A Latent Eigen Probit Model with Link Uncertainty for Prediction of Protein-Protein Interactions

• Activity Number: 347
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Xiaoyu Jiang*+ and Eric D Kolaczyk
• Companies: Novartis Institutes for BioMedical Research and Boston University
• Keywords: protein-protein interaction ; link uncertainty ; latent eigenmodel ; probit model ; hierarchical Bayesian ; prediction

Abstract

Protein-protein interactions (PPI's) are of fundamental importance in biology and biomedicine. Identifying and characterizing protein interactions based on various genomic and proteomic data has become a canonical problem in computational biology. Approaching this task as a binary classification problem, we propose a hierarchical Bayesian probit-based framework, incorporating multiple sources of relational protein data as covariates, for modeling binary network topology. More importantly, this model has two distinctive features - (1) capturing the latent characteristics of nodes in the network by an eigenmodel, and (2) accounting for and correcting the link uncertainty in the training data, a well-known critical issue with protein interactions generated by high-throughput technology. We evaluate and compare the predictive performance of the proposed model with three submodels without one or both of these features. Results from two yeast functional subnetworks have demonstrated that both the latent eigenmodel and accounting for link uncertainty are important for better predictions, and the latter can yield substantial improvement in predictive precision.