Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

KOCOBrain: Kuramoto-Guided Graph Network for Uncovering Structure-Function Coupling in Adolescent Prenatal Drug Exposure

Badhan Mazumder, Lei Wu, Sir-Lord Wiafe, Vince D. Calhoun, Dong Hye Ye

TL;DR

KOCOBrain addresses the challenge of discovering structure-function signatures of prenatal drug exposure by unifying structural and functional connectomes through a differentiable Kuramoto-guided coupling mechanism and cognition-aware graph attention. The key innovations are a Kuramoto-based phase module constrained by anatomy, phase-informed graph tokenization, cognition-gated MoE attention, and a joint objective that includes reconstruction and contrastive components to improve robustness under class imbalance. Empirical results on the ABCD cohort show that KOCOBrain achieves higher accuracy, sensitivity, and specificity than state-of-the-art baselines, with ablations confirming the necessity of dynamic coupling and cognition-aware routing. The approach yields interpretable patterns of disrupted network coordination linked to PDE and offers a scalable framework for exploring developmentally relevant structure-function dynamics in large multisite datasets.

Abstract

Exposure to psychoactive substances during pregnancy, such as cannabis, can disrupt neurodevelopment and alter large-scale brain networks, yet identifying their neural signatures remains challenging. We introduced KOCOBrain: KuramotO COupled Brain Graph Network; a unified graph neural network framework that integrates structural and functional connectomes via Kuramoto-based phase dynamics and cognition-aware attention. The Kuramoto layer models neural synchronization over anatomical connections, generating phase-informed embeddings that capture structure-function coupling, while cognitive scores modulate information routing in a subject-specific manner followed by a joint objective enhancing robustness under class imbalance scenario. Applied to the ABCD cohort, KOCOBrain improved prenatal drug exposure prediction over relevant baselines and revealed interpretable structure-function patterns that reflect disrupted brain network coordination associated with early exposure.

KOCOBrain: Kuramoto-Guided Graph Network for Uncovering Structure-Function Coupling in Adolescent Prenatal Drug Exposure

TL;DR

KOCOBrain addresses the challenge of discovering structure-function signatures of prenatal drug exposure by unifying structural and functional connectomes through a differentiable Kuramoto-guided coupling mechanism and cognition-aware graph attention. The key innovations are a Kuramoto-based phase module constrained by anatomy, phase-informed graph tokenization, cognition-gated MoE attention, and a joint objective that includes reconstruction and contrastive components to improve robustness under class imbalance. Empirical results on the ABCD cohort show that KOCOBrain achieves higher accuracy, sensitivity, and specificity than state-of-the-art baselines, with ablations confirming the necessity of dynamic coupling and cognition-aware routing. The approach yields interpretable patterns of disrupted network coordination linked to PDE and offers a scalable framework for exploring developmentally relevant structure-function dynamics in large multisite datasets.

Abstract

Exposure to psychoactive substances during pregnancy, such as cannabis, can disrupt neurodevelopment and alter large-scale brain networks, yet identifying their neural signatures remains challenging. We introduced KOCOBrain: KuramotO COupled Brain Graph Network; a unified graph neural network framework that integrates structural and functional connectomes via Kuramoto-based phase dynamics and cognition-aware attention. The Kuramoto layer models neural synchronization over anatomical connections, generating phase-informed embeddings that capture structure-function coupling, while cognitive scores modulate information routing in a subject-specific manner followed by a joint objective enhancing robustness under class imbalance scenario. Applied to the ABCD cohort, KOCOBrain improved prenatal drug exposure prediction over relevant baselines and revealed interpretable structure-function patterns that reflect disrupted brain network coordination associated with early exposure.
Paper Structure (14 sections, 5 equations, 2 figures, 2 tables)

This paper contains 14 sections, 5 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Kuramoto-Guided Coupling & Graph Tokenization
  4. Kuramoto Dynamics for Structure--Function Coupling
  5. Phase-Informed Graph Tokenization
  6. Cognition-Gated Phase-Locked Graph Attention
  7. Joint Training Objective
  8. Results and Discussion
  9. Dataset and Pre-processing
  10. Experimental Settings
  11. Quantitative Evaluation
  12. Qualitative Evaluation
  13. Conclusions
  14. Compliance with Ethical Standards

Figures (2)

  • Figure 1: KOCOBrain overview: SC from DTI and FNC from rs-fMRI were jointly modeled via a Kuramoto-guided coupling layer that simulates oscillatory phase synchronization over the anatomical wiring. The resulting phase dynamics yielded structure-informed functional representations, which combined with local structural features formed phase-aware node tokens. These tokens passed through cognition-gated, phase-locked graph-transformer layers where cognitive scores modulated query-key interactions via low-rank MoE adapters, and attention weights were biased by phase synchrony and SC strength. Global pooling of the refined embeddings produced subject-level representations optimized end-to-end for PDE prediction.
  • Figure 2: Axial depiction of group-mean attention maps highlighting the top 3% connections for exposed and unexposed groups across NeuroMark networks: subcortical (SCN), auditory (ADN), sensorimotor (SMN), visual (VSN), cognitive control (CON), default mode (DMN), and cerebellar (CBN) network. Intra-network links are color-coded, while inter-network edges appear in gray, with line thickness scaled to attention strength to indicate relative connection importance.