Inferring gene functions from metabolic reactions

Metabolic networks model the physiological processes that transform metabolites in organisms. A metabolic network is considered to be in steady state if the rate at which all such transformations remain unchanged. Analyzing steady states has been essential in understanding the contribution of individual molecules to long term characteristics of the underlying organism. In this paper, we develop a novel method to establish the relationship between the functions of genes that take part in a given metabolic network and the steady states of that network systematically. To do this, we first characterize the impact of each reaction on the steady states of the network. Then, using their impacts, we group every reaction in the network into clusters of genes with similar impacts. We conjecture that genes with similar impacts on the set of possible steady states tend to serve similar functions. Following from this conjecture, for each group we formed, we calculate the enrichment of each gene ontology (GO) term that exists for at least one gene in that group. Given a new gene with missing annotations in the network, we find the cluster that is closest to that gene in the steady state space. We predict the enriched GO terms of in that cluster as possible annotations to that gene. Our experiments demonstrate that enrichment values correlate highly with the actual GO terms of each reaction, and thus, our method can predict the GO terms of less known genes accurately.