Learning individual-specific brain-behavior manifolds via latent space alignment
Poster No:
1087
Submission Type:
Abstract Submission
Authors:
Yingjie Peng1, Ang Li1, Xiaohan Tian2, Shangzheng Huang1, Tongyu Zhang1
Institutions:
1State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences, Beijing, China, 2beijing normal university, Beijing, China
First Author:
Yingjie Peng
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
Co-Author(s):
Ang Li
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
Shangzheng Huang
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
Tongyu Zhang
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
State Key Lab of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences
Beijing, China
Introduction:
Establishing reliable brain-behavior relationships is crucial for advancing precision psychiatry and personalized medicine. While univariate analyses have identified specific brain-behavior correlations, and multivariate techniques like canonical correlation analysis (CCA) have revealed distributed patterns, several limitations still persist (Vieira et al., 2024): First, these methods assume linear relationships between neural activity and behavior, while real-world brain-behavior associations likely involve complex non-linear interactions across multiple brain regions. Second, they tend to identify correlational patterns without considering potential causal relationships between brain activity and behavioral outcomes. Furthermore, their focus on group-level patterns through averaging may overlook crucial individual differences that are particularly relevant for clinical applications. Recent advances in deep learning, especially (Generative Adversarial Networks) GANs and contrastive learning, offer promising solutions for modeling non-linear brain-behavior relationships while maintaining biological interpretability (Aglinskas et al., 2022).
Methods:
We developed a CycleGAN framework utilizing the UK Biobank data (N=10,000; 70% training, 10% validation, 20% testing). Regional gray matter volumes from T1 MRI were used as imaging features. Behavioral data included cognitive demographic, and social domains.The model combined reconstruction (self, cross-domain, cycle), feature alignment (contrastive, MMD), and interpretability (structural consistency, orthogonalization) losses, optimized via validation. All loss components were weighted through validation performance for optimal results. The study has been conducted under the UK Biobank application ID 85139.
Results:
Our model successfully learned a unified latent space that preserves both neuroanatomical organization and behavioral information. T-SNE visualization revealed clear anatomical coherence in brain-derived embeddings (Fig. 1a) and distinct behavioral clustering in behavior-informed embeddings (Fig. 1b), with systematic variations by sex (Fig. 1d) and gray matter volume (Fig. 1e). The model outperformed canonical correlation analysis in behavioral prediction across multiple domains, including fluid intelligence and household size (Fig. 1c, f, g), demonstrating robust performance in cross-validation experiments.
Analysis identified two opposing brain-behavior modes with distinct cortical distributions (Fig. 2a). Mode 1, characterized by negative loadings in prefrontal regions and positive loadings in posterior temporal and parietal regions, showed positive correlations with age-related variables (age: r=0.22, P<0.05; reaction time: r=0.18, P<0.05) and negative correlations with gender (r=-0.58, P<0.05) (Fig. 2b). Mode 2 exhibited an inverse pattern of cortical distributions and demonstrated significant positive associations with serotonin receptor distribution (5-HT2a, r=0.24, P<0.05, validated through 100 null model permutations). This mode maintained significant correlations with fluid intelligence r=0.06, P<0.01) and social behavior (r=0.06, p<0.01) after controlling for age and gender.
Using a 63-year-old subject, the model's predictions of age-dependent cortical patterns validated its ability to capture established aging patterns, particularly the progressive alterations in frontal and temporal regions (Fig. 2c).
Analysis identified two opposing brain-behavior modes with distinct cortical distributions (Fig. 2a). Mode 1, characterized by negative loadings in prefrontal regions and positive loadings in posterior temporal and parietal regions, showed positive correlations with age-related variables (age: r=0.22, P<0.05; reaction time: r=0.18, P<0.05) and negative correlations with gender (r=-0.58, P<0.05) (Fig. 2b). Mode 2 exhibited an inverse pattern of cortical distributions and demonstrated significant positive associations with serotonin receptor distribution (5-HT2a, r=0.24, P<0.05, validated through 100 null model permutations). This mode maintained significant correlations with fluid intelligence r=0.06, P<0.01) and social behavior (r=0.06, p<0.01) after controlling for age and gender.
Using a 63-year-old subject, the model's predictions of age-dependent cortical patterns validated its ability to capture established aging patterns, particularly the progressive alterations in frontal and temporal regions (Fig. 2c).
Conclusions:
Our model demonstrates improved predictive performance in modeling brain-behavior relationships compared to traditional CCA. Our preliminary experiments revealed two distinct modes with specific anatomical distributions - Mode 1 reflecting age-related structural variations in frontal-temporal regions, and Mode 2 showing associations with serotonin receptor distribution and cognitive measures. Additionally, our framework shows potential utility in generating individual-level brain patterns, suggesting possible applications in investigating personalized aging trajectories.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural)
Methods Development
Multivariate Approaches 2
Keywords:
ADULTS
Aging
Cognition
Computational Neuroscience
Computing
Cortex
Machine Learning
Neurotransmitter
NORMAL HUMAN
Structures
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Aglinskas, A. (2022). Contrastive machine learning reveals the structure of neuroanatomical variation within autism. Science, 376(6597), 1070–1074.