Performance of epistasis detection methods in semi-simulated GWAS

Area of Science:

• Genetics
• Statistical genetics
• Computational biology

Background:

• Missing heritability in Genome Wide Association Studies (GWAS) is partly attributed to genetic interactions (epistasis).
• Existing statistical methods for epistasis detection face challenges including computational demands, complex linkage disequilibrium (LD), and definition inconsistencies.
• Limited experimental data hinders validation of statistical method performance in realistic GWAS settings.

Purpose of the Study:

• To introduce a simulation pipeline for generating large-scale GWAS data with epistasis and realistic LD.
• To evaluate the performance of five exhaustive bivariate interaction methods: fastepi, GBOOST, SHEsisEpi, DSS, and IndOR.
• To compare these methods across 234 disease scenarios, assessing false positive rate, power, AUC, and computation time.

Main Methods:

• Development of a simulation pipeline for realistic GWAS data generation, incorporating epistasis and LD.
• Exhaustive bivariate analysis of five interaction detection methods: fastepi, GBOOST, SHEsisEpi, DSS, and IndOR.
• Performance evaluation using extensive simulations across 234 disease scenarios, including analysis on a real type 2 diabetes GWAS dataset.

Main Results:

• GBOOST, SHEsisEpi, and DSS demonstrated satisfactory control of false positive rates.
• fastepi and IndOR showed increased false positive rates with LD between causal SNPs.
• DSS exhibited superior power and AUC in scenarios with no or weak LD between causal SNPs.
• All evaluated methods achieved exhaustive GWAS analysis (6.10^5 SNPs, 15,000 samples) within hours using GPU acceleration.

Conclusions:

• Computational time is no longer a barrier for exhaustive epistasis searches in large GWAS.
• Utilizing DSS on SNP pairs with limited LD is recommended for optimal statistical performance.
• A combined approach using DSS and GBOOST shows promise, as evidenced by detecting distinct epistatic genes in the WTCCC dataset.
• Weak epistasis among common variants is becoming detectable with current methods as GWAS sample sizes increase.