An automated method for finding molecular complexes in large protein interaction networks

The paper addresses the growing need to analyze large biomolecular interaction networks produced by proteomics technologies such as two-hybrid, phage display, and mass spectrometry. It presents Molecular Complex Detection (MCODE), a novel graph-theoretic clustering algorithm that identifies densely connected regions in protein-protein interaction networks that may correspond to molecular complexes. The method weights vertices by local neighborhood density and then traverses outward from a locally dense seed protein to isolate dense regions, offering a directed mode for fine-tuning clusters of interest and examining cluster interconnectivity.

Using protein interaction and complex data from the yeast Saccharomyces cerevisiae, the authors show that dense regions can be found based solely on connectivity, with many corresponding to known protein complexes. Importantly, the algorithm is not affected by the known high rate of false positives in high-throughput interaction data, making it a robust tool for knowledge discovery, and the program was made publicly available.