Optimizing Antidepressant Efficacy: Multimodal Neuroimaging Biomarkers for Prediction of Treatment Response
===========================================================================================================

* Xiaoyu Tong
* Kanhao Zhao
* Gregory A. Fonzo
* Hua Xie
* Nancy B. Carlisle
* Corey J. Keller
* Desmond J. Oathes
* Yvette Sheline
* Charles B. Nemeroff
* Leanne M. Williams
* Madhukar Trivedi
* Amit Etkin
* Yu Zhang

## Abstract

Major depressive disorder (MDD) is a common and often severe condition that profoundly diminishes quality of life for individuals across ages and demographic groups. Unfortunately, current antidepressant and psychotherapeutic treatments exhibit limited efficacy and unsatisfactory response rates in a substantial number of patients. The development of effective therapies for MDD is hindered by the insufficiently understood heterogeneity within the disorder and its elusive underlying mechanisms. To address these challenges, we present a target-oriented multimodal fusion framework that robustly predicts antidepressant response by integrating structural and functional connectivity data (sertraline: R2 = 0.31; placebo: R2 = 0.22). Through the model, we identify multimodal neuroimaging biomarkers of antidepressant response and observe that sertraline and placebo show distinct predictive patterns. We further decompose the overall predictive patterns into constitutive *network constellations* with generalizable structural-functional co-variation, which exhibit treatment-specific association with personality traits and behavioral/cognitive task performance. Our innovative and interpretable multimodal framework provides novel insights into the intricate neuropsychopharmacology of antidepressant treatment and paves the way for advances in precision medicine and development of more targeted antidepressant therapeutics.

**Trial Registration** Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care for Depression (EMBARC), NCT#01407094

## Introduction

Major depressive disorder (MDD) is a pervasive condition that profoundly diminishes quality of life for individuals across ages and demographic groups. Unfortunately, current depression treatments are marked by limited effectiveness and unsatisfactory response rates. The development of effective treatments for MDD is hindered by its insufficiently understood heterogeneity and elusive pathophysiology. Encouragingly, over the past decade, neuroimaging studies have successfully identified promising biomarkers for diagnosing MDD1, 2, delineating symptom profiles3, and predicting treatment responsiveness4–8. Recent research has further delved into individual-level antidepressant response9, 10, shedding light on the heterogeneity among MDD patients. With the ability to measure physiological neural activity and potentially apply these measures to inform treatment of real patients, neuroimaging studies hold great potential in translating knowledge into precision medicine, thus advancing clinical practice for MDD patients.

Nonetheless, two major challenges persist. First, a substantial proportion of MDD patients experience relapse following discontinuation of an initial successful treatment11–13, suggesting that antidepressant medications primarily induce transient neural changes. This raises an intriguing question: how do these transient neural alterations become long-lasting to enable sustained remission? Notably, recent advances in antidepressant response biomarkers are largely based on functional connectivity (FC)9, 10, which captures brain region connectivity but may not effectively reflect the enduringness of connections14. Alternatively, structural connectivity (SC) provides information about the anatomical basis of neural connections that is complementary to the regional synchronization information provided by FC. Therefore, leveraging a multimodal framework to integrate structural and functional information is essential. While numerous recent studies have pioneered multimodal fusion techniques in an unsupervised manner15–18, the resultant fused features are task-agnostic, thus may contain out-of-interest information that potentially leads to overfitting and compromises interpretability. In contrast, a supervised multimodal fusion process guided by prediction targets holds significant potential for improving prediction performance and interpretability. Moreover, the complicated mechanisms of action underlying antidepressants remain poorly understood. Comprehending MDD’s intricate psychopharmacology requires dissecting the overall predictive pattern into constitutive components. Furthermore, the placebo effect in antidepressant response remains inadequately explored19. A recent study has suggested a significant contribution of the placebo effect to rapid antidepressant effect20. Therefore, caution is warranted when interpreting antidepressant biomarkers, and further research is necessary to identify characteristics that differentiate placebo and drug effects.

To address these challenges, we have developed an innovative target-oriented multimodal fusion (TOMMF) framework, which extracts treatment-specific latent space neuroimaging features by integrating SC and FC data to predict antidepressant response (Fig. 1a). By combining SC and FC, our objective is to identify comprehensive pre-treatment biomarkers to achieve MDD patient stratification with more predictable antidepressant response. Importantly, by integrating the multimodal fusion with prediction task, we introduced guidance from target on latent features, achieving the multimodal fusion in a supervised way that improved prediction performance. By further incorporating the L0-regularization technique21, which discourages the contribution of a single feature to multiple dimensions thus resulting in pseudo-exclusive dimension compositions, our approach has the ability in isolating distinct and highly interpretable components within antidepressant response biomarkers (Fig. 1b). Leveraging these distinct components in biomarkers, which we named as *network constellations* (Fig. 1c), we further investigate the characteristic SC-FC co-variation in MDD patients, along with their treatment-specific associations with personality traits and behavioral/cognitive task performance. Our study aims to unravel the distinct roles of structural and functional neural connectivity in predicting antidepressant response, their interplay, and how their treatment-specific patterns can differentiate placebo and drug effects. These efforts offer novel insights into the intricate neuropsychopharmacology of antidepressant treatment and lays the groundwork for improved clinical practice in MDD treatment.

![Fig. 1](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2024/04/12/2024.04.11.24305583/F1.medium.gif)

[Fig. 1](http://medrxiv.org/content/early/2024/04/12/2024.04.11.24305583/F1)

Fig. 1 Target-oriented multimodal fusion (TOMMF) framework.
**a** The overall framework. SC and FC features are extracted from structural and resting-state functional MRI. Subsequently, the framework optimizes coefficients of multimodal fusion (WS and WF) to transform the SC and FC features (XS and XF) into a latent space 𝓖 that distills the co-variation in SC and FC. Importantly, 𝓖 may lean towards a particular modality given its relative importance. Meanwhile, the framework optimizes the coefficients (𝜷) of latent features to predict the target (e.g., antidepressant response). Notably, in this framework with bidirectional propagation, the target not only benefits from the multimodal-derived latent features, but also informs their calculation. The latent space 𝓖 essentially relays the information between the prediction target and each of the data modalities, thus integrating the information from SC, FC, and the target. L0-regularization is imposed to control model complexity and facilitates the identification of distinct latent dimensions. Guidance from target is confined to the training set to prevent information leakage. Multimodal biomarkers for the target are derived from coefficients in cross-validated TOMMF-based prediction models. **b** Dissection of latent space. The overall latent space is essentially a compilation of dimensions, with each dimension incorporating a unique treatment-specific combination of associated SC and FC features. The L0- regularization applied on WS, WF, and 𝜷 discourages the contribution of a single feature to multiple dimensions, resulting in pseudo-exclusive dimension compositions. **c** Network constellations. Network constellations are constructed by parsimoniously aggregating SC and FC features with the strongest correlations at the network level. These constellations allocate network-level connectivity features in an exclusive manner, enhancing the clarity of the resulting psychopharmacological pathways. Importantly, these network constellations (as representations of latent dimensions) contribute to the treatment response prediction in a collective manner.

