Probabilistic model of human PPI, Rhodes, 2005, Nat Biotech

Probabilistic model of the human protein-protein interaction network

Daniel R Rhodes1,2,7, Scott A Tomlins2,7, Sooryanarayana Varambally2,7, Vasudeva Mahavisno2, Terrence Barrette2, Shanker Kalyana-Sundaram2, Debashis Ghosh3, Akhilesh Pandey6 & Arul M Chinnaiyan1,2,4,5

2005, Nature Biotechnology

Rhodes05 proposed a probabilistic model integrating PPI, protein domain, and expression data, and annotation data for 40K PPI.

Rhodes05 used otholog interaction from Sce, worm, and fruit fly to predict interactions in human proteins.

A semi-naive Bayes model was used to predict human protein interactions. A maximum likelihood ratio was used as the final score. Co-expression data from oncomine were used for the prediction. (My critique: oncomine may contain abnormal co-expressions).

My  summary: The probabilistic network model of Rhodes05 is used for prediction, not to describe the stochastic nature of protein-interactions or conditional protein-interactions depending on cell types or states, such as cell cycles, growth stages. 

Human Interaction Map  www.himap.org

Rhodes05 used both gold positive control and gold negative control. The gold negative control is an good idea which I can add into my future network analysis project.