Approaches to integrate diverse data types: applications to inferring functional regulatory networks

Transcriptional regulatory networks are networks of genes and transcription factor proteins that determine the context-specific expression pattern of a gene, where a context could represent a time point, a spatial location or a combination of these. Although these networks are a key to accurate information processing and function in living organisms, the networks are not completely known, that is, for a large majority of the genes, we don’t know what set of transcription factors regulate them, and if, and, what types of combinatorial functions determine the relationship between the levels of these genes and the levels of their transcriptional regulators. Machine learning approaches, including both supervised and unsupervised learning approaches have been used to infer the regulatory networks of several organisms. However, inference of regulatory networks is a computationally and experimentally difficult problem, and the success of these approaches have been modest. It is becoming increasingly clear that integrating different sources of regulatory evidences is crucial to better infer these networks. In this talk I will present some regression and classification approaches that we used to infer the regulatory network for the model organism, Drosophila melanogaster, and discuss some new directions that we are taking to integrate different types of information to help infer higher quality regulatory networks.