Abstract
Motivation Residue-residue coevolution has been used to elucidate structural information of enzymes. Networks of coevolution patterns have also been analyzed to discover residues important for the function of individual enzymes. In this work, we take advantage of the functional importance of coevolving residues to perform network-based clustering of subsets of enzyme families based on similarities of their coevolution patterns, or “Coevolution Similarity Networks”. The power of these networks in the functional analysis of sets of enzymes is explored in detail, using Sequence Similarity Networks as a benchmark.
Results A novel method to produce protein-protein networks showing the similarity between proteins based on the matches in the patterns of their intra-residue residue coevolution is described. The properties of these co-evolution similarity networks (CSNs) was then explored, especially in comparison to widely used sequence similarity networks (SSNs). We focused on the predictive power of CSNs and SSNs for the annotation of enzyme substrate specificity in the form of Enzyme Commission (EC) numbers using a label propagation approach. A method for systematically defining the threshold necessary to produce the optimally predictive CSNs and SSNs is described. Our data shows that, for the two protein families we analyse, CSNs show higher predictive power for the reannotation of substrate specificity for previously annotated enzymes retrieved from Swissprot. A topological analysis of both CSNs and SSNs revealed core similarities in the structure, topology and annotation distribution but also reveals a subset of nodes and edges that are unique to each network type, highlighting their complementarity. Overall, we propose CSNs as a new method for analysing the function enzyme families that complements, and offers advantages to, other network based methods for protein family analysis.
Availability Source code available on request.
