Automatic Identification and Classification of Protein Domains

Motivation: Proteins are comprised of one or several domains. Such domains can be classified into families according to their biological function. Whereas sequencing technologies have advanced immensely in recent years, there are no matching computational tools for large-scale determination of protein domains and their boundaries. The present paper addresses the challenge of developing computational tools to identify protein domains and to classify them into their families. The eventual goal of our research is to automatically identify and classify correctly all protein domains. Results: Our method, called EVEREST, combines methodologies from the fields of finite metric spaces, machine learning and statistical modeling and achieves state of the art results. Our process begins by constructing a database of protein segments that emerge in an all vs. all pairwise sequence comparison. It then proceeds to cluster these segments, choosing the best clusters using machine learning techniques, and creating a statistical model for each of the these clusters. This procedure is then iterated: The aforementioned statistical models are used to scan all protein sequences, to recreate a segment database and to cluster them again. Performance tests show that EVEREST recovers 63% of Pfam families and 40% of SCOP families with high accuracy, and suggests new families with about 40% fidelity. EVEREST domains are frequently a combination of domains as defined by Pfam or SCOP and frequently subdomains of such domains. The paper is concluded with a discussion of research avenues to improve these results. Availability: A database of statistical models (HMMER HMMs), one per domain family is available for download at http://www.cs.huji.ac.il/ elonp/everest.