Identifying functional modules using expression profiles and confidence-scored protein interactions.

MOTIVATION: Microarray-based gene expression studies have great potential but are frequently difficult to interpret due to their overwhelming dimensions. Recent studies have shown that the analysis of expression data can be improved by its integration with protein interaction networks, but the performance of these analyses has been hampered by the uneven quality of the interaction data.
RESULTS: We present Co-Expression Zone ANalysis using NEtworks (CEZANNE), a novel confidence-based method for extraction of functionally coherent co-expressed gene sets. CEZANNE uses probabilities for individual interactions, which can be computed by any available method. We propose a probabilistic model and a weighting scheme in which the likelihood of the connectivity of a subnetwork is related to the weight of its minimum cut. Applying CEZANNE to an expression dataset of DNA damage response in Saccharomyces cerevisiae, we recover both known and novel modules and predict novel protein functions. We show that CEZANNE outperforms previous methods for analysis of expression and interaction data. AVAILABILITY: CEZANNE is available as part of the MATISSE software at http://acgt.cs.tau.ac.il/matisse. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.