Extended Data Fig. 6: Characterization of cis-eQTLs across tissues.
(a) Factor weight variance as a function of PEER factors computed up to 75 factors for each of 23 distinct tissues. Factor weight variances become small for most of tissues when the number of inferred hidden PEER factors reaches 10. (b) Pearson’s correlation between inferred factors and known covariates in adipose. The color in each cell denotes −log10(FDR) after the Benjamini-Hochberg correlation of P values (the two-sided Student’s t-test). Only significant correlations (FDR < 0.05) are shown in cells. (c) The proportion of cis-eQTLs with |log2(aFC)| ≤ 1 over all cis-eQTLs as a function of sample size across 23 distinct tissues. |log2(aFC)|, that is, the log2 transformed allelic fold change, which is used to measure the effect size of cis-eQTL. (d) The cis-eQTL cumulative proportion plot of |log2(aFC)| distribution across 9 tissues with variable sample sizes. The arrow indicates tissues in legend were listed from largest to smallest sample size. (e) Distribution of cis-eQTLs around TSS (1 Mb up- and down-stream) in adipose. All gene-variant pairs tested as null; ‘Significant’ indicates the top eQTLs for significant eGenes; ‘Top, Not significant’ indicates the top associated SNP for non-significant genes (non-eGenes). (f) Distribution of cis-eQTLs around the TSS (1 Mb up- and down-stream) across all 23 distinct tissues. (g) Correlation of effect sizes (fastQTL slope) of cis-eQTLs and aFC of matched loci tested by allelic specific expression (ASE) analysis in adipose (Spearman’s rho = 0.75, the two-sided Student’s t-test: P < 2.2 × 10−308) and muscle (Spearman’s rho = 0.68, the two-sided Student’s t-test: P = 2.1 × 10−162). (h) Percentage of cis-eQTLs in the combined muscle data that are replicated in multi-subspecies meta-analysis at different P-value cutoffs used for defining cis-eQTLs. The cis-eQTLs with higher significant levels are more likely to be specifically detected in the combined population. (i) Effects sizes (|log2(aFC)|) of cis-eQTLs specifically detected in combined population are significantly (the two-sided Welch two sample t-test: P = 1.25 × 10−26) smaller than those that are replicated in multi-breed meta-analysis.
