For each probeset, surrogate variable analysis (SVA) [20] was performed on the matrix of expression measurements, after controlling for the effects of hybridization protocol, age, sex, and a principal component analysis based quantification of genetic ancestry. For each probeset, we then constructed a linear mixed effects model y ∼ m + P + A + C + R + I + W + SVi..n + e, where y is the log2 transformed probe intensity, m is the expected probe intensity, P is a factor controlling for the effect of subtle variations in hybridization protocol (e.g., the identity of the technician who performed the experiment), A is the effect of individual age, and C is the effect of individual sex, and R is the effect of genetic ancestry. I is the random effect of each individual, W is the random effect of the oligonucleotide probe, SVi..n represents the effects of a matrix of 55 surrogate variables, and e is the residual error. The model was fitted to each gene by residual maximum likelihood using the lmer function in the R package
