Title: PhyloGibbs: A Gibbs sampler for motif detection incorporating phylogenetic information
Speaker: Rahul Siddharthan, Rockefeller University
Date: Wednesday, February 25, 2004, 1:00 pm
Location: Hill Center, Room 260, Rutgers University, Busch Campus, Piscataway, NJ
We have developed a new Gibbs sampler algorithm with the motivation of finding motifs, representing candidate binding sites for transcription factors, in closely related species. Since much conservation here arises not from the existence of functional sites but simply from the lack of sufficient evolutionary divergence between the species, a conventional Gibbs sampler will fail. This approach, used on known test cases, is effective in comparison with conventional methods(ignoring phylogeny). Some examples from recently sequenced yeast species will be presented. This algorithm is also applicable to single-species or phylogenetically-unrelated sequences, and has further improvements over previous Gibbs samplers, including accounting for correlations in the ``background'' model, an option to search for ``dimers'' (pairs of motifs with variable spacing), and a ``tracking'' strategy that allows us to assess the significance of candidate motifs. Using the algorithm on yeast regulatory regions, we reproduce known results and make new predictions.
Seminar sponsored by DIMACS/BIOMAPS Seminar Series on Quantitative Biology and Epidemiology.