Biological pathway prediction from multiple data sources using iterative Bayesian updating

There is a diversity of functional genomics data, such as gene expression data from microarray experiments, phenotypic data from gene deletion experiments, protein-protein interaction data, and data from manually curated databases of gene function. Each data source finds certain types of relationships between genes and misses other types of relationships. A method that can combine multiple data sources might then be able to uncover more relationships than a method that depends on a single data source. This paper presents a method that uses an iterative Bayesian updating technique to combine data from multiple sources, represented as undirected weighted graphs, in order to estimate the probability that a gene is part of a given biological pathway. This method improves performance over a simple neighbor based approach for several well characterized biological pathways.