## Results

### Multimodal Neuroimaging Biomarkers for Sertraline-induced Antidepressant Response

We first applied the TOMMF framework to 103 MDD patients receiving sertraline treatment from the EMBARC dataset22, developing a prediction model for individual antidepressant response based on SC and FC extracted for 100 cortical23 and 35 subcortical24, 25 regions-of-interest (ROIs). Antidepressant response was quantified as the pre- to post-treatment change in 17-item Hamilton Depression Rating Scale scores. The model exhibited strong cross-validation (10x ten-fold) performance in predicting sertraline response (R2 = 0.3149, r = 0.5689, p = 3.63 x 10-10, Fig. 2a), with its significance confirmed by permutation testing (pperm < 0.001, SFig. 1a). Notably, the sertraline response prediction model demonstrated no predictability to placebo response (R2 < 0, r = −0.0324, p = 0.7754, Fig. 2b), suggesting its specificity to sertraline. Furthermore, the TOMMF framework outperformed various baseline methods, including regression models trained with concatenated SC and FC features, sparse canonical correlation analysis-based fused features, partial least squares regression, and single data modalities (one-tailed Wilcoxon signed-rank test: all pFDR ≤ 0.0020, Fig. 2c). Collectively, these findings underscore the potential of the TOMMF framework in providing enhanced predictive power for sertraline response on the individual patient level and identifying neural phenotypic characteristics that can be exploited for treatment stratification.

![Fig. 2](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2024/04/12/2024.04.11.24305583/F2.medium.gif)

[Fig. 2](http://medrxiv.org/content/early/2024/04/12/2024.04.11.24305583/F2)

Fig. 2 Multimodal neuroimaging biomarkers for sertraline-induced antidepressant response.
**a** Multimodal- based sertraline response prediction model. The model demonstrated robust prediction performance for sertraline response evaluated by 10x 10-fold cross-validation. **b** Specificity of sertraline response prediction model to treatment. The model for sertraline response showed no prediction capability for placebo response. **c** Comparison of target-oriented multimodal fusion (TOMMF) framework with baseline methods. The TOMMF framework showed significantly higher prediction performance than baseline methods, including partial least squares regression with concatenated SC and FC features (Concat_PLS), LASSO regression with only SC, only FC, sparse canonical correlation analysis-based fused features (sCCA), or concatenated SC and FC features (Concat). The statistical significance is evaluated by one-tailed paired t-test (p = 0.0024) with 100x 10-fold cross-validation. The error bars indicate the standard error. **d** Important FC and SC features for sertraline response. The predictive pattern exhibited a multivariate foundation encompassing every functional module of the brain. Specifically, the connection between left fusiform gyrus and left middle frontal gyrus showed the strongest contribution among structural connectivity, while the connection between right precuneus and left amygdala displayed the strongest contribution among functional connectivity. The top 20 connectivity features are shown for each data modality for clarity. **e** Important brain regions in the multimodal predictive pattern for sertraline response. The right orbital part of the middle frontal gyrus (R-oMFG) and the left fusiform gyrus (L-FG) showed most substantial contribution in abnormal structural connectivity. The bilateral striata and the right precuneus played crucial roles in abnormal functional connectivity. Within the left striatum, the region with rich connections to the limbic network (L-LIM-Striatum) demonstrated highest importance. Within the right striatum, the region functionally associated with the ventral attention network (R-VAN-Striatum) held particular importance. For visualization purpose, the striata are projected to the nearest surface. The top 10 brain regions are shown for each data modality for clarity. **f** Multimodal network connectivity importance for sertraline response prediction. The network connectivity importance is calculated as the average absolute value of non-zero ROI-level connectivity weights of the prediction model. Overall, the SC features show greater importance than FC features. The VN-DAN and VN-DMN SCs make the strongest contribution to sertraline response prediction. **VN**: Visual Network. **SOM**: Somatomotor Network. **DAN**: Dorsal Attention Network. **VAN**: Ventral Attention Network. **LIM**: Cortical Limbic Network. **FPCN**: Fronto-Parietal Control Network. **DMN**: Default-Mode Network. **CB**: Cerebellum. **SCN**: Subcortical Network.

Afterward, we examined the multimodal neuroimaging biomarkers for sertraline response as dictated by the prediction model weights. Overall, the predictive pattern exhibited a robust multivariate foundation encompassing SC and FC involving every functional brain network (Fig. 2d). Specifically, the connection between left fusiform gyrus and left middle frontal gyrus displayed the most substantial contribution among SC, while the connection between right precuneus and left amygdala demonstrated the strongest contribution among FC. Further investigation of pivotal brain regions in the structural and functional profiles of sertraline response biomarkers (Fig. 2e) revealed that the right orbital part of the middle frontal gyrus and the left fusiform gyrus played key roles in the SC-based biomarker, and the bilateral striata and the right precuneus played were prominent in the FC-based biomarker. Notably, within the left striatum, the region functionally associated with the ventral attention network held particular importance, while the region with rich connections to the limbic network played a crucial role in the right striatum. Lastly but importantly, we examined the network connectivity importance by averaging the absolute non-zero weights of ROI-level connectivity features. As a result, we found that SC exhibited overall greater importance than FC, while the SC between visual network (VN) and dorsal attention network (DAN) and between VN and default mode network (DMN) displayed the strongest contribution to sertraline response prediction (Fig. 2f).

### Multimodal Neuroimaging Biomarkers for Placebo-induced Antidepressant Response

Subsequently, we applied the TOMMF framework to derive a prediction model for predicting treatment response in 120 placebo-medicated MDD patients. As a result, the framework exhibited high cross-validation performance for placebo response prediction (R2 = 0.2190, r = 0.4682, p = 6.96 x 10-8, Fig. 3a), whose significance was further confirmed by permutation test (pperm = 0.001, SFig. 1b). Importantly, the placebo response prediction model was unable to predict sertraline response (R2 < 0, r = −0.0849, p = 0.4370, Fig. 3b), suggesting its specificity to placebo. Furthermore, our target-oriented multimodal framework showed superiority of placebo response prediction over baseline methods (one-tailed Wilcoxon signed-rank test: all pFDR ≤ 9.77 x 10-4, Fig. 3c). Together, these results demonstrated the potential of this framework in providing enhanced predictive power for placebo response and showcased its generalizability to different clinical prediction targets.

![Fig. 3](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2024/04/12/2024.04.11.24305583/F3.medium.gif)

[Fig. 3](http://medrxiv.org/content/early/2024/04/12/2024.04.11.24305583/F3)

Fig. 3 Multimodal neuroimaging biomarkers for placebo-induced antidepressant response.
**a** Multimodal-based placebo response prediction model. The model demonstrated robust prediction performance for placebo response evaluated by 10x 10-fold cross-validation. **b** Specificity of placebo response prediction model to treatment. The model for placebo response showed no prediction capability for sertraline response. **c** Comparison of TOMMF framework with baseline methods. The TOMMF framework showed significantly higher prediction performance than baseline methods, including partial least squares regression with concatenated SC and FC features (Concat_PLS), LASSO regression with only SC, only FC, sparse canonical correlation analysis-based fused features (sCCA), or concatenated SC and FC features (Concat). The statistical significance is evaluated by one-tailed paired t-test (p = 9.74 x 10-6) with 100x 10-fold cross-validation. The error bars indicate the standard error. **d** Important FC and SC features for placebo response. The predictive pattern exhibited a multivariate foundation encompassing every functional module of the brain. Specifically, the connection between right precentral gyrus (R-PreCG) and left anterior hippocampus showed the strongest contribution among structural connectivity, while the connection between left middle frontal gyrus (L-MFG) displayed the strongest contribution among functional connectivity. The top 20 connectivity features are shown for each data modality for clarity. **e** Important brain regions in the multimodal predictive pattern for placebo response. The right precuneus and the R-PreCG showed most substantial contribution in abnormal structural connectivity. The left inferior frontal gyrus (L-IFG) and the L-MFG played crucial roles in abnormal functional connectivity. The top 10 brain regions are shown for each data modality for clarity. **f** Multimodal network connectivity importance for placebo response prediction. The network connectivity importance is calculated as the average absolute value of non-zero ROI-level connectivity weights of the prediction model. Overall, the FC features show greater importance than SC features. The within-FPCN, within-SMN FCs, and SMN-SCN SC make the strongest contribution to placebo response prediction. **VN**: Visual Network. **SOM**: Somatomotor Network. **DAN**: Dorsal Attention Network. **VAN**: Ventral Attention Network. **LIM**: Cortical Limbic Network. **FPCN**: Fronto-Parietal Control Network. **DMN**: Default-Mode Network. **CB**: Cerebellum. **SCN**: Subcortical Network.

