CoMM-S²: a collaborative mixed model using summary statistics in transcriptome-wide association studies

Abstract

Motivation: Although genome-wide association studies (GWAS) have deepened our understanding of the genetic architecture of complex traits, the mechanistic links that underlie how genetic variants cause complex traits remains elusive. To advance our understanding of the underlying mechanistic links, various consortia have collected a vast volume of genomic data that enable us to investigate the role that genetic variants play in gene expression regulation. Recently, a collaborative mixed model (CoMM) [42] was proposed to jointly interrogate genome on complex traits by integrating both the GWAS dataset and the expression quantitative trait loci (eQTL) dataset. Although CoMM is a powerful approach that leverages regulatory information while accounting for the uncertainty in using an eQTL dataset, it requires individual-level GWAS data and cannot fully make use of widely available GWAS summary statistics. Therefore, statistically efficient methods that leverages transcriptome information using only summary statistics information from GWAS data are required. Results: In this study, we propose a novel probabilistic model, CoMM-S 2 , to examine the mechanistic role that genetic variants play, by using only GWAS summary statistics instead of individual-level GWAS data. Similar to CoMM which uses individual-level GWAS data, CoMM-S 2 combines two models: the first model examines the relationship between gene expression and genotype, while the second model examines the relationship between the phenotype and the predicted gene expression from the first model. Distinct from CoMM, CoMM-S 2 requires only GWAS summary statistics. Using both simulation studies and real data analysis, we demonstrate that even though CoMM-S 2 utilizes GWAS summary statistics, it has comparable performance as CoMM, which uses individual-level GWAS data.

A tissue-specific collaborative mixed model for jointly analyzing multiple tissues in transcriptome-wide association studies

“…To demonstrate the utility of the TisCoMM-S 2 tissue-specific test, we applied the tissue-specific test to all identified 92 candidate genes of LOAD and 200 candidate genes of asthma by using the TisCoMM-S 2 joint test, and compared analysis results with those from CoMM [10, 11]. Table 3 shows the distributions of identified tissues with which candidate genes are associated in LOAD and asthma, respectively (see details in Supplementary Tables S6 and S7).…”

“…Transcriptome-wide association studies (TWAS) has been widely used to integrate the expression regulatory information from these eQTL studies with GWAS to prioritize genome-wide trait-associated genes [7, 8, 9]. A variety of TWAS methods have been proposed using different prediction models for expression imputation, including the parametric imputation models, e.g., PrediXcan [7], TWAS [8], CoMM [10] and CoMM-S 2 [11], and the nonparametric imputation model, e.g., Tigar [12]. These methods have been used for analyzing many complex traits with expression profiles from different tissues, successfully enhancing the discovery of genetic risk loci for complex traits [13, 9].…”

“…Second, both MultiXcan and UTMOST rely on a step-wise inference framework, ignoring the uncertainty in the process of expression imputation and thus losing power, especially when cellular-heritability is small [10]. Recently, CoMM [10] and its variant for summary-level data, CoMM-S 2 [11], have been proposed to account for uncertainty in the process of expression imputation. Third, MultiXcan and UTMOST do not make efficient use of the shared patterns of eQTLs across tissues, where MultiXcan uses principal component analysis (PCA) regularization on the predicted expression data, and UTMOST uses penalized regularization on coefficients for eQTL effects.…”

MR-LDP: a two-sample Mendelian randomization for GWAS summary statistics accounting for linkage disequilibrium and horizontal pleiotropy

“…We can easily formulate the problem (5) as a statistical test for the null hypothesis that the health risk factor is not associated with the disease of interest, or . Testing this hypothesis requires evaluating the marginal log-likelihood of observed data in MR-LD or MR-LDP as has previously been done in [31, 32]; details are given in supplementary document. As VB searches within a factorizable family for posterior distribution, one can only obtain an approximation for the posterior distribution of latent variables.…”

“…Thus, the evidence lower bound (ELBO) from VB-type algorithm cannot be directly used to conduct a likelihood-based test. In this paper, we follow Yang et al [32] and adopt the similar strategy to calibrate ELBO as well as mitigate the bias of variance. Details for the PX-VBEM algorithm and the calibration of ELBO can be found in the supplementary document.…”