Genome-wide association studies (GWAS) have successfully mapped many thousands of loci for complex phenotypes, yet the manner by which such loci influence these phenotypes has proven elusive as the majority of associations have unclear biological significance. Recent work has shown that GWAS associations are enriched in transcription regulatory and enhancer regions. To leverage this information for studying complex phenotypes, current studies map molecular quantitative trait loci (QTL) with respect to multi-omics (i.e., epigenetic, transcriptomic, proteomic, and metabonomic) data and then incorporate molecular QTL in GWAS for functional association studies. However, the impact of this approach is limited because existing methods usually only analyze cis-acting molecular QTL and fail to consider the complicating effects that linkage disequilibrium (LD) has on the mapping uncertainty of molecular QTL (disentangling true causal variation from nearby correlated null variations). These limitations reduce the yield of functional association studies for considering incomplete information about molecular QTL. This proposal will develop novel Bayesian statistical methods for improved integrative multi-omics studies with real applications for validation. Our proposed methods have potential to elucidate the genomic etiology of many complex phenotypes, by increasing the precision of mapping molecular QTL and identification of risk genes. These novel Bayesian methods are built upon our recent work and will account for prior knowledge for the parameters of interest through flexible prior distribution assumptions and account for LD by jointly modeling genome-wide variants. (i) First, we will extend our recently proposed Bayesian GWAS method to enable mapping both cis- and trans-acting (genome-wide) molecular QTL. We will model different genetic architectures for cis- and trans-acting variants by assuming respective prior distributions. Our previously derived scalable Bayesian inference algorithm will also be adapted for this new model. (ii) Next, we will develop novel Bayesian methods for functional association studies, which will take the mapping uncertainty of molecular QTL into account through flexible prior assumptions for variant effect sizes. (iii) Finally, to make the most use of public summary-level multi-omics data of large sample sizes, we will derive new Bayesian inference algorithms using only summary-level data while obtaining equivalent results as using individual-level data for our proposed Bayesian methods. (iv) We will validate the proposed methods by applying them to multi-omics and GWAS data from well-characterized older adults and relevant public summary-level data to study Alzheimer's disease (AD) dementia and other complex phenotypes. My lab has access to the well- characterized AD dementia related phenotypic, multi-omics, and GWAS data from older adults participating in the Religious Orders Study (ROS) and Memory and Aging Project (MAP) studies by Rush Alzheimer Disease Center. We will release free software to implement the novel Bayesian statistical tools developed in this proposal.
This proposal will develop novel Bayesian statistical tools for integrating multi-omics (i.e., epigenetic, transcriptomic, proteomic, and metabolomic) data with GWAS data to elucidate the genomic etiology of complex phenotypes. We will validate our methods by applying them to unique multi-omics and GWAS data profiled from older adults with well-characterized phenotypic data, along with relevant public summary-level data, for studying Alzheimer's disease dementia. These novel methods have the potential to increase the information obtained from integrative multi-omics studies to improve our understanding about the genomic etiology of complex phenotypes, thus providing novel gene targets for further drug discovery and improve identification of at-risk individuals.