We then explored the multimodal neuroimaging biomarkers for placebo response as guided by the prediction model weights. Similar to the biomarkers for sertraline response, the predictive pattern for placebo response exhibited a robust multivariate foundation encompassing every functional brain network (Fig. 3d). Specifically, the connection between right precentral gyrus and left anterior hippocampus displayed the most substantial contribution among SC, while the connection between left middle frontal gyrus and right angular gyrus demonstrated the strongest contribution among FC. Further investigation of important brain regions in the structural and functional profiles of placebo response biomarkers (Fig. 3e) revealed that the right precuneus and the right precentral gyrus played key roles in the SC-based biomarker, while the left inferior frontal gyrus and the left middle frontal gyrus were prominent in the FC-based biomarker. Network connectivity importance revealed that FC provided overall greater contribution for placebo response than SC, while the FCs within somatomotor network (SMN) and within frontoparietal control network (FPCN) and the SC between SMN and subcortical network (SCN) showed particular importance (Fig. 3f).

### Distinct Data Modality Importance for Sertraline and Placebo

Next, we investigated how SC and FC contribute to the sertraline and placebo response respectively. To this end, we examined the effect of the modality importance parameter 𝜌! on prediction performance (see **Methods** for details). Interestingly, our findings indicated that the sertraline response biomarker exhibited a preference for SC, while the placebo response biomarker showcased a more substantial influence of FC (SFig. 2). These results suggested the distinct biomarkers for sertraline and placebo responses, implying the potential difference in their psychopharmacological mechanisms.

### Network Constellations with Generalizable SC-FC Co-variation for MDD patients

To further understand the essence of latent features and parse the intricate psychopharmacology of MDD, we constructed network constellations by aggregating network components with the highest cross-modality correlations based on the sertraline response biomarker (see **Methods** for details). Essentially, we aimed to identify co-variations of SC and FC that are predictive to antidepressant response. As a result, we identified three network constellations with generalizable SC-FC co-variation spanning the intrinsic-extrinsic functioning axis. We named these network constellations based on the pattern of findings, including the default-mode regulatory constellation, the affective constellation, and the sensory processing constellation (Fig. 4a). The default-mode regulatory constellation consisted of the intra-connectivity within DMN, SCN, and their connectivity with DAN and FPCN. The affective constellation was primarily composed of the intra-connectivity of cortical limbic network (LIM) and its connectivity with VN, SMN, and DAN. The sensory processing constellation encompassed the VN’s connectivity to DMN and SCN, as well as connectivity involving attention networks and cerebellum. For the sertraline response biomarker, the sertraline-medicated patients indeed showed significant correlations between the SC and FC constellation scores (default-mode regulatory: r = 0.4385, p = 1.96 x 10-12; affective: 0.3027, p = 1.62 x 10-6; sensory processing: r = 0.1375, r = 0.0190), which were successfully generalized to the placebo-medicated patients (default-mode regulatory: r = 0.1394, p = 0.0154; affective: r = 0.1579, p = 0.0072; sensory processing: 0.1445, p = 0.0126, top row of Fig. 4b).

![Fig. 4](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2024/04/12/2024.04.11.24305583/F4.medium.gif)

[Fig. 4](http://medrxiv.org/content/early/2024/04/12/2024.04.11.24305583/F4)

Fig. 4 Network constellations in antidepressant response biomarkers.
**a** Composition of network constellations. First, network components are quantified as the sum of region-level connectivity scores involving a particular network pair, weighted by antidepressant response model coefficients. Subsequently, network constellations are constructed by aggregating network components with the highest cross-modality correlations. Three generalizable network constellations spanning the intrinsic-extrinsic functioning axis are identified, including the default-mode regulatory constellation, the affective constellation, and the sensory processing constellation. Notably, SC and FC contributions are not equivalent. **b** Generalizable SC-FC co-variation of MDD patients in network constellations. Top row: SC-FC co-variation of sertraline response biomarker. The SC and FC contributions of constellation scores exhibit significant positive correlation on the sertraline arm, demonstrating the framework have successfully identified the link between SC and FC. As SC-FC co-variation is an intrinsic property of pre-treatment neuroimaging data, we expect observe the the same relationship in the other treatment arm. Encouragingly, the patients receiving placebo indeed show significant positive correlation between SC and FC contributions of all three constellation scores, suggesting the generalizability of SC-FC co-variation of sertraline response. Bottom row: SC-FC co-variation of placebo response biomarker. The network constellations constructed on sertraline response biomarker also show robust SC-FC co-variation with placebo response biomarker for all three network constellations. This SC-FC co-variation is generalizable across treatment arms. **c** Treatment-specific traits of network constellations. Top row: Association of treatment-specific network constellations with NEO personality traits. For default-mode regulatory constellation, the sertraline-specific biomarker exhibits a robust association with neuroticism and conscientiousness, while the placebo-specific biomarker significantly associates with openness. The sertraline-specific biomarker in affective constellation also exhibits significant association with neuroticism, while no biomarkers in sensory processing constellation showed significant association with NEO personality traits. Bottom row: Association of treatment-specific network constellations with behavioral/cognitive task performance. The sertraline-specific biomarker in default-mode regulatory constellation significantly associates with WF performance, while the placebo-specific biomarker in sensory processing constellation shows a significant association with CRT. No Biomarkers in the affective constellation exhibits significant correlations with the task performance we have investigated. **A-not-B**: Accuracy in A-not-B task. **CRT**: Choice reaction time. **Flanker_Acc**: Accuracy in flanker test. **Flanker_RT**: Reaction time in flanker test. PRT: Response bias in probabilistic reward task. **WF**: Total number of valid words in word fluency task.

Afterward, we examined whether these network constellations captured SC-FC co-variation in placebo response biomarker. Encouragingly, these network constellations also exhibited generalizable SC-FC co-variation for the placebo response biomarker (placebo-medicated patients: default-mode regulatory: r = 0.4349, p = 8.48 x 10-13; affective: r = 0.2909, p = 2.29 x 10-6; sensory processing: r = 0.1769, p = 0.0032; sertraline-medicated patients: default-mode regulatory: r = 0.1242, p = 0.0305; affective: r = 0.1518, p = 0.0109; sensory processing: r = 0.1801, p = 0.0030, bottom row of Fig. 4b), demonstrating the generalizability of the SC-FC co-variation within these identified network constellations for MDD patients.

### Treatment-specific Traits of Network Constellations

While the network constellations displayed generalizable SC-FC co-variation between treatment arms, they might each exhibit distinct contributions to clinical improvements. Therefore, we then investigated whether these network constellations showed treatment-specific traits as reflected in antidepressant response biomarkers. Notably, the SC and FC constellation scores were summed together to get a combined score in this analysis.

First, as a recent study demonstrated significant association between personality traits and neuropsychopathology26, we examined the correlations between network constellation scores and NEO personality traits of MDD patients (top row of Fig. 4c). Interestingly, in the default-mode regulatory constellation, the sertraline-specific biomarker exhibited a robust association with neuroticism (r = 0.1897, pFDR = 2.12 x 10-4) and conscientiousness (r = −0.1343, pFDR = 0.0098), while the placebo-specific biomarker was significantly associated with openness (r = 0.1524, pFDR = 0.0052). The sertraline-specific biomarker in affective constellation also exhibited significant association with neuroticism (r = 0.1216, pFDR = 0.0451), while no biomarkers in sensory processing constellation showed significant association with NEO personality traits.

