An inverse normal transformation was applied to the results for each transcript to avoid the effect of outliers. Briefly, the procedure involves first transforming all observations to ranks and then converting these ranks to deviates from a standard normal distribution. Narrow-sense heritability for each transcript was estimated by using a variance component model, and a variance component–based score test was used to evaluate the evidence for association at each SNP (Chen and Abecasis 2007). This variance component–based association analysis results in an estimate of the additive genetic effect at each SNP and accounts for the correlation in phenotypes between siblings. Both procedures are implemented in MERLIN (Abecasis et al. 2002; Abecasis and Wigginton 2005).
