Symmetric Bilinear Regression for Signal Subgraph Estimation.

Area of Science:

• Neuroscience
• Computational Biology
• Network Science

Background:

• Brain connectomics involves analyzing interconnections between brain regions to understand cognitive traits.
• Large-scale brain networks from neuroimaging present challenges for identifying functionally relevant subnetworks.
• Interpreting associations between network predictors and responses requires identifying informative subgraphs.

Purpose of the Study:

• To develop an accurate and efficient method for identifying small, outcome-relevant subgraphs in large brain networks.
• To improve the interpretability of the relationship between brain structure and cognitive traits.
• To discover key interconnected brain regions associated with human cognitive variations.

Main Methods:

• A symmetric bilinear model with L1 penalty was proposed to identify small clique subgraphs.
• A coordinate descent algorithm was employed for model estimation, featuring analytical solutions for conditional convex optimizations.
• The method was applied to human connectome and language comprehension data.

Main Results:

• The method successfully identified relevant interconnections within small sets of brain regions.
• The proposed approach demonstrated superior predictive performance compared to existing methods.
• Key interconnected brain regions related to language comprehension were discovered.

Conclusions:

• The developed method is effective for discovering small, informative subgraphs in complex brain networks.
• This approach enhances the understanding of the neural mechanisms underlying cognitive traits.
• The findings offer valuable insights into brain connectomics and predictive modeling.