Marker-Trait Complete Analysis

A recurring problem in genomics involves testing association of one or more traits of interest to multiple genomic features. Feature-trait squared correlations r2 are commonly-used statistics, sensitive to trend associations. It is often of interest to perform testing across collections {r2} over markers and/or traits using both maxima and sums. However, both trait-trait correlations and marker-marker correlations may be strong and must be considered. The primary tools for multiple testing suffer from various shortcomings, including p-value inaccuracies due to asymptotic methods that may not be applicable. Moreover, there is a lack of general tools for fast screening and follow-up of regions of interest.To address these difficulties, we propose the MTCA approach, for Marker-Trait Complete Analysis. MTCA encompasses a large number of existing approaches, and provides accurate p-values over markers and traits for maxima and sums of r2 statistics. MTCA uses the conditional inference implicit in permutation as a motivational frame-work, but provides an option for fast screening with two novel tools: (i) a multivariate-normal approximation for the max statistic, and (ii) the concept of eigenvalue-conditional moments for the sum statistic. We provide examples for gene-based association testing of a continuous phenotype and cis-eQTL analysis, but MTCA can be applied in a much wider variety of settings and platforms.