Evaluating Genetic Associations with Traits

We performed a burden test and a nonburden sequence kernel association test (SKAT) to assess the cumulative effect of all variants within one kilobase of each annotated gene. The weighted burden test weights the contribution of each variant in a gene by the reciprocal of the standard deviation of its estimated minor allele frequency and uses the weighted averages to estimate a score statistic (Madsen and Browning 2009 (link); Han and Pan 2010 (link)). The SKAT kernel function builds a relationship matrix detailing relatedness of individuals based upon all variants within a gene. This relationship matrix is fit as the covariance matrix of a random effect in a linear mixed model framework and used to estimate a variance component score to discern the significance of a trait association (Wu et al. 2011 (link)). The SKAT kernel function used was linear and did not up-weight the relative contribution of minor alleles.
We performed both the weighted burden test and SKAT using the SKAT package (Wu et al. 2011 (link)) in R v3.0.1 (R Development Core Team 2013 ). For both methods, male and female starvation resistance and genome size were fit with an identity link function and fixed effect covariates for Wolbachia infection status, major inversions, and the 11 principal components explaining the most genetic variation in the DGRP (Tracy-Winom P-value < 0.01). Wolbachia infection status was fit with a logit link function in a likewise manner, excluding the fixed effect of Wolbachia infection status. We performed gene-based tests for all variants, and for common (MAF ≥ 0.05) and rare (MAF < 0.05) variants separately.