112 psychopathology items, 108 contributed significantly to the first component, suggesting that this component reflected overall psychopathology. Moreover, the network contribution weights of this component were mainly negative in association networks and were mainly positive in sensorimotor networks (Figure 6C). The cortical spatial distribution of contribution weights (Figure 6D) was strongly correlated (r = 0.89, spin testing p < .001) (Figure 6E) with the contribution pattern from the PLS-R prediction model of the overall psychopathology factor. See Supplemental Results for more details.

Figure 6.

Figure 6.

Open in a new tab

Linking the topography with item-level psychopathology symptoms revealed the association between functional topography and overall psychopathology factor. (A) The out-of-sample correlation between the first pair of topography and clinical dimension was r = 0.18 (p < .001). Clinical dimension score was a weighted combination of all psychopathology items (Table S2 in Supplement 1 for weight of each item), while topography dimension score was a weighted combination of all network loadings. p value derived from permutation testing indicated that the actual out-of-sample correlation was significantly higher than that expected by chance. (B) Distribution of out-of-sample correlation from permutation testing (small dots and histogram/boxplot) and the actual out-of-sample correlation (large red dot). (C) Examining the contribution weights of the topography pattern in the first latent component by summing the weights of all vertices within each network revealed that the contribution weights were highly negative in the association networks, including frontoparietal (FP) (networks 9 and 17) and ventral attention (VA) (network 4) networks. (D) At each location on the cortex, the absolute contribution weight of each network was summed, revealing that the prefrontal cortex and temporal-occipital junction contributed the most to the topography pattern in the first component. (E) This contribution pattern was highly correlated with the one (Figure 5F) from the multivariate prediction model of overall psychopathology factor. AU, auditory; DA, dorsal attention; DM, default mode; PLS, partial least square; SM, somatomotor; VS, visual.

DISCUSSION

We found that the spatial topography of large-scale, individual-specific functional networks was associated with individual differences in symptom severity of 4 major dimensions of psychopathology: fear, psychosis, externalizing, and anxious-misery. Furthermore, we demonstrated that these associations between symptoms and functional topography were mainly driven by an individual’s level of overall psychopathology (p factor). Critically, reduced cortical representation in association networks contributed the most to the prediction of overall psychopathology. Taken together, these findings suggest that individual differences in the spatial layout of association networks are systematically related to psychopathology in youth.

This work builds on convergent studies from multiple independent efforts that have demonstrated that there is marked individual variability in the topography of functional networks (3–11,26,40–42). Previous studies have reported that the individual variability of functional topography is maximum in association networks in adults (5–7,10); we previously found that as in adulthood, this is also true in childhood and adolescence (10). Our data further revealed the clinical relevance of this variability by demonstrating that individual variation in functional topography is associated with major dimensions of psychopathology in youth.

Current diagnostic systems (i.e., DSM-5) for psychiatric disorders assign patients into discrete categories based on signs and symptoms. However, efforts including the Research Domain Criteria initiative and the Hierarchical Taxonomy of Psychopathology have been proposed given increasing empirical evidence that psychopathology is a dimensional phenomenon that is highly comorbid in nature (13,15,16,43). In this study, we identified 4 correlated major dimensions of psychopathology that cut across the boundaries of traditional diagnoses. To parsimoniously account for shared vulnerability to all 4 symptom dimensions, we also identified 4 orthogonal dimensions and one overall psychopathology factor that reflects the shared burden of psychopathology across the 4 correlated dimensions (17–20,29).

We found that personalized functional topography significantly predicted the 4 major dimensions of psychopathology in youth. We observed a substantial overlap between brain areas that strongly contributed to prediction across all 4 of the models. This overlap could be related to the clinical overlap among these dimensions, which reflects the high level of comorbid psychopathology in individuals (17,20). Further analysis supported this interpretation by showing that the predictions were largely explained by overall psychopathology. In particular, the contribution pattern predicting overall psychopathology overlapped with the features that predicted each of the 4 correlated dimensions. After accounting for the overall psychopathology factor, functional topography no longer significantly predicted specific factors representing psychosis, externalizing, or anxious-misery. Together, these results suggest that individual variation in functional topography may represent a broad vulnerability factor for transdiagnostic psychopathology.

