Prediction of interaction partners for orphan nuclear receptors by prior-based protein sequence profiles

We present a prior-based profile method for the prediction of protein-protein interaction partners that is here applied to the nuclear receptor superfamily. In this method, the diagnostic features are locally encoded in the physicochemical properties of residues in the interaction surface that are conserved in all proteins belonging to the defining set. The procedure models the positional variation based on that observed in the defining set and a prior-based substitution matrix derived from over 20,000 highly conserved positions in a set of 147 functional protein families. The method clusters sets of nuclear receptors known to interact with retinoid X receptor or corepressor proteins with predictive sets of receptors in C. elegans and higher metazoans. The method effectively reduces the search space of all possible interactions and yields experimentally testable predictions. Applications of this novel approach extend to interaction prediction problems in general, particularly to those that are not amenable to analysis by the rigid-body approximation.