MOSTWAS: Multi-Omic Strategies for Transcriptome-Wide Association Studies

Abstract

Traditional predictive models for transcriptome-wide association studies (TWAS) consider only single nucleotide polymorphisms (SNPs) local to genes of interest and perform parameter shrinkage with a regularization process. These approaches ignore the effect of distal-SNPs or other molecular effects underlying the SNP-gene association. Here, we outline multi-omics strategies for transcriptome imputation from germline genetics to allow more powerful testing of gene-trait associations by prioritizing distal-SNPs to the gene of interest. In one extension, we identify mediating biomarkers (CpG sites, microRNAs, and transcription factors) highly associated with gene expression and train predictive models for these mediators using their local SNPs. Imputed values for mediators are then incorporated into the final predictive model of gene expression, along with local SNPs. In the second extension, we assess distal-eQTLs (SNPs associated with genes not in a local window around it) for their mediation effect through mediating biomarkers local to these distal-eSNPs. Distal-eSNPs with large indirect mediation effects are then included in the transcriptomic prediction model with the local SNPs around the gene of interest. Using simulations and real data from ROS/MAP brain tissue and TCGA breast tumors, we show considerable gains of percent variance explained (1–2% additive increase) of gene expression and TWAS power to detect gene-trait associations. This integrative approach to transcriptome-wide imputation and association studies aids in identifying the complex interactions underlying genetic regulation within a tissue and important risk genes for various traits and disorders.

Integrative Transcriptome-Wide Analyses Uncover Novel Risk-Associated MicroRNAs in Hormone-Dependent Cancers

“…Our study did not consider the effect of distal SNPs and mediating biomarkers, such as CpG sites, copy number alteration, and transcription factors (Bhattacharya et al, 2021). In HEIDI analysis, a few SMR associations were not considered as their top-associated cis-eQTL had only one or two SNPs in the cis region.…”

TIGAR-V2: Efficient TWAS Tool with Nonparametric Bayesian eQTL Weights of 49 Tissue Types from GTEx V8

“…We provide trained eQTL weights of genes that have 5fold CV ܴ ଶ 0.005 in the Synapse data base with a link given in Web Resources of in this paper. Since we used a more liberal threshold than the 0.01 used by previous studies 2, 21,22 to allow more genes to be tested in follow-up TWAS, we would suggest users to investigate the 5-fold CV ܴ ଶ and test p-values of the expression prediction models, as well as the biological functions of significant TWAS risk genes 26 .…”

“…There are many other alternative TWAS tools available to address these two limits. For example, BGW-TWAS 4 and MOSTWAS 22 uses both cis-and trans-genotype data to train gene expression prediction model of the target gene, while CoMM 75 and PMR-Egger 5,76 assume a joint model with reference and test data that can achieve higher power when both data sets are homogeneous.…”

“…The 5-fold CV ܴ ଶ can be used to evaluate if the trained gene expression prediction model is "valid" for follow-up TWAS (e.g., using the threshold of 5-fold CV ܴ ଶ 0.005). Here, we use a more liberal threshold than the threshold 0.01 used by previous studies 2,21,22 to allow more genes to be tested in follow-up TWAS. Because the follow-up gene-based association Z-score test statistic is essentially a weighted average of single variant GWAS Zscore statistics with variant weights provided by the eQTL effect sizes (Equations 2 and 3), poorly estimated eQTL weights would only reduce power but will not increase false positive rate under null hypothesis.…”

Genetic control of fetal placental genomics contributes to development of health and disease

“…Here, we set out to identify the following: (1) which genes show associations between their placental genetically-regulated expression (GReX) and various traits across the life course, (2) which traits along the life course can be explained by placental GReX, in aggregate, and (3) which transcription factors, miRNAs, or CpG sites potentially regulate trait-associated genes in the placenta (Figure 1). We leveraged multi-omic data from fetal-side placenta tissue from the Extremely Low Gestational Age Newborn (ELGAN) Cohort Study 21 to train predictive models of gene expression enriched for distal SNPs using MOSTWAS, a recent TWAS extension that integrates multi-omic data 22 . Using 40 GWAS of European-ancestry subjects from large consortia [23][24][25][26][27] , we performed a series of TWAS for noncommunicable health traits and disorders that may be influenced by the placenta to identify GTAs and functional hypotheses for regulation (Figure 2).…”