Iterative Usage of Fixed and Random Effect Models for Powerful and Efficient Genome-Wide Association Studies

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genomics

Background:

• Genome-Wide Association Studies (GWAS) face challenges with false positives and compromised true positives when controlling for population structure and kinship using Mixed Linear Models (MLM).
• Existing methods like Multiple Loci Linear Mixed Model (MLMM) partially address confounding but do not completely eliminate it.

Purpose of the Study:

• To develop a novel statistical method that completely eliminates confounding in GWAS association tests.
• To improve statistical power and detection of true positives in GWAS.
• To enhance computational efficiency for large-scale genetic datasets.

Main Methods:

• Introduced FarmCPU (Fixed and random model Circulating Probability Unification), an iterative method combining Fixed Effect Model (FEM) and Random Effect Model (REM).
• FEM tests markers one at a time with associated markers as covariates to control false positives.
• REM estimates associated markers to define kinship, preventing model over-fitting and unifying P-values iteratively.

Main Results:

• FarmCPU demonstrated improved statistical power compared to existing GWAS methods in both real and simulated data analyses.
• The method effectively controls false positives while increasing the detection of true positives.
• Achieved linear computational time complexity with respect to the number of individuals and markers.

Conclusions:

• FarmCPU offers a robust and efficient solution for GWAS, enhancing the accuracy of genetic association studies.
• The method's computational efficiency enables analysis of very large datasets (e.g., half a million individuals and markers within three days).
• FarmCPU represents a significant advancement in statistical genomics for identifying genetic variants associated with traits.