Computational methods for the prediction of protein interaction partners

Several computational methods have been recently developed for the prediction of protein interactions. The first class of these methods exclusively uses sequence information for the predictions, including three methods rooted in comparative genomics (phylogenetic profiles, conserved gene neighborhood, and gene fusion), and two other methods that use multiple sequence alignments as input information (in silico two-hybrid and MirrorTree methods). Interestingly, these two methods can be extended beyond the prediction of interaction partners to the detection of regions of interactions and key functional residues, as demonstrated in various experimental systems. The second type of computational methods exploits the information of protein structures by analyzing the observed combinatorial of protein domains or by extrapolating the structural information of protein complexes to compatible sequences. Several studies have addressed the interesting possibilities of combining predicted and experimentally derived interactions. The current view is that the overlap between the interactions predicted by different methods is relatively small, and the various experimental and computational methods seem to be able to detect specific types of protein interactions. A number of computational methods for predicting functional and/or physical relations between proteins have been developed in the last 4 years. The computational methods are strongly rooted in the traces left by the evolution in the organization and composition of bacterial genomes (see Microbial genomes). The methods developed can be divided into three categories: those methods that only use the information from genomes and sequences for the prediction of interaction partners, those methods that use the information of protein complexes of known structures (see Fundamentals of protein structure and function, Large complexes by X-ray methods, and Large complexes and molecular machines by electron microscopy), and those methods that use the puzzle composition of proteins in domains (see Classification of proteins into families, Pfam: the protein families database, and COGs) to predict the probability of interaction between the corresponding proteins. Keywords: protein interaction; in silico predictions; methods comparison; bioinformatics; computational methods