Non-parametric GWAS: Another View on Genome-wide Association Study

Abstract

Genome-wide association study (GWAS) is a fundamental step for understanding the genetic link to traits (phenotypes) of interest, such as disease, BMI and height. Typically, GWAS estimates the effect of SNP on the phenotype using a linear model by coding SNP as working code, {0, 1, 2}, according to the minor allele frequency. Looking inside the linear model, we find that the coding strategy of SNP plays a key role in detecting SNPs contributed to the phenotype. Specifically, a partial mismatch between the order of the working code and that of the underlying true code will lead to false negatives, which has been ignored for a long time. Motivated by this phenomenon, we propose an indicator of possible false negatives and several non-parametric GWAS methods independent of coding strategy. Results from both simulations and real data analysis show the advantages of new methods in identifying significant loci, indicating their important complementary role in GWAS.