Introduction
In our companion review published in the last issue , we outlined the experimental techniques for the identification and characterization of protein interactions. We showed that high-throughput experimental methods produce a large amount of data which needs to be analyzed and verified. Despite this, interactomes of many organisms are far from complete. The low interaction coverage along with the experimental biases toward certain protein types and cellular localizations reported by most experimental techniques call for the development of computational methods to predict whether two proteins interact. These methods can be very useful for choosing potential targets for experimental screening or for validating experimental data (see ) and can provide information about interaction details (in the case of domain prediction methods) which might not be apparent from the experimental techniques. Many methods use combinations of experimental and computational techniques to different extent (for example, gene co-expression and synthetic lethality methods were covered among experimental approaches in our companion paper ) and do not predict physical interactions directly but rather infer the functional associations between potentially interacting proteins.
In this review, we report on several methods to predict protein or domain interaction partners. Some computational methods are based on the co-localization of potentially interacting genes in the same gene clusters or protein chains (gene cluster, gene neighborhood, and Rosetta stone methods), on co-evolution patterns in interacting proteins (sequence co-evolution methods), and on the co-expression of genes. Some methods find patterns of co-occurences in interacting proteins, protein domains, and phenotypes (phylogenetic profiles and synthetic lethality methods), while others use the presence of sequence/structural motifs characteristic only for interacting proteins (classification methods, association methods). To analyze interaction specificity at the domain level, in this second paper of the review we describe methods that are aimed at identifying specific domains mediating interactions in an interacting protein pair.