Predicting protein complexes by data integration of different types of interactions.

The explosion of high throughput interaction data from proteomics studies gives us the opportunity to integrate Protein-Protein Interactions (PPI) from different type of interactions. These methods rely on the assumption that proteins within a complex have more interactions across the different data sets which translate into the identification of dense subgraphs. However, the relative importance of the types of interaction are not equivalent in their reliability and accuracy consequently they should be analysed separately. Here we propose a method that use graph theory and mathematical modelling to solve this problem. Our approach has four steps that: i) score independently each type of interaction; ii) build an interaction specific networks for each type; iii) weight the specific networks; and iv) combine and normalise the scores. Using this approach to the BRCA1 Associated genome Surveillance Complex (BASC), we correctly identified the known core components of the complex and subcomplexes that have solved structures as well as predicted new interactions and core complexes. The method presented in this study is of general use. It is flexible enough to allow the development of any scoring system and can be applied to any protein complex to provide the latest knowledge in its interactions and structure.