DETECTING MUTUALLY EXCLUSIVE INTERACTIONS IN PROTEIN-PROTEIN INTERACTION MAPS

Proteins are responsible for an impressive large variety of functions. To properly understand the significance of protein-protein interactions in the cell it is important to address two problems: first, is identification of the different interactions that are involved in each biological function, and, second, is determining how proteins interact and the consequences of the interaction. The identification of protein interactions by high-throughput experiments has led to the development of a number of methods for their analysis, producing, in the last years, a vast amount of interacting data. However, there are at least two issues that arise from the analysis of such experimental maps, these are, on one side, the significant number of false positives they contain and, on the other, the difficulty in distinguishing whether, when more than one protein interact with the same partner, they can do so simultaneously, i.e. whether their interaction is mutually exclusive. The general strategy we describe is based on the combination of known three-dimensional structures with protein-protein interaction networks to determine which of the multiple interactions or connections that are made by a hub can occur in mutually exclusive fashion, and, in such cases, identify, whenever is possible, the shared similarities in their binding regions, concluding that their interaction has to be mutually exclusive (i.e. not simultaneous) and that the region identified by similarity is indeed the interaction site. We applied this strategy to the interactomes of seven organisms. We show that our methodology allows the identification of mutually exclusive interactions with accuracy higher than 77%. The procedure also allows us to predict which residues are likely to be in the binding interface of the nodes, and in a significant number of cases (between 63% and 75%) we correctly identify at least one of them (5 on average) and this has obvious implications for helping to reduce the search space in docking procedures. The coverage of the method varies substantially for different organisms, as it could be expected, however it does reach 42% for human and more than 36% for yeast averaging at about 25%. These figures are bound to increase with time both thanks to the progress in experimental methods and, possibly, to the increasing reliability of modeling techniques. For this reason, we also introduce here the Estrella server that embodies this strategy, is designed for users interested in validating specific hypotheses about the functional role of a protein-protein interaction and it also allows access to pre-computed data for seven organisms.