Statistical power and heritability in whole-genome association studies for quantitative traits

Abstract

The genome-wide association studies (GWAS) have made many advances in identifying the underlying genetics of diseases and quantitative traits. Still, GWAS has major problems, such as the lack of justification for genetic variation and the inability to fully recognize the rare variants. Probably the next generation sequencing (NGS) technology can solve these problems. This study investigated factors affecting statistical power to detect causal variants and heritability in GWAS using sequencing data or a whole-genome association study (WGAS). The WGAS was simulated for quantitative traits with different levels of heritability, frequency of the causal variants, polygenic variance, and linkage disequilibrium (LD). Association studies (using linear and linear mixed models (LMM)) and estimation of heritabilities were conducted and factors affecting statistical power and heritability were determined by analyses of variance. This study indicated that high causal variant frequencies (Linear P = 0.022, LMM P < 0.01) and the high heritability (Linear P < 0.01, LMM P < 0.01) were led to the increase of statistical power for detecting causal variants. Also the polygenic structure with LMM (Linear P = 0.19, LMM P < 0.01) and high LD with both linear model and LMM (Linear P < 0.01, LMM P = 0.04) led to a decrease in statistical power. This study suggests that association study methods for genome sequencing as well as methods for estimating variance components need further development. The LD pattern is an important factor in association studies with sequencing data and needs to be efficiently pruned before estimating variance components. These results can help designing studies and developing methods in association studies.