Consistent with our findings, prior studies have reported transdiagnostic disruptions in functional connectivity across mental disorders (25,44–46) and the associations between overall psychopathology and both within- and between-network functional connectivity (23,24). However, these studies calculated functional connectivity by applying a group-level functional atlas to individuals. Our results provide novel evidence that the spatial topography of personalized functional networks is associated with the level of overall psychopathology. This is critical because topography and connectivity make distinct contributions to individual differences, and using a group atlas for individuals’ data aliases the topographical signal into the measurement of connectivity (9,26). Future studies could leverage personalized functional parcellations for the calculation of individuals’ functional connectivity to isolate the effects of both topography and connectivity (40,47,48). Moreover, to rigorously validate the association between topography and psychopathology, we used a rigorous 2F-CV, which inherently evaluates the generalizability of the association with an unseen individual compared with traditional group-level analyses (36,37).

We found that association networks, most notably the ventral attention and frontoparietal networks, predominantly displayed negative contribution weights in the prediction model of overall psychopathology. These results suggest that greater symptom burden is associated with a reduced cortical representation of association networks, which involve multiple brain regions across the prefrontal, parietal, and temporal cortices (49–51). Previous studies have demonstrated that dysfunction of the frontoparietal (25,45,52) and ventral attention (25,53) networks is a common factor across a broad range of mental disorders. Examining the contribution of cortical location, we observed that the association cortex, which sits at the top of the cortical functional hierarchy (54), contributed the most to the prediction of psychopathology. This pattern is consistent with a prior report that reduced cortical thickness in the association cortex may represent a transdiagnostic feature associated with overall psychopathology (55). These results may emerge given that association cortices support executive, social, and emotional mental functions implicated in psychopathology (52). Furthermore, the development of association networks is defined by a prolonged plastic period that enhances vulnerability to abnormal development and thus dysfunction (51,56).

Several potential limitations should be noted. First, all data presented were cross-sectional; future studies with longitudinal sampling could further inference within-individual development effects. Second, it should be noted that most effect sizes in our work are small. However, prior work has consistently demonstrated that small samples systematically inflate the apparent effect size (57,58), where large samples provide a much more accurate estimate of the true effect size. Third, we combined data from three fMRI runs, including two with task-regressed data. This approach was motivated by prior evidence that functional networks are primarily defined by individual-specific rather than task-specific factors (34) and that intrinsic networks during task performance are similar to those present at rest (32,33). Fourth, certain important psychopathological classes, such as autism spectrum disorder and substance abuse, were not part of the screening interview and thus were not included in this analysis. Future studies may address this by considering broader assessments of psychopathology. Finally, an additional limitation of our approach is that it did not allow for an effective comparison of functional topography and functional connectivity. Prior work has demonstrated that functional topography and connectivity provide unique information in studies of individual differences (9,26). However, we could not directly evaluate this in this work due to the much higher dimensionality of topography features compared with the lower-dimensional number of connectivity features between our personalized 17 networks. Moving forward, use of techniques that delineate higher-resolution functional networks would facilitate this line of investigation.

Conclusions

This study provides novel evidence that personalized functional network topography is related to overall psychopathology in youth. These findings emphasize the relevance of personalized functional neuroanatomy to the neurobiological mechanisms of comorbidity across psychiatric disorders. Because overall psychopathology in part explains a person’s liability to diverse symptoms of mental illness, the potential predictive value of functional topography could potentially aid in the early identification of youths who are at risk of psychopathology. Finally, these results motivate clinical trials of neuromodulatory interventions that are targeted using personalized functional neuroanatomy.

