Quantification of Gene Expression, Splicing, and ASE
We used Tophat and Cufflinks to obtain gene-expression levels from RNA-seq. We used Tophat to map RNA reads to the human reference genome (UCSC Genome Browser, hg19) and Cufflinks to quantify transcript-expression levels. Gene-expression levels were the sum of transcript-expression levels. Gencode28 v.12 was used as the input annotation for Cufflinks. We calculated transcript ratios to quantify alternative splicing patterns. Gene-expression and transcript-ratio data for Geuvadis samples were downloaded from the Geuvadis website; we used quantified gene-level reads per kilobase per million both before (for assessing effect sizes) and after (for eQTL mapping) normalization via probabilistic estimation of expression residuals.29 We assessed ASE by counting RNA read depth at heterozygous sites. We performed multiple quality-control steps to reduce known technical artifacts (see Figure S2). We obtained read counts at each heterozygous site by using SAMtools30 mpileup and our own ASE pipeline based on a binomial test modified for reference-mapping bias with a filter for observing at least five reads for each allele and a minimum read depth of 20× per site.21,31 To assess the quality of ASE estimates, we compared ASE correlation between double-IBD (identical-by-descent) siblings, half-IBD siblings, and non-IBD siblings. Indeed, we observed an expected increase in correlation between degree of IBD and allelic ratio measured across all sites (Figures S28–S30).