Deciphering protein-protein interactions. Part II. Computational methods to predict protein and domain interaction partners

Abstract

Recent advances in high-throughput experimental methods for the identification of protein interactions have resulted in a large amount of diverse data that are somewhat incomplete and contradictory. As valuable as they are, such experimental approaches studying protein interactomes have certain limitations that can be complemented by the computational methods for predicting protein interactions. In this review we describe different approaches to predict protein interaction partners as well as highlight recent achievements in the prediction of specific domains mediating protein-protein interactions. We discuss the applicability of computational methods to different types of prediction problems and point out limitations common to all of them.

Figure 1. Different Methods of Protein Interaction Prediction

(A) Gene cluster and gene neighborhood methods, different boxes showing different genes. (B) Phylogenetic profile method, showing the presence/absence of four proteins in three genomes. (C) Rosetta Stone method. (D) Sequence co-evolution method looking for the similarity between two phylogenetic trees/distance matrices (E) Classification methods shown with the example of RFD method, where five different features/domains are used and each interacting protein pair is encoded as a string of 0, 1, and 2. The decision trees are constructed based on the training set of interacting protein pairs and decisions are made if proteins under the question interact or not (“yes” for interacting, “no” for non-interacting).

Figure 2. Strategies to Predict Domain Interactions from Protein Interactions

(A) Shows that due to the abundance of domains x and y in protein interaction pairs shown on the same line, the domains x and y are predicted to interact. (B) Illustrates the same dataset revealing that the actual domain interactions (dotted lines) do not include domains x and y. It shows that accounting for other domains in a protein pair in addition to x and y can result in alternative predictions. (C) Considers the case of several paralogous protein pairs (from Family_1 and Family_2) containing the same two domains. In this case each paralog from one domain family (represented by a shade of red for Family_1) interacts with only one specific paralog (represented by a shade of blue for Family_2) of the other domain family. While there are examples of specific interacting domains (shown by dotted line), there are even more cases where they do not interact (shown with an “X”), meaning that the larger abundance of non-interacting examples can mask the few, specific interacting cases.