Supplementary Material

Supplementary Materials
Supplementary Table

ACKNOWLEDGMENTS AND DISCLOSURES

This study was supported by grants from the National Institutes of Health (Grant Nos. R01MH113550 [to TDS and DSB], R01EB022573 [to CD, YF, and TDS], R01MH120482 [to TDS], RF1MH116920 [to TDS and DSB], R37MH125829 [to DAF and TDS], R01MH112847 [to RTS and TDS], R01MH112070-01 [to CD], R01MH096773 [to DAF], R01MH115357 [to DAF], R01MH119219 [to RCG and REG], R01NS060910 [to RTS], R01MH123563 [to RTS], F31MH123063-01 [to ARP], T32MH019112 [to JWV], and T32MH014654 [to BL]). VJS was supported by a National Science Foundation Graduate Research Fellowship (Grant No. DGE-1845298). ZC was supported by funds from Beijing Nova Program (Grant No. Z211100002121002) and Chinese Institute for Brain Research, Beijing. The Philadelphia Neurodevelopmental Cohort was supported by Grant Nos. MH089983 and MH089924. Additional support was provided by the Lifespan Brain Institute at Penn and the Children’s Hospital of Philadelphia and the Dowshen Program for Neuroscience. The content is solely the responsibility of the authors and does not represent the official views of any of the funding agencies.

TDS and ZC designed the study. ZC and TDS performed the analyses with support from ARP, HL, and JWV. HL and YF provided parcellation tools. ARP replicated all analyses. AA, TMM, and TDS completed data preprocessing. RCG, REG, CD, and TDS provided resources. ZC, ARP, BL, VJS, and TDS wrote the manuscript. HL, AA, AFA-B, DSB, MB, MEC, CD, DAF, RCG, REG, TMM, SS, RTS, JWV, CHX, and YF reviewed and edited the manuscript.

A previous version of this article was published as a preprint on bioRxiv: https://www.biorxiv.org/content/10.1101/2021.08.02.454763v1.

RTS has consulting income from Genentech/Roche and Octave Bioscience. All other authors report no biomedical financial interests or potential conflicts of interest.

Footnotes

Supplementary material cited in this article is available online at https://doi.org/10.1016/j.biopsych.2022.05.014.

