Protein clusters analysis via directed diffusion

Motivation: Graph-theoretical approaches are useful for elucidating the modular compositions of protein-protein interaction networks, which are known to consist of regions of increased network connectivity (clusters), corresponding to known molecular complexes or functional pathways. In this work, we introduce the concept of semisupervised directed diffusion as a graph-based methodology for cluster analysis. Results: We show that our scheme allows both similarity propagation and cluster boundary detection. It is experimentally verified by analyzing known biological pathways, and we show that it can accurately identify an entire pathway, given only 10%-20% of its proteins. Thus, we submit that directed diffusion is a promising approach for evidence propagation in biological networks and clustering of functional groups. Availability: A Matlab implementation of the proposed scheme is available at http://pantheon.yale.edu/∼yk253/software/.