Subsequently, we explored the association between network constellation scores and behavioral/cognitive task performance, including the A-not-B task, choice reaction time, the flanker test, the probabilistic reward task, and the word fluency test (bottom row of Fig. 4c). Results revealed that the sertraline-specific biomarker in default-mode regulatory constellation was significantly associated with word fluency performance (r = 0.2431, pFDR = 0.0463), while the placebo-specific biomarker in sensory processing constellation showed a significant association with choice reaction time (r = −0.3131, pFDR = 0.0114). No biomarkers in the affective constellation exhibited significant correlations with the task performance we investigated.

### Enhanced Robustness of Biomarkers Endowed by TOMMF

Lastly, we examined the association between TOMMF-based multimodal biomarkers and SC/FC biomarkers from unimodal baseline methods for both sertraline and placebo responses. Predictive SC exhibited a strong correlation between unimodal- and multimodal-based patterns (rSER = 0.8186, rPLA = 0.8158, SFig. 3), while predictive FC showed a relatively modest correlation (rSER = 0.3401, rPLA = 0.4352, SFig. 3). This highlighted that the TOMMF framework effectively utilized inputs from each data modality for improved prediction. Notably, the pattern correlation for FC aligned with its relative importance for placebo compared to sertraline, and the overall stronger correlation for SC echoed its relative persistence compared to FC.

Subsequently, the robustness of predictive patterns across cross-validation folds was investigated. We respectively examined the intraclass correlation coefficients (ICC) of predictive SC and FC patterns from 10x10 cross-validation folds demonstrated robustness for both sertraline and placebo response in the TOMMF framework (ICCSC,SER = 0.5326, 95% CI [0.5253,0.5400]; ICCFC,SER = 0.3916, 95% CI [0.3846,0.3988]; ICCSC,PLA = 0.4801, 95% CI [0.4727, 0.4875]; ICCFC,PLA = 0.4926, 95% CI [0.4853, 0.5001]). Importantly, these were higher than the robustness of predictive patterns obtained from unimodal methods (ICCSC,SER = 0.4395, 95% CI [0.4322,0.4468]; ICCFC,SER = 0.3018, 95% CI [0.2955,0.3081]; ICCSC,PLA = 0.3972, 95% CI [0.3901,0.4044]; ICCFC,PLA = 0.3169, 95% CI [0.3105, 0.3234]), with statistical significance confirmed by one-tailed two-sample z-tests (pSC,SER = 1.11 x 10-16, pFC,SER = 3.09 x 10-12, pSC,PLA = 2.41 x 10-12, pFC,PLA < 10-323). Collectively, these results suggested that the TOMMF framework enhanced the predictability of antidepressant response by integrating information across data modalities and improved the robustness of identified patterns for each data modality.

## Discussion

In this study, we introduced the TOMMF framework to investigate the intricate neuropsychopharmacology of antidepressant treatment. We explored the roles and interplay of structural and functional neural connectivity and their treatment-specific patterns that distinguish placebo and drug effects. The integration of SC and FC facilitated the development of predictive models for sertraline- and placebo-induced antidepressant responses that outperformed unimodal methods. These models revealed critical brain connections/regions linked to antidepressant response and the associations between SC and FC abnormalities. Furthermore, we parsed the overall predictive patterns into distinct components, revealing constitutive network constellations with generalizable SC-FC co-variation and treatment-specific traits. The TOMMF framework, with its conceptual generalizability and high interpretability, holds promise for applications in investigating other psychiatric disorders. Our findings offer enhanced predictive power for antidepressant responses with multimodal biomarkers, paving the way for precision medicine in MDD and providing a valuable tool for future multimodal neuroimaging biomarker studies.

Multimodal fusion is a technique emergingly employed by neuroimaging-based psychiatric studies, aiming to deepen our understanding of psychopathological neural alterations27–30 and improve predictability in psychiatric diagnosis31–33 and treatment effect34. Our innovative approach incorporates prediction target guidance into the multimodal fusion process, yielding substantial benefits over unsupervised fusion for downstream prediction tasks. As a recent study showed that the general neuroimaging variability across subjects mostly represent subclinical heterogeneity35, the target-oriented latent features ensured its relevance to antidepressant responses. Notably, the learning objective minimized the distance between canonical dimensions of SC and FC. Therefore, the latent features essentially reflect the information in agreement, thus identifying the association between SC and FC. Importantly, TOMMF’s linear basis simplifies the complexity, enhancing generalizability and facilitating neurophysiological interpretation of identified biomarkers. Furthermore, the L0-regularization achieved a high degree of orthogonality in latent features without explicit constraints, achieving data-driven dimensionality and subdimension identification in the overall predictive patterns, which elucidates the psychopharmacological intricacies of MDD.

Interestingly, our findings suggest that SC is more informative for predicting sertraline response, while FC is more indicative of placebo response. Conceptually, SC reflects the stable and enduring anatomical brain structure, while FC signifies the more adaptable active interactions between brain regions17. As previous studies have noted a significantly higher relapse rate with placebo compared to antidepressant medications after treatment discontinuation following remission36–38, we speculate that the greater variation in SC may contribute to the relatively persistent sertraline response, whereas the intermittent placebo response is associated with greater variation in FC. Due to the design of the clinical trial, we were unable to examine the heterogeneity in relapse using the EMBARC dataset. Additional research is warranted to provide further evidence for this theory, including investigating whether patients experiencing relapse earlier exhibit greater FC alteration compared to those experiencing relapse later or without relapse, irrespective of the treatment received. Its generalizability to other antidepressant medications should also be investigated.

Leveraging the innovative TOMMF framework, we identified neuroimaging biomarkers for antidepressant response on a multimodal basis. Notably, while sertraline and placebo responses exhibited distinct biomarkers, the precuneus emerged as a consistently significant predictor for both treatments. This underscores the crucial role of the precuneus in modulating antidepressant effects, aligning with previous findings indicating its abnormalities in MDD patients39 and predictive value for treatment outcome10, 40. Our study also provided insights into the contribution of subcortical regions to sertraline response, a facet not extensively explored in prior individual-level antidepressant response prediction studies10, 40. Specifically, we found functional connectivity involving striatum is particularly important for sertraline response, echoing with previous findings that FC involving striatum substantially associates with MDD symptoms41 and responses to repetitive transcranial magnetic stimulation42. Our models did not attribute as much importance to hippocampal structural abnormalities for antidepressant response, despite the hippocampus being consistently identified as a biomarker for MDD diagnosis43. This highlights the unique information provided by SC compared with anatomical volume of brain regions and suggests a possible discrepancy between the psychopathology of psychopharmacology of MDD. Moreover, the subcortical regions exhibited less substantial contribution in placebo response, in line with a recent finding highlighting differences in brain gradient between sertraline and placebo responses40. Importantly, our biomarker findings discerned between the contributions of SC and FC, providing novel insights into the respective roles of persistent and adaptable neural connections in treatment efficacy. Inspired by these findings, we advocate for future research into how persistent and adaptable neural connectivity influences antidepressant response and how therapeutics can facilitate the transition from transient treatment-induced neurophysiological changes to persistent ones.

Finally, the construction of latent features effectively disentangled the overarching predictive patterns for antidepressant response into distinct components. Our findings unveiled three distinct network constellations (i.e., clusters of networks) that synergistically informed the prediction of antidepressant response. These constellations, including the default-mode regulatory, affective, and sensory processing constellations, encompassed a spectrum from intrinsic to extrinsic functions and demonstrated a high degree of independence from each other. As validated by their robust SC-FC co-variation and generalizability across treatment arms, these network constellations offered novel insights into the intricate psychopharmacology of MDD. They also suggested personality traits and behavioral/cognitive task performance as potential indicators of treatment outcome. To further validate and extend our discoveries, we encourage future studies to scrutinize the heterogeneity in response to other antidepressant therapeutics and explore potential distinctions between pharmacological and pathological aspects43.

