Genotyped and imputed variants as described above were phased using SHAPEIT2 (Delaneau et al., 2011) for allele-specific expression (ASE) analyses. Haplotype phasing was performed based on reference haplotypes in the 1000 Genomes Project multi-ethnic panel (Phase 3) as described above for the genotype imputation. All variants without high phasing probability or missing 1000 Genomes references were excluded from ASE analyses. For allele-specific RNA-seq quantification, we used hg19/GRCh37 genome-based alignment of STAR for the second pass in the 2-pass mode (Dobin et al., 2013). Reference-mapping bias and potential PCR duplicates were removed with WASP (van de Geijn et al., 2015). Allele-specific read counts were quantified using SAMtools mpileup (Li et al., 2009). Single-cell-type ASE was quantified using QuASAR (Harvey et al., 2015), which quantifies ASE with a beta-binomial model, and differential ASE was quantified with Fisher’s exact test. False discovery rates were estimated using the method of Benjamini and Hochberg (1995).
