Interpretable network-guided epistasis detection

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Epistatic interactions between genes are vital for complex disease etiology.
• Genome-wide association studies (GWAS) face statistical challenges in detecting these interactions.
• Current methods struggle with mapping gene interactions to single-nucleotide polymorphism (SNP) interaction hypotheses.

Purpose of the Study:

• To develop and evaluate a multi-step protocol for epistasis detection using gene-gene co-function networks.
• To reduce the number of statistical tests and control for type I errors in GWAS.
• To provide interpretable gene interaction insights for complex diseases.

Main Methods:

• A network-based approach utilizing gene-gene co-function networks.
• Comparison of three SNP-gene mapping strategies: positional overlap, expression quantitative trait loci (eQTLs), and 3D structure proximity.
• Application of the adaptive truncated product method for non-parametric gene pair score computation.

Main Results:

• The protocol was applied to a GWAS dataset for inflammatory bowel disease (IBD).
• Different protocol configurations yielded distinct results, suggesting diverse IBD mechanisms.
• Results showed overlap with known IBD characteristics and outperformed conventional non-network approaches.

Conclusions:

• Incorporating prior biological knowledge via networks enhances epistasis detection in GWAS.
• The proposed method offers a powerful strategy for discovering gene interactions underlying complex diseases.
• This network-guided approach has the potential for significant additional discoveries in genetic association studies.