eQTL mapping using allele-specific count data is computationally feasible, powerful, and provides individual-specific estimates of genetic effects

Abstract

Using information from allele-specific gene expression (ASE) can improve the power to map gene expression quantitative trait loci (eQTLs). However, such practice has been limited, partly due to computational challenges and lack of clarification on the size of power gain or new findings besides improved power. We have developed geoP, a computationally efficient method to estimate permutation p-values, which makes it computationally feasible to perform eQTL mapping with ASE counts for large cohorts. We have applied geoP to map eQTLs in 28 human tissues using the data from the Genotype-Tissue Expression (GTEx) project. We demonstrate that using ASE data not only substantially improve the power to detect eQTLs, but also allow us to quantify individual-specific genetic effects, which can be used to study the variation of eQTL effect sizes with respect to other covariates. We also compared two popular methods for eQTL mapping with ASE: TReCASE and RASQUAL. TReCASE is ten times or more faster than RASQUAL and it provides more robust type I error control.

A computational method for Cell type-specific Expression Quantitative Trait Loci mapping using bulk RNA-seq data

“…When performing bulk eQTL, CSeQTL is equivalent to the TReCASE method [28] that combines total expression and allele-specific expression to improve the power of eQTL mapping. Consistent with our previous results [11, 28], CSeQTL has much higher power than OLS for bulk eQTL mapping. For example, considering the results for CMC schizophrenia samples (n=250), for bulk eQTL mapping, CSeQTL and OLS identified around 10,000 and 3,000 eGenes (genes with at least one significant eQTL) respectively.…”

“…A brute-force implementation, which requires permuting the data many times and running CSeQTL on each permuted dataset, is computationally prohibitive. Instead, we employed a computationally efficient method named geoP [11, 27] that calculates the permutation p-value by estimating the effect number of independent tests. To account for multiple testing across genes, we selected a permutation p-value cutoff to control the false discovery rate (FDR).…”

“…We derived these 21 categories by grouping similar traits to study whether cell type-specific eQTLs were associated with some but not all traits. Calculations of GWAS enrichment and confidence intervals followed the procedure presented in Vasyl et al [11].…”

“…Most eQTL studies use RNA-seq data from bulk samples and ignore the heterogeneity of cell type compositions. Gene expression variation due to cell type composition may be partially captured by principal components or latent factors such as the probabilistic estimation of expression residuals (PEER) factors [10,11]. A few studies have conducted cell type-specific eQTL mapping using bulk tissue gene expression by testing the interaction between genotype and cell type proportion in a linear model [3,4,9,[12][13][14][15][16].…”

Detecting isoform-level allelic imbalance accounting for inferential uncertainty

“…It may also be of interest to detect in which cell types the allelic signal may be strongest or exclusively present, as has been investigated in recent methods for single cell allelic expression or accessibility datasets [78–80]. Finally, we note that a number of methods have shown that AI can be effectively integrated with total expression across individuals of all three genotypes [5, 81–84, 13]. This approach uses more information, and so should produce a gain in sensitivity, as well as extending beyond genes harboring exonic variants, which is a limitation for AI-based methods.…”