Prediction of Protein-Protein Interactions: Looking Through the Kaleidoscope

A comprehensive map of functional protein interactions is still lacking, due to the complexity in their detection by experimental techniques and their associated high rate of false hits. Computational methods can help in filling this knowledge gap by efficiently exploiting the sparse information available. Here we review some of the most commonly used computational approaches to the analysis and prediction of protein interaction data, with a particular emphasis on the prediction of binary protein complexes and their interfaces. These approaches may play an important role in the validation and targeting of known, but poorly characterised, complexes.