REFERENCES

  • 1.Yeo BTT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, et al. (2011): The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol 106:1125–1165. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Power JD, Cohen AL, Nelson SM, Wig GS, Barnes KA, Church JA, et al. (2011): Functional network organization of the human brain. Neuron 72:665–678. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Laumann TO, Gordon EM, Adeyemo B, Snyder AZ, Joo SJ, Chen MY, et al. (2015): Functional system and areal organization of a highly sampled individual human brain. Neuron 87:657–670. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Glasser MF, Coalson TS, Robinson EC, Hacker CD, Harwell J, Yacoub E, et al. (2016): A multi-modal parcellation of human cerebral cortex. Nature 536:171–178. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Wang D, Buckner RL, Fox MD, Holt DJ, Holmes AJ, Stoecklein S, et al. (2015): Parcellating cortical functional networks in individuals. Nat Neurosci 18:1853–1860. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Kong R, Li J, Orban C, Sabuncu MR, Liu H, Schaefer A, et al. (2019): Spatial topography of individual-specific cortical networks predicts human cognition, personality, and emotion [published correction appears in Cereb Cortex 2021; 31:3974]. Cereb Cortex 29:2533–2551. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Gordon EM, Laumann TO, Gilmore AW, Newbold DJ, Greene DJ, Berg JJ, et al. (2017): Precision functional mapping of individual human brains. Neuron 95:791–807.e7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Gordon EM, Laumann TO, Adeyemo B, Petersen SE (2017): Individual variability of the system-level organization of the human brain. Cereb Cortex 27:386–399. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Li M, Wang D, Ren J, Langs G, Stoecklein S, Brennan BP, et al. (2019): Performing group-level functional image analyses based on homologous functional regions mapped in individuals. PLoS Biol 17:e2007032. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Cui Z, Li H, Xia CH, Larsen B, Adebimpe A, Baum GL, et al. (2020): Individual variation in functional topography of association networks in youth. Neuron 106:340–353.e8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Braga RM, Buckner RL (2017): Parallel interdigitated distributed networks within the individual estimated by intrinsic functional connectivity. Neuron 95:457–471.e5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Psychiatric Association American (2013): Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition Washington, DC: American Psychiatric Publishing. [Google Scholar]
  • 13.Kotov R, Krueger RF, Watson D, Achenbach TM, Althoff RR, Bagby RM, et al. (2017): The Hierarchical Taxonomy of Psychopathology (HiTOP): A dimensional alternative to traditional nosologies. J Abnorm Psychol 126:454–477. [DOI] [PubMed] [Google Scholar]
  • 14.Satterthwaite TD, Feczko E, Kaczkurkin AN, Fair DA (2020): Parsing psychiatric heterogeneity through common and unique circuit-level deficits. Biol Psychiatry 88:4–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Insel TR (2014): The NIMH Research Domain Criteria (RDoC) Project: Precision medicine for psychiatry. Am J Psychiatry 171:395–397. [DOI] [PubMed] [Google Scholar]
  • 16.Insel T, Cuthbert B, Garvey M, Heinssen R, Pine DS, Quinn K, et al. (2010): Research domain criteria (RDoC): Toward a new classification framework for research on mental disorders. Am J Psychiatry 167:748–751. [DOI] [PubMed] [Google Scholar]
  • 17.Lahey BB, Moore TM, Kaczkurkin AN, Zald DH (2021): Hierarchical models of psychopathology: Empirical support, implications, and remaining issues. World Psychiatry 20:57–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Kaczkurkin AN, Park SS, Sotiras A, Moore TM, Calkins ME, Cieslak M, et al. (2019): Evidence for dissociable linkage of dimensions of psychopathology to brain structure in youths. Am J Psychiatry 176:1000–1009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Shanmugan S, Wolf DH, Calkins ME, Moore TM, Ruparel K, Hopson RD, et al. (2016): Common and dissociable mechanisms of executive system dysfunction across psychiatric disorders in youth. Am J Psychiatry 173:517–526. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Caspi A, Moffitt TE (2018): All for one and one for all: Mental disorders in one dimension. Am J Psychiatry 175:831–844. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Caspi A, Houts RM, Belsky DW, Goldman-Mellor SJ, Harrington H, Israel S, et al. (2014): The p factor: One general psychopathology factor in the structure of psychiatric disorders? Clin Psychol Sci 2:119137. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Lahey BB, Applegate B, Hakes JK, Zald DH, Hariri AR, Rathouz PJ (2012): Is there a general factor of prevalent psychopathology during adulthood? J Abnorm Psychol 121:971–977. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Kebets V, Holmes AJ, Orban C, Tang S, Li J, Sun N, et al. (2019): Somatosensory-motor dysconnectivity spans multiple transdiagnostic dimensions of psychopathology. Biol Psychiatry 86:779–791. [DOI] [PubMed] [Google Scholar]
