Predicting protein-peptide interactions via a network-based motif sampler.

MOTIVATION: Many protein-protein interactions are mediated by peptide recognition modules (PRMs), compact domains that bind to short peptides, and play a critical role in a wide array of biological processes. Recent experimental protein interaction data provide us with an opportunity to examine whether we may explain, or even predict their interactions by computational sequence analysis. Such a question was recently posed by the use of random peptide screens to characterize the ligands of one such PRM, the SH3 domain.
RESULTS: We describe a general computational procedure for identifying the ligand peptides of PRMs by combining protein sequence information and observed physical interactions into a simple probabilistic model and from it derive an interaction-mediated de novo motif-finding framework. Using a recent all-versus-all yeast two-hybrid SH3 domain interaction network, we demonstrate that our technique can be used to derive independent predictions of interactions mediated by SH3 domains. We show that only when sequence information is combined with such all versus all protein interaction datasets, are we capable of identifying motifs with sufficient sensitivity and specificity for predicting interactions. The algorithm is general so that it may be applied to other PRM domains (e.g. SH2, WW, PDZ). AVAILABILITY: The Netmotsa software and source code, as part of a general Gibbs motif sampling library, are available at http://sf.net/projects/netmotsa