Using coevolution to improve protein subfamily classification

FRANCO SIMONETTI; ARIEL BERENSTEIN; ARIEL CHERNOMORETZ; CRISTINA MARINO BUSLJE

San Carlos de Bariloche

Congreso; V Congreso Argentino de Bioinformática y Biología Computacional; 2014

Asociación Argentina de Bioinformática y Biología Computacional

Background The common approach for protein subfamily classification relies on grouping protein sequences according to their degree of similarity. However, there is no single sequence similarity threshold for accurately grouping sequences into isofunctional groups. Most methods rely on protein superfamilies as a starting point for subfamily classification. Superfamilies are defined as a set of homologous proteins in which conserved sequence or structural characteristics can be associated with conserved functional characteristics. Superfamily members can be highly divergent and catalyze quite different overall reactions. A subfamily is defined as a set of homologous proteins within a superfamily that perform an identical function by the same mechanism. Current subfamily classification methods use bottom-up clustering to construct a cluster hierarchy, then cut the hierarchy at the most appropriate locations to obtain a single partitioning [1, 2]. These methods usually integrate data such as protein sequence similarity, residue conservation within groups and HMM profiles. Moreover, results usually predict a great number of subfamilies with few members and limited biological meaning.The goal of this study is to identify subsets of functionally closely related sequences within a given superfamily. Since all proteins within a superfamily share a common ancestor, we hypothesize that functional diversity within superfamilies has arisen through a series of concerted changes that must have left an identifiable coevolutionary signal. Material and Methods The challenge is to be able to separate the subfamilies coevolutionary signals and use them in the process of subfamily classification. This information can be used to guide a hierarchical clustering. Our approach uses Mutual Information to calculate covariation [3] and commonly used clustering methods based on sequence similarity. We have defined a select group of superfamilies from the Structure Function Linkage Database as our gold standard dataset [4]. Results Different approaches were considered for integrating Mutual Information data in sequence clustering. Since Mutual Information can only be calculated for a group of sequences, a preliminary sequence clustering is performed. Using solely covariation data, our method can cluster groups of sequences from the same subfamily. For a complete clustering solution, it performs almost as good as a hierarchical clustering based on sequence similarity. The next step will be to integrate both methods.   Conclusions Automated protein classification remains an active topic of research and state of the art methods are far from predicting biologically meaningful results. Covariation data has never been used before in this context and further analysis are needed to improve the method.