In summary, we present a novel, adaptable, and highly explainable multimodal framework capable of predicting antidepressant response. Beyond the identification of biomarkers, our framework illuminates the intricate interplay between data modalities and discerns distinct components within the overall predictive patterns, which establishes it as a promising tool for investigating psychiatric disorders with multimodal neuroimaging data. Our key observation emphasizes that sertraline and placebo responses correspond to greater alterations in SC and FC, underscoring the significance of persistent neurophysiological changes for optimal treatment outcomes. Furthermore, we have delineated three network constellations constituting the overarching antidepressant response biomarkers, which unveil potential distinct aspects within the complex psychopharmacology of MDD. These novel insights serve as inspiration for future research, advocating the development of therapeutics that induce transient changes to persistent alterations and address each constitutive neural circuitry within the comprehensive psychopharmacological pathways of MDD. This approach has the potential to advance the field toward precision medicine for MDD and foster the development of more efficacious antidepressant treatments.

## Methods

### Participants

The EMBARC study recruited 296 MDD patients and randomly assigned these subjects into the sertraline and placebo treatment arms22. After excluding disqualified/withdrawn subjects and the subjects with missing or low-quality neuroimaging data, we included 103 sertraline-medicated and 120 placebo-medicated MDD patients aged between 18 and 65 in our analyses (SFig. 4). The EMBARC study was conducted according to the FDA guidelines and the Declaration of Helsinki and approved by Institutional Review Board of each clinical site. All participants have given signed informed consent and agreement to all procedures prior to entry. The inclusion criteria of EMBARC study included: 1) MDD as the primary diagnosis by the Structured Clinical Interview for DSM-IV Axis I Disorders44, 2) Quick Inventory of Depressive Symptomatology score ≥ 14, 3) a MDD episode onset prior to age 30, either a chronic recurrent episode (duration ≥ 2 years) or recurrent MDD (≥ two lifetime episodes), and 4) no antidepressant failure during the current episode. The exclusion criteria of EMBARC study included: 1) ongoing pregnancy or breastfeeding, 2) being sexually active without contraception, 3) lifetime history of psychosis or bipolar disorder, 4) substance dependence in last 6 months or substance abuse in last 2 months, 5) unstable psychiatric or general medical conditions that require hospitalization, 6) study medication contraindication, 7) clinically significant laboratory abnormalities, 8) history of epilepsy or conditions requiring an anticonvulsant, 9) electroconvulsive therapy, vagal nerve stimulation, TMS or other somatic treatments in the current episode, 10) ongoing medication intake (including but not limited to antipsychotics and mood stabilizers), 11) ongoing psychotherapy, 12) significant suicide risk, and 13) unresponsiveness to any antidepressant at adequate dose and duration in the current episode.

For patients in either treatment arm, an eight-week course of sertraline or placebo was administered. Stratification was enforced by site, depression severity, and chronicity for the randomization of treatment arm. The dosing of medications started at 50mg and increased to a maximum of 200mg given patient’s tolerance and unresponsiveness to lower dosing. The treatment response was evaluated using HAMD17. Unavailable endpoint HAMD17 scores were imputed from subjects’ intermediate treatment response where possible to maximally retain the sample size. Additionally, neuroimaging data (including diffusion and functional MRI) were scanned for MDD patients before the treatment.

### MRI Acquisition

#### Diffusion MRI Acquisition

Different scanners were utilized at the four clinical sites, including the General Electric 3T scanner (Columbia University), the Siemens 3T scanner (Massachusetts General Hospital), and the Philips 3T scanner (University of Texas Southwestern Medical Center and University of Michigan). All four clinical sites acquired diffusion MRI data with a voxel size of 1 × 1 x 1 mm3. The four clinical sites collected diffusion MRI data with different parameters. Specifically, the acquisition parameters were: Columbia University: 6 msec repetition time, 2.4 msec echo time, 9° flip angle, 5 min duration; Massachusetts General Hospital: 2.3 msec repetition time, 2.54 msec echo time, 9° flip angle, 4.3 min duration; University of Texas Southwestern Medical Center: 8 msec repetition time, 3.7 msec echo time, 12° flip angle, 4.24 min duration; University of Michigan: 8.1 msec repetition time, 3.7 msec echo time, 12° flip angle, 5.29 min duration.

#### Functional MRI Acquisition

Resting-state fMRI data were collected using a single-shot gradient echo-planar pulse sequence. fMRI recordings were acquired with the same parameters across study sites (2000 msec repetition time, 28 msec echo time, 90° flip angle, 64 × 64 matrix size, 3.2 × 3.2 x 3.1 mm3 voxel size, with 39 axial slices and 180 image volumes). The same scanners were used as data acquisition of diffusion MRI.

### MRI Preprocessing

#### Diffusion MRI Preprocessing

Images with a b-value less than 100 s/mm2 were first designated as b=0 images. A denoising procedure was then implemented using the MRtrix3’s method45 with a 5-voxel window. Subsequently, correction for B1 field inhomogeneity was conducted using the dwibiascorrect function from MRtrix3 with the N4 algorithm incorporated46. Afterward, the mean intensity of the DWI series was adjusted to calibrate the mean intensity of the b=0 images across each individual DWI scanning sequence. FSL’s eddy was employed for head motion and Eddy correction47, with a q-space smoothing factor of 10, 5 iterations, and 1000 voxel-based hyperparameter estimation. Linear first- and second-level models were employed to characterize Eddy current-related spatial distortion and q-space coordinates were assigned to the respective shells. Field offset was separated from subject movement. Post-eddy, Eddy’s outlier replacement48 was conducted. Data were further organized by slices, ensuring the inclusion of values from slices with a minimum of 250 intracerebral voxels. Outlier groups that deviated by more than 4 standard deviations from the prediction were replaced with imputed values.

#### Functional MRI Preprocessing

The acquired resting-state fMRI data were preprocessed using the fMRIPrep pipeline49. Specifically, the T1-weighted images were corrected for intensity nonuniformity and stripped skull. Spatial normalization was conducted via nonlinear registration with the T1-weighted reference50. Brain tissue was segmented from the reference brain-extracted T1-weighted image51. The fieldmap information was used to correct distortion in low- and high-frequency components caused by field inhomogeneity. Subsequently, a corrected echo-planar imaging reference was obtained from a more accurate co-registration with the anatomical reference. The blood-oxygen-level-dependent (BOLD) reference was then transformed to the T1-weighted image with a boundary-based registration method52. Automatic removal of motion artifacts using independent component analysis53 was performed on the preprocessed BOLD time-series after removal of non-steady-state volumes and spatial smoothing with an isotropic Gaussian kernel of 6 mm full-width half-maximum. Lastly, quality control was enforced to removed data with high head motion (>0.5 mm motion framewise displacement or BOLD signal displacements > 0.5%)54, 55.

