Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Biologically Plausible Brain Graph Transformer

Ciyuan Peng, Yuelong Huang, Qichao Dong, Shuo Yu, Feng Xia, Chengqi Zhang, Yaochu Jin

TL;DR

The paper addresses the lack of biological plausibility in brain graph representations produced by existing graph transformers by introducing BioBGT, a Biologically Plausible Brain Graph Transformer. BioBGT employs two key mechanisms: network entanglement based node importance encoding via density matrix spectral entropy to identify hubs, and a functional module aware self-attention guided by a Louvain driven contrastive module extractor and a kernel based attention to preserve functional segregation and integration. The approach yields state of the art performance on ABIDE, ADNI, and ADHD-200 for disease detection tasks and demonstrates improved biological plausibility through correlations with NEff and module aligned attention patterns. This work advances brain graph analysis by integrating principled biologically motivated priors into transformer architectures, with potential impact on digital health and neuroscience research.

Abstract

State-of-the-art brain graph analysis methods fail to fully encode the small-world architecture of brain graphs (accompanied by the presence of hubs and functional modules), and therefore lack biological plausibility to some extent. This limitation hinders their ability to accurately represent the brain's structural and functional properties, thereby restricting the effectiveness of machine learning models in tasks such as brain disorder detection. In this work, we propose a novel Biologically Plausible Brain Graph Transformer (BioBGT) that encodes the small-world architecture inherent in brain graphs. Specifically, we present a network entanglement-based node importance encoding technique that captures the structural importance of nodes in global information propagation during brain graph communication, highlighting the biological properties of the brain structure. Furthermore, we introduce a functional module-aware self-attention to preserve the functional segregation and integration characteristics of brain graphs in the learned representations. Experimental results on three benchmark datasets demonstrate that BioBGT outperforms state-of-the-art models, enhancing biologically plausible brain graph representations for various brain graph analytical tasks

Biologically Plausible Brain Graph Transformer

TL;DR

The paper addresses the lack of biological plausibility in brain graph representations produced by existing graph transformers by introducing BioBGT, a Biologically Plausible Brain Graph Transformer. BioBGT employs two key mechanisms: network entanglement based node importance encoding via density matrix spectral entropy to identify hubs, and a functional module aware self-attention guided by a Louvain driven contrastive module extractor and a kernel based attention to preserve functional segregation and integration. The approach yields state of the art performance on ABIDE, ADNI, and ADHD-200 for disease detection tasks and demonstrates improved biological plausibility through correlations with NEff and module aligned attention patterns. This work advances brain graph analysis by integrating principled biologically motivated priors into transformer architectures, with potential impact on digital health and neuroscience research.

Abstract

State-of-the-art brain graph analysis methods fail to fully encode the small-world architecture of brain graphs (accompanied by the presence of hubs and functional modules), and therefore lack biological plausibility to some extent. This limitation hinders their ability to accurately represent the brain's structural and functional properties, thereby restricting the effectiveness of machine learning models in tasks such as brain disorder detection. In this work, we propose a novel Biologically Plausible Brain Graph Transformer (BioBGT) that encodes the small-world architecture inherent in brain graphs. Specifically, we present a network entanglement-based node importance encoding technique that captures the structural importance of nodes in global information propagation during brain graph communication, highlighting the biological properties of the brain structure. Furthermore, we introduce a functional module-aware self-attention to preserve the functional segregation and integration characteristics of brain graphs in the learned representations. Experimental results on three benchmark datasets demonstrate that BioBGT outperforms state-of-the-art models, enhancing biologically plausible brain graph representations for various brain graph analytical tasks

Paper Structure

This paper contains 33 sections, 3 theorems, 29 equations, 16 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem Definition
  4. Graph Transformers
  5. Biologically Plausible Brain Graph Transformer
  6. Network Entanglement-based Node Importance Encoding
  7. Functional Module-Aware Self-Attention
  8. Community Contrastive Strategy-based Functional Module Extractor
  9. Updated Self-Attention Mechanism
  10. Experiments
  11. Experimental Setup
  12. Results
  13. Ablation Studies
  14. Comparative Analysis of Node Importance Measurement
  15. Biological Plausibility Analysis
  16. ...and 18 more sections

Key Result

Proposition 1

The structural information of a brain graph $G$, including the connection strength between nodes and the degree distribution of nodes, is encoded by its density matrix, which stands as a normalized information diffusion propagator and formulated as $\rho_G= \frac{e^{-\gamma \mathbf{L}}}{Z}$. Here, $

Figures (16)

  • Figure 1: Small-world architecture of brain graphs.
  • Figure 2: Overall framework of BioBGT.
  • Figure 3: Model performance of BioBGT and its altered models.
  • Figure 4: The NE and NEff values of 50 randomly selected nodes from a sample in the ABIDE dataset.
  • Figure 5: The heatmaps of the average self-attention scores. Compared to other methods, heatmap (c) shows that learned attention scores of BioBGT align better with the division of functional modules.
  • ...and 11 more figures

Theorems & Definitions (7)

  • Proposition 1: Density matrix as structural information
  • Definition 1: Node importance degree
  • Theorem 1: Quantification analysis of entanglement
  • Theorem 2: Controllability analysis of functional module-aware self-attention
  • proof
  • proof
  • proof