A Model-based Approach to Transcription Regulatory Network Reconstruction from Time-Course Gene Expression Data

Time-course gene expression profiling provides valuable data on dynamic behavior of cellular responses to external stimulation. Investigation of transcription factors (TFs) that regulate co-expressed genes in a dynamic process can reveal insights on the underlying molecular mechanisms. As the ChIP-seq technology is only suitable for a fraction of TFs in mammalian organisms, the computational identification of relevant TFs remains to be critical. We propose a regression-based model to infer the functional binding sites of TFs from time-course gene expression profiles. Our approach incorporates an association strength for each potential TF and target gene pair based on computational analysis of binding sites in promoter sequences of co-expressed genes. Our model further uses the Lasso-penalized technique to search for the most informative TF-target pairs. The application of our method to a gene expression study on E2-induced apoptosis in a variant of MCF-7 cells revealed that the findings are biologically meaningful.