Heterodimeric protein complex identification

It is a challenging problem to predict heterodimeric protein complexes accurately in size and membership, because, in yeast, those complexes are the majority of curated protein complexes, and structures of those complexes are much simpler than those of complexes consisting of three or more proteins. In this paper, we characterize heterodimeric protein complexes by supervised-learning of a naïve Bayes classifier from heterogeneous genomic data, including protein-protein interaction data, gene expression data, and gene ontology annotations. We have examined predictability of the trained classifier and compared it with those of existing popular protein complex prediction tools. The result shows that our method outperforms the others.