### Connectivity Feature Calculation

#### Structural Connectivity Calculation

We calculated the structural connectivity (SC) of all subjects using DSI studio program ([https://dsi-studio.labsolver.org](https://dsi-studio.labsolver.org)). Specifically, we first created a white matter mask with thresholding. Afterward, the spin distribution function of each voxel of masked image was estimated using generalized q-sampling imaging (GQI) reconstruction56 in the T1-weighted space with a diffusion sampling length ratio of 1.25. The tensor metrics were calculated using DWI with b-value lower than 1750 s/mm². The whole brain fibers were reconstructed using a modified streamline deterministic tracking algorithm57 with augmented tracking strategies58 and whole brain seeding strategy. The parameters were set as angular threshold 45°, step size 1 mm, fiber length extent 20–300 mm, and maximum subvoxel search seeds 1,000,000. In each tracking iteration, a random voxel coordinate within the whole brain seed was selected and fiber tracking algorithm started from the coordinate and tracks in two directions until the fiber reaches the brain border. Tracks with lengths shorter than 30 or longer than 200 mm were discarded. Finally, we extracted SC by calculating the counts of connected fibers linked pair-wise ROIs defined from the Schaefer atlas of 100 parcels23 and subcortical parcellations24, 25. The 35 subcortical ROIs include striatal and cerebellar regions, amygdala regions, anterior and posterior hippocampal regions, and the thalamus. We excluded the SC which do not exist in real brain or only exist in a proportion of subjects (< 67%) to improve the stability of predictive patterns across cross-validation folds. Lastly, the fiber numbers were log-transformed to enhance normality.

#### Functional Connectivity Calculation

Functional connectivity (FC) was extracted based on the same cortical and subcortical ROIs as used for structural connectivity extraction. The preprocessed fMRI data were averaged into 135 ROI-level time series. For each subject, FC was calculated as the Pearson’s correlation coefficient between the time series of each pair of ROIs. Fisher’s r-to-z transformation was implemented to ensure the normality of the FC features.

### The TOMMF framework (Multimodal-based Regression Model with L0-regularization)

To integrate information from multiple data modalities for improvement in prediction performance, we developed a framework that combines the multimodal fusion and regression task (Fig. 1a). Additionally, because the neural connectivity features outnumbered the sample size of subjects, sparsity constraint is introduced to control the model complexity thus alleviating the potential overfitting issue. Notably, here we utilized L0-regularization as the sparsity constraint for its non-shrinking variance and natural interpretability (see **Appendix B** for details), as compared with the commonly used L1-regularization (a.k.a. LASSO). We employed a technique named continuous sparsification21 to realize the optimization of the non-convex and non-differentiable L0-regularization (see **Appendix A** for details).

The overall loss function of the TOMMF framework can be formularized as: ![Formula][1]</img>  The notation 𝛬 denotes the set of data modalities, in our case it contains the structural (S) and functional (F) modalities. The multimodal fusion term ![Graphic][2]</img> aims to construct a latent space 𝒢 that represents the information in agreement across data modalities, where the modality importance ratio 𝜌𝛬 is a hyperparameter that controls the relative contribution of each data modality. Incorporating 𝜌𝛬 into the multimodal fusion term enables the framework to lean the latent space toward the data modality that is more informative for a particular prediction target. Remarkably, as we minimize this multimodal fusion term, the distance between 𝑋$𝑊$ and 𝑋&𝑊&is also minimized. Therefore, the multimodal fusion step also identifies canonical association between SC and FC features.

The prediction task term ![Graphic][3]</img> is a standard loss for prediction residue, where 𝐾4567 modulates the relative importance of multimodal fusion and prediction loss. Both multimodal fusion and prediction task are regularized by the L0-norm of their feature weights (𝑊$, 𝑊&, and 𝛽). This L0-regularization endows the framework with two significant characteristics – data-driven dimensionality and pseudo-exclusive compositions of the latent features (Fig. 1b, see **Appendix B** for details). These unique properties enable the framework to determine the number of features in the latent space based on their informativeness instead of an empirical percentage, as well as significantly enhance the interpretability of the latent features.

Additionally, by incorporating both multimodal fusion and prediction loss, our approach directs the fusion process with guidance from target – the optimal latent space 𝒢 is dependent on the target variable 𝑌. This crucial relationship ensures that the latent space is tailored to the prediction target, resulting in distinct features for sertraline and placebo responses. In contrast to unsupervised multimodal fusion techniques (Fig. 2c, 3c), our proposed TOMMF framework conducts multimodal fusion in a unsupervised manner, enhancing the relevance of latent features and, thereby improving the prediction performance. With both multimodal fusion and prediction model relying on linear transformations, the overall predictive pattern from original SC and FC features to the antidepressant response is also linear. As a result, the TOMMF framework offers excellent end-to-end interpretability. Specifically, the equation for generating prediction 𝑌𝛬 based on original SC and FC features (𝑋𝑆 and 𝑋𝐹) can be formularized as (see **Appendix A** for details): ![Formula][4]</img>  We can observe from the above formula that ![Graphic][5]</img> and ![Graphic][6]</img> are the linear transformation matrices for the original SC and FC features, therefore granting the linear interpretability of the predictive biomarkers for antidepressant response.

### Network Constellations

The network constellations were constructed to represent the distinct dimensions in the common space (Fig. 1c). To enhance their interpretability, we leveraged the dissection of functional networks which partition the brain into functional modules59. Specifically, we quantified the network components within biomarkers by summing ROI-level connectivity features involving a particular pair of networks, weighted by their coefficients from the antidepressant response prediction model. We calculated these components for SC and FC, respectively. Afterwards, we constructed network constellations by parsimoniously aggregating network components with the highest cross-modality correlations. Notably, SC and FC components for the same network were not considered equivalent. To derive psychopharmacological components with high generalizability and reliability, the network constellations were constructed based on sertraline response biomarker and then validated by the placebo response biomarker.

### NEO Personality Traits and Cognitive Tasks

We aimed to explore the relationship between personality traits and the network constellation components identified within antidepressant response biomarkers, building on recent findings of extensive associations between personality traits and psychopathology26. Personality traits were assessed using the NEO-five factor inventory, including agreeableness, neuroticism, extraversion, openness, and conscientiousness in MDD patients, as part of the EMBARC study22.

Additionally, behavioral/cognitive task performance has shown substantial associations with symptom profiles and treatment responsiveness in MDD22. The EMBARC study conducted five cognitive tasks, which we analyzed in this study. These tasks include the A-not-B task, choice reaction, the flanker test, the probabilistic reward task, and the word fluency test. Specifically, we examined patients’ accuracy in the A-not-B task, choice reaction time, accuracy and reaction time in the flanker test, response bias in the probabilistic reward task, and the total number of valid words in the word fluency test. After excluding subjects with missing neuroimaging data, our analysis included 49 MDD patients for the A-not-B task, 48 for choice reaction time, 39 for the flanker test, 50 for the probabilistic reward task, and 60 for the word fluency test.

### Statistical Tests

#### Comparing Intraclass Correlation Coefficients

To compare the intraclass correlation coefficients (ICCs) of predictive patterns derived from unimodal and multimodal frameworks, we performed two-sample z-test on the z-scores of ICCs. Specifically, Fisher z-transformation was first conducted on ICCs ![Graphic][7]</img> to ensure the normality of variables60. The z-statistic for two-sample z-test can be formularized as ![Graphic][8]</img>, where ![Graphic][9]</img> is the standard error of the difference between means and 𝑆𝐸, (𝑖 = 1,2) is the standard error of each mean. Given the approximated standard error of Fisher’s ![Graphic][10]</img>, we can further formularize the z-statistic as: ![Formula][11]</img>  𝑁1and 𝑁2 are the sample sizes for the two ICCs, which are the numbers of total connectivity features. Essentially, we examined the importance of each connectivity feature with the models in each of the cross-validation folds as judges. The p-value corresponding to the z-statistic can be found based on standard normal distribution.

## Data Availability

The EMBARC data are publicly available through the National Institute of Mental Health Data Archive (NDA) ([https://nda.nih.gov/edit_collection.html?id=2199](https://nda.nih.gov/edit_collection.html?id=2199)).

## Code Availability

All analyses were conducted in MATLAB (version R2022b) and the code will be made available upon the publication of this paper.

