EPISFA

EPISFA detects gene-gene (G × G) interactions in family-based genetic datasets using unsupervised Epistasis Sparse Factor Analysis to help explain heritability not accounted for by single-nucleotide polymorphisms (SNPs).

Key Features:

• Epistasis Sparse Factor Analysis: Uses unsupervised sparse factor analysis to screen for gene-gene interactions in high-dimensional genetic data.
• EPISFA-LD (Epistasis Sparse Factor Analysis for Linkage Disequilibrium): Provides a variant that accounts for linkage disequilibrium when detecting interactions.
• Family-based focus: Tailored to family-based datasets and leverages familial structure to mitigate biases from population stratification.
• High-dimensional screening: Performs efficient screening of large SNP datasets for epistatic signals.
• Validation and power: Demonstrates high power in simulations and shows favorable performance compared with Family-Based Multifactor Dimensionality Reduction (FAM-MDR).
• Real-data discoveries: Identified five significant G × G pairs in the Fangshan/family-based Ischemic Stroke Study in China, with three corroborated by FAM-MDR and logistic regression and two uniquely detected by EPISFA/EPISFA-LD.
• Addresses missing heritability: Targets interaction signals that may explain the gap between estimated heritability and variance explained by identified SNPs.

Scientific Applications:

• Epistatic architecture of complex disease: Identify gene-gene interactions that contribute to missing heritability in complex diseases.
• Family-based cohort analysis: Analyze G × G interactions in family-based cohorts, including the Fangshan/family-based Ischemic Stroke Study in China.
• LD-aware interaction detection: Detect interactions in datasets with linkage disequilibrium using EPISFA-LD.
• Comparative validation: Compare and validate epistatic signals against Family-Based Multifactor Dimensionality Reduction (FAM-MDR) and logistic regression analyses.

Methodology:

Implements unsupervised machine learning via sparse factor analysis, includes an LD-aware variant (EPISFA-LD), performs efficient screening of high-dimensional SNP datasets, and uses simulations plus comparisons with FAM-MDR and logistic regression for validation.