MOSTWAS

MOSTWAS integrates germline genetics, multi-omic mediators, and distal-SNP effects into transcriptome imputation to improve prediction of gene expression and detection of gene–trait associations.

Key Features:

• Identification and incorporation of mediating biomarkers: Identifies mediators such as CpG sites, microRNAs, and transcription factors, trains predictive models for these mediators using their local SNPs, and integrates imputed mediator values into final gene expression models alongside local SNPs.
• Assessment of distal-eQTLs and distal-eSNP mediation: Evaluates distal-eQTLs for mediation through biomarkers proximal to distal-eSNPs and incorporates distal-SNPs with significant indirect mediation effects into transcriptomic prediction models.
• Enhanced predictive performance and power: Leverages multi-omic mediation to increase the percent variance explained of gene expression by an additive ~1–2% and to improve TWAS power.
• Mechanistic insights via added-last test: Implements an added-last test that assesses the additional information provided by distal variants beyond local associations to infer regulatory mechanisms within tissues.
• Application to real datasets: Demonstrated with simulations and real data including ROS/MAP brain tissue and TCGA breast tumors to facilitate identification of risk genes for traits and disorders.

Scientific Applications:

• Transcriptome-wide association studies (TWAS): Improves transcriptome imputation and increases power to detect gene–trait associations by incorporating distal-SNPs and multi-omic mediators.
• Gene regulation mechanism discovery: Supports inference of regulatory mechanisms within tissues by assessing mediation via CpG sites, microRNAs, and transcription factors and by using the added-last test.
• Risk gene identification in disease cohorts: Facilitates identification of risk genes for traits and disorders in datasets such as ROS/MAP and TCGA breast tumors.
• Distal-eQTL mediation evaluation: Enables assessment and inclusion of distal-eQTLs mediated through proximal biomarkers into predictive models.

Methodology:

Trains predictive models for mediators using local SNPs, imputes mediator values and integrates them with local SNPs into final transcriptomic prediction models, evaluates distal-eQTLs for mediation via proximal biomarkers, applies an added-last test to quantify additional contribution of distal variants, and validates performance using simulations and datasets from ROS/MAP and TCGA.