## Supporting information

Supplementary Material [[supplements/305583_file03.pdf]](pending:yes)

Appendix A [[supplements/305583_file04.pdf]](pending:yes)

Appendix B [[supplements/305583_file05.pdf]](pending:yes)

## Financial Disclosures

G.A.F. received monetary compensation for consulting work for SynapseBio AI and owns equity in Alto Neuroscience. A.E. reports salary and equity from Alto Neuroscience and equity in Akili Interactive and Mindstrong Health. C.J.K reports equity from Alto Neuroscience. The remaining authors declare no competing interests.

## Acknowledgements

This work was supported by NIH grant nos. R01MH129694, R21MH130956, R21AG080425, Alzheimer’s Association Grant (AARG-22-972541), and Lehigh University FIG (FIGAWD35), CORE, and Accelerator grants. Portions of this research were conducted on Lehigh University’s Research Computing infrastructure partially supported by NSF Award 2019035. G.A.F. was also supported by philanthropic funding and NIH grant nos. K23MH114023 and R01MH125886, and grants from the One Mind - Baszucki Brain Research Fund, the SEAL Future Foundation, and the Brain and Behavior Research Foundation.

*   Received April 11, 2024.
*   Revision received April 11, 2024.
*   Accepted April 12, 2024.


*   © 2024, Posted by Cold Spring Harbor Laboratory

This pre-print is available under a Creative Commons License (Attribution-NonCommercial-NoDerivs 4.0 International), CC BY-NC-ND 4.0, as described at [http://creativecommons.org/licenses/by-nc-nd/4.0/](http://creativecommons.org/licenses/by-nc-nd/4.0/)

## References

1.  1.Lai, C.-H. Promising neuroimaging biomarkers in depression. Psychiatry investigation 16, 662 (2019).
    
    

2.  2.Kambeitz, J. et al. Detecting neuroimaging biomarkers for depression: a meta-analysis of multivariate pattern recognition studies. Biological psychiatry 82, 330–338 (2017).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.biopsych.2016.10.028&link_type=DOI) 

3.  3.Buch, A.M. & Liston, C. Dissecting diagnostic heterogeneity in depression by integrating neuroimaging and genetics. Neuropsychopharmacology 46, 156–175 (2021).
    
    

4.  4.Kraus, C., Kadriu, B., Lanzenberger, R., Zarate Jr, C.A. & Kasper, S. Prognosis and improved outcomes in major depression: a review. Translational psychiatry 9, 127 (2019).
    
    

5.  5.Runia, N. et al. The neurobiology of treatment-resistant depression: a systematic review of neuroimaging studies. Neuroscience & Biobehavioral Reviews 132, 433–448 (2022).
    
    

6.  6.Enneking, V., Leehr, E.J., Dannlowski, U. & Redlich, R. Brain structural effects of treatments for depression and biomarkers of response: a systematic review of neuroimaging studies. Psychological Medicine 50, 187–209 (2020).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1017/S0033291719003660&link_type=DOI) 

7.  7.Fu, C.H. et al. Neuroanatomical dimensions in medication-free individuals with major depressive disorder and treatment response to SSRI antidepressant medications or placebo. Nature Mental Health, 1–13 (2024).
    
    

8.  8.Widge, A.S. et al. Electroencephalographic biomarkers for treatment response prediction in major depressive illness: a meta-analysis. American Journal of Psychiatry 176, 44–56 (2019).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1176/appi.ajp.2018.17121358&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=30278789&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

9.  9.Wu, W. et al. An electroencephalographic signature predicts antidepressant response in major depression. Nature biotechnology 38, 439–447 (2020).
    
    

10. 10.Zhao, K. et al. Individualized fMRI connectivity defines signatures of antidepressant and placebo responses in major depression. Molecular psychiatry, 1–10 (2023).
    
    

11. 11.Geddes, J.R. et al. Relapse prevention with antidepressant drug treatment in depressive disorders: a systematic review. The Lancet 361, 653–661 (2003).
    
    

12. 12.Kanai, T. et al. Time to recurrence after recovery from major depressive episodes and its predictors. Psychological Medicine 33, 839–845 (2003).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1017/S0033291703007827&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=12877398&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

13. 13.Breedvelt, J.J.F. et al. An individual participant data meta-analysis of psychological interventions for preventing depression relapse. Nature Mental Health (2024).
    
    

14. 14.Honey, C.J. et al. Predicting human resting-state functional connectivity from structural connectivity. Proceedings of the National Academy of Sciences 106, 2035–2040 (2009).
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoicG5hcyI7czo1OiJyZXNpZCI7czoxMDoiMTA2LzYvMjAzNSI7czo0OiJhdG9tIjtzOjUwOiIvbWVkcnhpdi9lYXJseS8yMDI0LzA0LzEyLzIwMjQuMDQuMTEuMjQzMDU1ODMuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

15. 15.Sui, J., Adali, T., Yu, Q., Chen, J. & Calhoun, V.D. A review of multivariate methods for multimodal fusion of brain imaging data. Journal of neuroscience methods 204, 68–81 (2012).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jneumeth.2011.10.031&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=22108139&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000299977400008&link_type=ISI) 

16. 16.Zhang, Y.-D. et al. Advances in multimodal data fusion in neuroimaging: Overview, challenges, and novel orientation. Information Fusion 64, 149–187 (2020).
    
    

17. 17.Tulay, E.E., Metin, B., Tarhan, N. & Arıkan, M.K. Multimodal neuroimaging: basic concepts and classification of neuropsychiatric diseases. Clinical EEG and neuroscience 50, 20–33 (2019).
    
    

18. 18.Maglanoc, L.A. et al. Multimodal fusion of structural and functional brain imaging in depression using linked independent component analysis. Human brain mapping 41, 241–255 (2020).
    
    

19. 19.Burke, M.J. et al. Placebo effects and neuromodulation for depression: a meta-analysis and evaluation of shared mechanisms. Molecular psychiatry 27, 1658–1666 (2022).
    
    

20. 20.Lii, T.R. et al. Randomized trial of ketamine masked by surgical anesthesia in patients with depression. Nature Mental Health, 1–11 (2023).
    
    

21. 21.Savarese, P., Silva, H. & Maire, M. Winning the lottery with continuous sparsification. Advances in neural information processing systems 33, 11380–11390 (2020).
    
    

22. 22.Trivedi, M.H. et al. Establishing moderators and biosignatures of antidepressant response in clinical care (EMBARC): Rationale and design. Journal of psychiatric research 78, 11–23 (2016).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jpsychires.2016.03.001&link_type=DOI) 

23. 23.Schaefer, A. et al. Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI. Cerebral Cortex 28, 3095–3114 (2017).
    
    

24. 24.Buckner, R.L., Krienen, F.M., Castellanos, A., Diaz, J.C. & Yeo, B.T. The organization of the human cerebellum estimated by intrinsic functional connectivity. Journal of neurophysiology 106, 2322–2345 (2011).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1152/jn.00339.2011&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=21795627&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000297690500019&link_type=ISI) 

25. 25.Choi, E.Y., Yeo, B.T. & Buckner, R.L. The organization of the human striatum estimated by intrinsic functional connectivity. Journal of neurophysiology 108, 2242–2263 (2012).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1152/jn.00270.2012&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=22832566&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000310057500014&link_type=ISI) 

26. 26.Zhang, Y.-R. et al. Personality traits and brain health: a large prospective cohort study. Nature Mental Health, 1–14 (2023).
    
    

27. 27.Sui, J. et al. In search of multimodal neuroimaging biomarkers of cognitive deficits in schizophrenia. Biological psychiatry 78, 794–804 (2015).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.biopsych.2015.02.017&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=25847180&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

28. 28.Foti, D., Carlson, J.M., Sauder, C.L. & Proudfit, G.H. Reward dysfunction in major depression: Multimodal neuroimaging evidence for refining the melancholic phenotype. NeuroImage 101, 50–58 (2014).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2014.06.058&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=24996119&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

29. 29.Scheepens, D.S. et al. The link between structural and functional brain abnormalities in depression: A systematic review of multimodal neuroimaging studies. Frontiers in Psychiatry 11, 485 (2020).
    
