unused factors, and Pickrell et al. (2010) chooses the number of PCs that gives the largest number of eQTLs. Our approach selects the number of PCs that gives the largest number of local-eQTLs (defined as a SNP within 1 Mb of a gene). We show above (Fig. 2) that for a given P-value threshold, local-eQTL are more reliable than distant-eQTL. Consequently local-eQTL should give a more reliable estimate of statistical power and a better indicator of the number of PCs to adjust for unmeasured confounders. Unfortunately, we have not found SVA and VBQTL to be computationally feasible for genome-wide eQTL mapping, because in practice, SVA will ignore SNPs in order to estimate significant surrogate variables and the fast estimates of VBQTL (fVBQTL) required for genome-wide analyses will also ignore SNPs to estimate nonzero factors.
