Wavelet Screaming: a novel approach to analyzing GWAS data
We present an alternative method for genome-wide association studies (GWAS) that is more powerful than the regular GWAS method for locus detection. The regular GWAS method suffers from a substantial multiple-testing burden because of the millions of single nucleotide polymorphisms (SNPs) being tested simultaneously. Furthermore, it does not consider the functional genetic effect on the response variable; i.e., it ignores more complex joint effects of nearby SNPs within a region. Our proposed method screens the entire genome for associations using a sequential sliding-window approach based on wavelets. A sequence of SNPs represents a genetic signal, and for every screened region, we transform the genetic signal into the wavelet space. We then estimate the proportion of wavelet coefficients associated with the phenotype at different scales. The significance of a region is assessed via simulations, taking advantage of a recent result on Bayes factor distributions. Our new approach reduces the number of independent tests to be performed. Moreover, we show via simulations that the Wavelet Screaming method provides a substantial gain in power compared to the classic GWAS modeling when faced with more complex signals than just single-SNP associations. To demonstrate feasibility, we re-analyze data from the large Norwegian HARVEST cohort. Keywords: Bayes factors, GWAS, SNP, Multiple testing, Polygenic
READ FULL TEXT