  • 24.Elliott ML, Romer A, Knodt AR, Hariri AR (2018): A connectome-wide functional signature of transdiagnostic risk for mental illness. Biol Psychiatry 84:452–459. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Xia CH, Ma Z, Ciric R, Gu S, Betzel RF, Kaczkurkin AN, et al. (2018): Linked dimensions of psychopathology and connectivity in functional brain networks. Nat Commun 9:3003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Bijsterbosch JD, Woolrich MW, Glasser MF, Robinson EC, Beckmann CF, Van Essen DC, et al. (2018): The relationship between spatial configuration and functional connectivity of brain regions. Elife 7:e32992. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Satterthwaite TD, Elliott MA, Ruparel K, Loughead J, Prabhakaran K, Calkins ME, et al. (2014): Neuroimaging of the Philadelphia Neurodevelopmental Cohort. Neuroimage 86:544–553. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Calkins ME, Merikangas KR, Moore TM, Burstein M, Behr MA, Satterthwaite TD, et al. (2015): The Philadelphia Neurodevelopmental Cohort: Constructing a deep phenotyping collaborative. J Child Psychol Psychiatry 56:1356–1369. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Moore TM, Calkins ME, Satterthwaite TD, Roalf DR, Rosen AFG, Gur RC, Gur RE (2019): Development of a computerized adaptive screening tool for overall psychopathology (“p”). J Psychiatr Res 116:26–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Kaczkurkin AN, Moore TM, Calkins ME, Ciric R, Detre JA, Elliott MA, et al. (2018): Common and dissociable regional cerebral blood flow differences associate with dimensions of psychopathology across categorical diagnoses. Mol Psychiatry 23:1981–1989. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Ciric R, Rosen AFG, Erus G, Cieslak M, Adebimpe A, Cook PA, et al. (2018): Mitigating head motion artifact in functional connectivity MRI. Nat Protoc 13:2801–2826. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Fair DA, Schlaggar BL, Cohen AL, Miezin FM, Dosenbach NUF, Wenger KK, et al. (2007): A method for using blocked and event-related fMRI data to study “resting state” functional connectivity. Neuroimage 35:396–405. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Elliott ML, Knodt AR, Cooke M, Kim MJ, Melzer TR, Keenan R, et al. (2019): General functional connectivity: Shared features of resting-state and task fMRI drive reliable and heritable individual differences in functional brain networks. Neuroimage 189:516–532. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Gratton C, Laumann TO, Nielsen AN, Greene DJ, Gordon EM, Gilmore AW, et al. (2018): Functional brain networks are dominated by stable group and individual factors, not cognitive or daily variation. Neuron 98:439–452.e5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Li H, Satterthwaite TD, Fan Y (2017): Large-scale sparse functional networks from resting state fMRI. Neuroimage 156:1–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Bzdok D, Meyer-Lindenberg A (2018): Machine learning for precision psychiatry: Opportunities and challenges. Biol Psychiatry Cogn Neurosci Neuroimaging 3:223–230. [DOI] [PubMed] [Google Scholar]
  • 37.Arbabshirani MR, Plis S, Sui J, Calhoun VD (2017): Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls. Neuroimage 145:137–165. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Karlaftis VM, Giorgio J, Vértes PE, Wang R, Shen Y, Tino P, et al. (2019): Multimodal imaging of brain connectivity reveals predictors of individual decision strategy in statistical learning. Nat Hum Behav 3:297–307. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Griffis JC, Metcalf NV, Corbetta M, Shulman GL (2019): Structural disconnections explain brain network dysfunction after stroke. Cell Rep 28:2527–2540.e9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Wang D, Li M, Wang M, Schoeppe F, Ren J, Chen H, et al. (2020): Individual-speciffc functional connectivity markers track dimensional and categorical features of psychotic illness. Mol Psychiatry 25:21192129. