Research Talk: Diffusion State Distance: a better way to understand protein-protein-interaction networks

Abstract

Most modern protein function prediction methods are based on the simple shortest path distance metric on the protein-protein- interaction (PPI) networks. However, PPI networks are “small world” networks, in which most nodes are close to all other nodes. So there is a limitation to relaying on ordinary shortest path distance metric: connection between two adjacent nodes can signify something very different from that between two other adjacent nodes. In order to capture the type of neighborhood similarity, we introduce a new metric based on a graph diffusion property. By replacing the shortest path metric by our metric, we improve the performance of several classical network-based function prediction methods.

This is joint work with: Mengfei Cao, Jisoo Park, Noah Daniels, Mark Crovella, Lenore Cowen, and Benjamin Hescott.