Figure 5  Predicting Rare and Common eQTLs
(A) Utility of diverse noncoding annotation for predicting rare and common eQTLs. We considered the enrichment in eQTLs (measured with the π1 statistic) for rare (MAF < 0.01) and common (MAF > 0.01) variants overlapping the following different functional annotations: ENCODE TF binding and DNase I hypersensitivity peaks, distance to TSS, PhyloP conservation scores, and motif disruption (score change > 10); annotations were added one at a time. We found that these functional annotations were significantly more powerful for detecting an eQTL when intersecting rare variants rather than common variants. Furthermore, on the right, we demonstrate that none of the genes possessing rare variants overlapping the different categories of annotation were disproportionally enriched in their ability to also be eQTLs in the population. A full matrix summarizing intersections of these annotations is provided in Table S7.
(B) Conservation scores and allele frequency predict genes with an eQTL. We restricted to variants within 100 kb of the TSS, within ENCODE TF binding and DNase I hypersensitivity peaks, and with different PhyloP scores and allele frequencies to assess each variant class’s enrichment in eQTLs. We observed that highly conserved and rare variants were strongly predictive of an eQTL.
(C) Conservation scores, the distance to splice site, and allele frequency predict genes with a sQTL. We considered different thresholds on distance to splice sites, PhyloP conservation scores, and allele frequencies. We observed that rare and conserved variants near splice sites (light blue) were highly predictive of a sQTL.