    

30. 30.Johnston, J.A. et al. Multimodal neuroimaging of frontolimbic structure and function associated with suicide attempts in adolescents and young adults with bipolar disorder. American journal of psychiatry 174, 667–675 (2017).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1176/appi.ajp.2016.15050652&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=28135845&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

31. 31.Shi, J., Zheng, X., Li, Y., Zhang, Q. & Ying, S. Multimodal neuroimaging feature learning with multimodal stacked deep polynomial networks for diagnosis of Alzheimer’s disease. IEEE journal of biomedical and health informatics 22, 173–183 (2017).
    
    

32. 32.Liu, S. et al. Multimodal neuroimaging feature learning for multiclass diagnosis of Alzheimer’s disease. IEEE transactions on biomedical engineering 62, 1132–1140 (2014).
    
    

33. 33.Lei, D. et al. Integrating machining learning and multimodal neuroimaging to detect schizophrenia at the level of the individual. Human brain mapping 41, 1119–1135 (2020).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/hbm.24863&link_type=DOI) 

34. 34.Weigand, A. et al. Predicting antidepressant effects of ketamine: the role of the pregenual anterior cingulate cortex as a multimodal neuroimaging biomarker. International Journal of Neuropsychopharmacology 25, 1003–1013 (2022).
    
    

35. 35.Tong, X. et al. Symptom dimensions of resting-state electroencephalographic functional connectivity in autism. Nature Mental Health (2024).
    
    

36. 36.Kato, M. et al. Discontinuation of antidepressants after remission with antidepressant medication in major depressive disorder: a systematic review and meta-analysis. Molecular psychiatry 26, 118–133 (2021).
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

37. 37.Hansen, R. et al. Meta-analysis of major depressive disorder relapse and recurrence with second-generation antidepressants. Psychiatric services 59, 1121–1130 (2008).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1176/appi.ps.59.10.1121&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=18832497&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000259703500009&link_type=ISI) 

38. 38.Williams, N., Simpson, A.N., Simpson, K. & Nahas, Z. Relapse rates with long-term antidepressant drug therapy: a meta-analysis. Human Psychopharmacology: Clinical and Experimental 24, 401–408 (2009).
    
    

39. 39.Cheng, W. et al. Functional connectivity of the precuneus in unmedicated patients with depression. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 3, 1040–1049 (2018).
    
    

40. 40.Tong, X. et al. Individual deviations from normative electroencephalographic connectivity predict antidepressant response. Journal of Affective Disorders (2024).
    
    

41. 41.Gabbay, V. et al. Striatum-based circuitry of adolescent depression and anhedonia. Journal of the American Academy of Child & Adolescent Psychiatry 52, 628–641. e613 (2013).
    
    

42. 42.Avissar, M. et al. Functional connectivity of the left DLPFC to striatum predicts treatment response of depression to TMS. Brain stimulation 10, 919–925 (2017).
    
    

43. 43.Otte, C., et al. Major depressive disorder. Nature reviews Disease primers 2, 1–20 (2016).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/nrdp.2016.22&link_type=DOI) 

44. 44.First, M.B., Spitzer, R.L., Gibbon, M. & Williams, J.B. (SCID-I/P New York, NY, USA:, 2002).
    
    

45. 45.Veraart, J. et al. Denoising of diffusion MRI using random matrix theory. Neuroimage 142, 394–406 (2016).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2016.08.016&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

46. 46.Tustison, N.J. et al. N4ITK: improved N3 bias correction. IEEE transactions on medical imaging 29, 1310–1320 (2010).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1109/TMI.2010.2046908&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=20378467&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000278535800009&link_type=ISI) 

47. 47.Andersson, J.L.R. & Sotiropoulos, S.N. An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. Neuroimage 125, 1063–1078 (2016).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2015.10.019&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=26481672&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

48. 48.Andersson, J.L.R., Graham, M.S., Zsoldos, E. & Sotiropoulos, S.N. Incorporating outlier detection and replacement into a non-parametric framework for movement and distortion correction of diffusion MR images. Neuroimage 141, 556–572 (2016).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2016.06.058&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

49. 49.Esteban, O. et al. fMRIPrep: a robust preprocessing pipeline for functional MRI. Nature methods 16, 111–116 (2019).
    
    

50. 50.Avants, B.B., Epstein, C.L., Grossman, M. & Gee, J.C. Symmetric diffeomorphic image registration with cross-correlation: evaluating automated labeling of elderly and neurodegenerative brain. Medical image analysis 12, 26–41 (2008).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.media.2007.06.004&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=17659998&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000254032800004&link_type=ISI) 

51. 51.Zhang, Y., Brady, J.M. & Smith, S. in Medical Imaging 2000: Image Processing, Vol. 3979 1126–1137 (SPIE, 2000).
    
    

52. 52.Greve, D.N. & Fischl, B. Accurate and robust brain image alignment using boundary-based registration. Neuroimage 48, 63–72 (2009).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2009.06.060&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19573611&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000269321100010&link_type=ISI) 

53. 53.Pruim, R.H. et al. ICA-AROMA: A robust ICA-based strategy for removing motion artifacts from fMRI data. Neuroimage 112, 267–277 (2015).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2015.02.064&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=25770991&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

54. 54.Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L. & Petersen, S.E. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. Neuroimage 59, 2142–2154 (2012).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2011.10.018&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=22019881&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000299494000017&link_type=ISI) 

55. 55.Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L. & Petersen, S.E. Steps toward optimizing motion artifact removal in functional connectivity MRI; a reply to Carp. Neuroimage 76 (2013).
    
    

56. 56.Yeh, F.C., Wedeen, V.J. & Tseng, W.Y. Generalized q-sampling imaging. IEEE Trans Med Imaging 29, 1626–1635 (2010).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1109/TMI.2010.2045126&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=20304721&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000283222600003&link_type=ISI) 

57. 57.Yeh, F.C., Verstynen, T.D., Wang, Y., Fernandez-Miranda, J.C. & Tseng, W.Y. Deterministic diffusion fiber tracking improved by quantitative anisotropy. PLoS One 8, e80713 (2013).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1371/journal.pone.0080713&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=24348913&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 

58. 58.Yeh, F.C. Shape analysis of the human association pathways. Neuroimage 223, 117329 (2020).
    
    

59. 59.Thomas Yeo, B., et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of neurophysiology 106, 1125–1165 (2011).
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1152/jn.00338.2011&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=21653723&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F04%2F12%2F2024.04.11.24305583.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000294775500007&link_type=ISI) 

60. 60.Weinberg, R. & Patel, Y.C. Simulated intraclass correlation coefficients and their z transforms. Journal of Statistical Computation and Simulation 13, 13–26 (1981).
    
    

61. 61.Fouladi, R.T. & Steiger, J.H. The Fisher transform of the Pearson product moment correlation coefficient and its square: Cumulants, moments, and applications. Communications in Statistics—Simulation and Computation® 37, 928–944 (2008).

 [1]: /embed/graphic-5.gif
 [2]: /embed/inline-graphic-1.gif
 [3]: /embed/inline-graphic-2.gif
 [4]: /embed/graphic-6.gif
 [5]: /embed/inline-graphic-3.gif
 [6]: /embed/inline-graphic-4.gif
 [7]: /embed/inline-graphic-5.gif
 [8]: /embed/inline-graphic-6.gif
 [9]: /embed/inline-graphic-7.gif
 [10]: /embed/inline-graphic-8.gif
 [11]: /embed/graphic-7.gif