The heritability of gene expression defines an upper bound to how well we can predict the trait. We estimated the narrow-sense heritability for each gene using a variance component model with a genetic relationship matrix (GRM) estimated from genotype data, as implemented in GCTA20. No pair of subjects from the 922 individuals in DGN shared genetic relatedness (π̂) in excess of 5% and thus all were included in the narrow-sense heritability estimation. SNPs in the vicinity of each gene (within 1Mb of gene start or end, as defined by the GENCODE45 version 12 gene annotation), with MAF > 0.05, and in Hardy-Weinberg Equilibrium (P > 0.05) were used to construct the GRM for each gene. We calculated the proportion of the variance of gene expression explained by these local SNPs using the following mixed-effects model37: Y=Xb+Glocal+e var(Y)=Alocalσlocal2+Iσe2 where Y is a gene expression trait and b a vector of fixed effects. Here Alocal is the GRM calculated from the local SNPs, and (the random effect) Glocal denotes the genetic effect attributable to the set of local SNPs with var (Glocal)=Alocalσlocal2
