New directions for diffusion-based network prediction of protein function: incorporating pathways with confidence

• DOI: 10.1093/bioinformatics/btu263
• Publication Date: 2014-06-16T21:55:09Z
• Authors: Mengfei Cao; Christopher M. Pi...; Xian Feng; Kathryn J. Doroschak; Thomas Schaffner; Jisoo Park; Hao Zhang; Lenore J. Cowen; Benjamin J. Hescott
• Keywords: Representation

Abstract Motivation: It has long been hypothesized that incorporating models of network noise as well edge directions and known pathway information into the representation protein–protein interaction (PPI) networks might improve their utility for functional inference. However, a simple way to do this not obvious. We find diffusion state distance (DSD), our recent diffusion-based metric measuring dissimilarity in PPI networks, natural extensions incorporate confidence, can even express coherent pathways by calculating DSD on an augmented graph. Results: define three incremental versions which we term cDSD, caDSD capDSD, where capDSD matrix incorporates directed edges, measure how similar each pair nodes is according structure network. test four popular function prediction methods (majority vote, weighted majority multi-way cut flow) using these different matrices Baker’s yeast cross-validation. The best performing method vote capDSD. then performance integrated heterogeneous set protein association edges from STRING database. superior context confirms treating probabilistic units more powerful than simply independently Availability: All source code confidences, extracting KEGG XML files, are available http://dsd.cs.tufts.edu/capdsd Contact: [e-mail] or [e-mail] Supplementary information: data at Bioinformatics online.