Bayesian Network-Based Integrative Genomics Methods for Precision Medicine

点击翻译按钮获取中文摘要

Project Summary/Abstract Modern multi-platform genomic data sets contain substantial molecular information potentially useful for discovering new precision therapeutic strategies. Integration across multi-platform data and across genes using network-based models is a key to extracting mechanistic molecular information embedded in these data. In this proposal, we develop integrative network-based methods that ll gaps in existing literature. They will be used to identify key pathways for a given disease and its subtypes, nd key upstream regulators of these pathways and determine which appear to be causal, construct pathway signatures potentially usable for patient selection, and identify factors modulating pathway associations. While our methods will be applicable to any disease setting, our initial focus will be to use multi-platform genomic data sets to provide a deep molecular characterization of four recently discovered consensus molecular subtypes (CMS) of colorectal cancer (CRC) to arm our biomedical and clinical collaborators with knowledge to devise and test new precision therapeutic strategies targeting these subtypes. For these purposes, we propose the following aims: Speci c Aim 1: We will devise a novel model formulation regressing pathway scores on upstream genetic and epigenetic factors to identify a sparse set of potential pathway drivers. We will identify characteristic pathways for each CMS and for each pathway identify potential drivers that our biomedical collaborators will functionally validate via CRISPR and identify potential matching drug targets. We will also develop novel Bayesian hierarchically linked regression models (BLINK) that will determine which cancers share common pathway drivers and thus are candidates for sharing a common targeted therapy, while increasing power for discovery of pathway drivers for rare cancers. Speci c Aim 2: We will develop network mediation analysis approaches to discover putative causal network edges in multi-layered graphs of multi-platform genomic data. We will use these methods to more deeply characterize the networks underlying key CMS-characteristic pathways and determine which potential pathway drivers appear to be causal, and which mediators are predictive of response to therapy. From these networks, we will devise methods to construct pathway signatures integrating multi-platform molecular information to provide a single-number, patient- speci c summary of pathway activity potentially useful for patient selection for precision therapeutics. Speci c Aim 3: We will develop novel Bayesian network regression methods for undirected and multi-layer networks that identify heterogeneous network structure varying linearly or nonlinearly across patient-speci c covariates. We will apply these methods to key networks identi ed for CRC data to discover how these networks vary across various covariates, including subtypes (CMS), biological factors (immune in ltration), and clinical response. Successful completion of this work will produce a broad set of rigorous tools for integrative and network modeling of multi-platform genomic data, and will provide our CRC collaborators with a short list of key CMS-speci c pathways and drivers for functional validation and clinical translation via CMS-based precision therapeutics. Our dissemination efforts will include software for our methods and Shiny apps for exploring biological underpinnings of CRC.