Annotating the Function of Protein-coding Genes Based on Gene Ontology Terms of Neighboring Co-expressed Genes

Abstract

Proteins are of key importance in virtually every cellular process but many proteins have still not been annotated with functions due to experimental difficulties involved with functional assays. To address this problem, many computational methods based on sequence homology, three-dimensional structure, genomic context, and gene expression were developed to predict functions of proteins. Here, we tested the performance of a novel approach that is motivated by the concept of bacterial operons. To predict the substrate specificities of membrane transporters we combined genomic context-based methods with Gene Ontology and gene expression data whereby using SVM for classifying genes. We found that in Escherichia coli, the substrate-specificities of membrane transporters can be predicted with ca. 90% accuracy from the biological functions of co-expressed neighboring genes. In Saccharomyces cerevisiae and Homo sapiens, the respective accuracies are lower at around 80%. When applying the same strategy to enzymes of four metabolic classes of Escherichia coli, we found lower accuracies of 77% (2-class prediction) and 68% (4-class prediction), respectively. This suggests that transfer of functional associations between co-expressed neighbor genes may be case-specific