YExp represents the vector of normalized expression estimates for a particular gene across individuals. µ is the mean gene expression level across all samples. XExp is an indicator variable identifying the class (high or low lymphocyte count) to which the sample belongs, while β is the fixed effect. ɛ is an error term assumed to be normally distributed with mean 0 and variance . To test for the effect of class membership on gene expression levels, we compared the null model where all samples are considered to be from the same class to an alternative where samples can be from one of two classes (see Supplementary Material, Fig. S6 for the distribution of P-values for the likelihood ratio test). Because the models are nested, we assumed that the differences in log-likelihood between the two models are χ2-distributed with 1 degree of freedom. The q values were estimated for P-values using the method of Storey and Tibshirani (76).