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Anderson KM, Ge T, Kong R, Patrick LM, Spreng RN, Sabuncu MR, et al. (2021): Heritability of individualized cortical network topography. Proc Natl Acad Sci U S A 118:e2016271118. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Greene DJ, Marek S, Gordon EM, Siegel JS, Gratton C, Laumann TO, et al. (2020): Integrative and network-speciff c connectivity of the basal ganglia and thalamus defined in individuals. Neuron 105:742–758.e6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Kotov R, Krueger RF, Watson D, Cicero DC, Conway CC, DeYoung CG, et al. (2021): The Hierarchical Taxonomy of Psychopathology (HiTOP): A quantitative nosology based on consensus of evidence. Annu Rev Clin Psychol 17:83–108. [DOI] [PubMed] [Google Scholar]
  • 44.Ma Q, Tang Y, Wang F, Liao X, Jiang X, Wei S, et al. (2020): Transdiagnostic dysfunctions in brain modules across patients with schizophrenia, bipolar disorder, and major depressive disorder: A connectome-based study. Schizophr Bull 46:699–712. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Baker JT, Dillon DG, Patrick LM, Roffman JL, Brady RO Jr, Pizzagalli DA, et al. (2019): Functional connectomics of affective and psychotic pathology. Proc Natl Acad Sci U S A 116:9050–9059. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Barron DS, Gao S, Dadashkarimi J, Greene AS, Spann MN, Noble S, et al. (2021): Transdiagnostic, connectome-based prediction of memory constructs across psychiatric disorders. Cereb Cortex 31:2523–2533. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Gratton C, Kraus BT, Greene DJ, Gordon EM, Laumann TO, Nelson SM, et al. (2020): Deff ning individual-speciff c functional neuroanatomy for precision psychiatry. Biol Psychiatry 88:28–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Sylvester CM, Yu Q, Srivastava AB, Marek S, Zheng A, Alexopoulos D, et al. (2020): Individual-speciffc functional connectivity of the amygdala: A substrate for precision psychiatry. Proc Natl Acad Sci U S A 117:3808–3818. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Cheng L, Zhang Y, Li G, Wang J, Sherwood C, Gong G, et al. (2021): Connectional asymmetry of the inferior parietal lobule shapes hemispheric specialization in humans, chimpanzees, and rhesus macaques. Elife 10:e67600. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Numssen O, Bzdok D, Hartwigsen G (2021): Functional specialization within the inferior parietal lobes across cognitive domains. Elife 10:e63591. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Sydnor VJ, Larsen B, Bassett DS, Alexander-Bloch A, Fair DA, Liston C, et al. (2021): Neurodevelopment of the association cortices: Patterns, mechanisms, and implications for psychopathology. Neuron 109:2820–2846. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Cole MW, Repovs G, Anticevic A (2014): The frontoparietal control system: A central role in mental health. Neuroscientist 20:652–664. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Sheffield JM, Kandala S, Tamminga CA, Pearlson GD, Keshavan MS, Sweeney JA, et al. (2017): Transdiagnostic associations between functional brain network integrity and cognition. JAMA Psychiatry 74:605–613. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Margulies DS, Ghosh SS, Goulas A, Falkiewicz M, Huntenburg JM, Langs G, et al. (2016): Situating the default-mode network along a principal gradient of macroscale cortical organization. Proc Natl Acad Sci U S A 113:12574–12579. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Romer AL, Elliott ML, Knodt AR, Sison ML, Ireland D, Houts R, et al. (2021): Pervasively thinner neocortex as a transdiagnostic feature of general psychopathology. Am J Psychiatry 178:174–182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Larsen B, Luna B (2018): Adolescence as a neurobiological critical period for the development of higher-order cognition. Neurosci Biobehav Rev 94:179–195. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Yarkoni T (2009): Big correlations in little studies: Inflated fMRI correlations reflect low statistical power–Commentary on Vul et al. (2009). Perspect Psychol Sci 4:294–298. [DOI] [PubMed] [Google Scholar]
  • 58.Marek S, Tervo-Clemmens B, Calabro FJ, Montez DF, Kay BP, Hatoum AS, et al. (2022): Reproducible brain-wide association studies require thousands of individuals [published correction appears in Nature, online ahead of print May 9, 2022]. Nature 603:654–660. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Materials
NIHMS1880507-supplement-Supplementary_Materials.pdf (3.3MB, pdf)
Supplementary Table
NIHMS1880507-supplement-Supplementary_Table.xlsx (34.1KB, xlsx)

RESOURCES