Predictive Learning of Transcriptional Networks

Princeton University, Princeton NJ

Investigators

Abstract

Abstract We propose to develop and apply a comprehensive set of computational and experimental methods in order to fully characterize the regulation of mammalian gene expression at the sequence level. At the core of our approach is an informationtheoretic framework for sensitive and highly specific identification of DNA and RNA regulatory elements from large-scale gene expression data and genomic sequence information. We will develop and apply a non-alignment based approach based on network-level conservation in order to identify comprehensive catalogues of regulatory elements conserved between pairs of mammalian genomes. These high-confidence predictions will then be used in order to identify distal regulatory elements composed of clusters of transcription factor binding sites. A Bayesian network learning algorithm will be employed to learn the combinatorial rules by which the discovered elements function to affect gene expression[unreadable]both within local promoters/3[unreadable]UTRs and through distal regulatory modules such as enhancers and silencers. We propose a versatile approach based on microarray profiling of phage-display selections in order to rapidly and efficiently identify the protein trans factors that specifically interact with the hundreds of novel DNA and RNA regulatory elements we expect to identify. The proposed research will significantly advance the rate and scale at which regulatory networks are characterized[unreadable]both in humans, but also across a range of other complex genomes of biomedical and industrial importance.