Delineating the network effects of mental disorder-associated variants using convex optimization methods

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Driven by international open scientific collaboration through groups such as the Psychiatric Genomics Consortium (PGC, in which co-I Mullins is a leading analyst) both genome-wide association studies (GWAS) and whole exome and genome sequencing studies of neuropsychiatric disorders (NPDs) are rapidly increasing in sample size. With this increased sample size comes increased statistical power to detect many more, smaller genetic effects on disease risk, known as the polygenic component. The challenge now is to understand what these findings tell us about NPD risk, etiology and biology. Here we propose a suite of methods for multi-trait analysis to determine underlying latent structure, causal networks of genes and traits, and enriched data-derived regulatory pathways. We make extensive use of convex optimization methods that allow both computational efficiency and guarantees on reproducibility. In Aim 1 we will decompose a wide range of NPDs and their subphenotypes into shared and unique genetic components using a novel convex formulation of observed-weighted principal components analysis (PCA) and develop extensions to handle sample overlap, linkage disequilibrium (LD), and different ancestries. In Aim 2 we will extend and customize our existing work on causal network inference using biconvex optimization to estimate both cis and trans gene regulatory networks in the brain using large-scale uniformly processed chromatin accessibility and expression quantitative trait loci (QTLs). We will regularize estimates of cis interactions using chromatin conformation data, model latent genetic confounders in these networks using an expectation-maximization (EM) approach and estimate networks over both genes and NPDs in order to determine the most direct causes (“core” genes in the omnigenic model). In Aim 3 we will analyze both rare and common genetic associations in their gene regulatory network context. Borrowing from cancer genomics, we will use heat diffusion models to propagate statistical information on the local network over both genes and regulatory elements (REs) and then use graph clustering algorithms to extract “hot” subnetworks, corresponding to pathways implicated in the NPD under study. The methods we develop for these analyses will be made publicly available under source licenses with extensive support in terms of documentation, tutorials, and vignettes. Through this we hope to empower future “post-GWAS” analyses that can leverage the genetic, subphenotype and trait networks underlying human neuropsychiatric health, and eventually point the way to therapeutic interventions.