Unified framework for non-coding mutation analysis based on information theory

点击翻译按钮获取中文摘要

My research program develops bioinformatic methods for interpreting genomic sequence variation, specifically DNA or RNA elements that regulate transcription and post-transcriptional processes.There is an acute need for better approaches that distinguish deleterious variants from benign sequence changes. This proposal develops a unified framework, based on information theory, for interpretation of variants of unknown significance (VUS) in complete gene sequences and genomes. Changes in information contents of nucleic acid binding sites are a surrogate measure of affinity and can be used to accurately predict deleterious mutant binding sites. We have produced software to determine information changes at individual sites efficiently in complete genomes. Mutations that alter exon definition can be inferred with multipartite information models that account for multiple binding sites and intersite distances (ie. gap surprisal). We validate predicted gene expression changes by comparing abnormal splice isoforms in genomes containing predicted mutations causing missplicing with controls lacking these variants. This proposal develops a system for interpreting non-coding VUS, regardless of the type of binding event. Mutations affecting transcription factor binding, and RNA-binding protein stabilization of mRNA, and exon definition comprising multiple RNA binding events will be modeled. Information position weight matrices (PWM) are computed for each protein-nucleic acid interaction. These models will be cross-validated using chromatin immunoprecipitation data (from the International Epigenome Consortium) from multiple cell lines with the same transcription factors to eliminate low affinity, non-specific binding events. RNA binding sequences are processed similarly from protein cross-linking (public I- and PAR-CLIP) data. The PWMs are used to detect mutations based on information changes. Combinatorial analysis of cooperative regulation and overlapping or adjacent sites will be evaluated using information densities of site clusters, gap surprisal, or mixed models. Mutations will then be reanalyzed with these combinatorial methods for changes that correspond to promoter strength. After predicting deleterious single nucleotide variants, we will modify our case-control gene expression validation methods by computing the likelihood that a variant produces steady state changes in expression of all genes with predicted transcription factor or RNA binding site mutations, relative to controls lacking these mutations. The project will culminate with development of a system to make this unified information theory-based framework broadly available. This software will generate an abbreviated set of potentially deleterious mutations from gene panels and complete genome sequences.