GenoGraph: An Interpretable Graph Contrastive Learning Approach for Identifying Breast Cancer Risk Variants.

Area of Science:

• Genetics
• Computational Biology
• Machine Learning

Background:

• Genome-wide association studies (GWASs) have identified numerous breast cancer-associated genetic variants.
• Conventional GWAS methods often fail to capture complex genetic interactions crucial for disease susceptibility.
• Machine learning (ML) and deep learning (DL) offer alternatives but face challenges like overfitting and limited interpretability in high-dimensional genetic data.

Purpose of the Study:

• To introduce GenoGraph, a novel graph-based contrastive learning framework.
• To address limitations of existing methods in modeling high-dimensional genetic data, especially in low-sample-size scenarios.
• To enhance breast cancer risk prediction and discover population-specific genetic interactions.

Main Methods:

• Developed GenoGraph, a graph-based contrastive learning framework.
• Applied GenoGraph to model high-dimensional genetic data for breast cancer case-control classification.
• Utilized the Biobank of Eastern Finland dataset for validation.

Main Results:

• GenoGraph achieved a high accuracy of 0.96 in breast cancer classification.
• Identified a key risk variant (rs11672773) in the Finnish population.
• Discovered significant interactions between rs11672773, rs10759243, and rs3803662, with confirmed biological relevance.

Conclusions:

• GenoGraph effectively models complex genetic interactions for improved breast cancer risk prediction.
• The framework shows promise for identifying population-specific genetic risk factors and interactions.
• Findings support the potential of GenoGraph for advancing personalized medicine in oncology.