Identification of protein homology using domain architecture comparison

Abstract

Homology identification is the first step for many genomic studies and can be divided into sequence-based and domain-based methods. The sequence-based homology searches can yield false mis-assignments, especially when comparing multidomain proteins. Here we propose a new method to detect homologs through explicit comparison of domain architecture. A major obstacle in comparison of domain architecture, the organization of protein domains, is the existence of ‘promiscuous’ domains, which carry out auxiliary functions and appear in many unrelated proteins. To distinguish these promiscuous domains from protein domains, we assigned a weight score to each domain extracted from RefSeq proteins, based on its abundance and versatility. To measure domain abundance and versatility, we defined the Inverse Versatility (IV) and the Inverse Abundance Frequency (IAF), derived from Inverse Document Frequency (IDF), a statistic commonly used in information retrieval. A domain’s score represents its importance in the “protein world” and is used in the comparison of domain architectures. We examined the weight scores of 215 known promiscuous domains to identify relationship between the obtained weight scores and domain promiscuity. We found that all of the known promiscuous domains have very low weight scores, most of which are below 10. In scoring domains, domain combinations were also considered as well as single domains. To measure domain architecture similarity, we compared domain content, domain order, and domain duplication between two domain architectures. To do this, we used several methods that are based on algorithms used in information retrieval (the cosine similarity, the Goodman-Kruskal gamma function, and domain duplication index). A final similarity score between two domain architectures were obtained by combining these three indices into a similarity score. To determine the best combination of three indices, we used the COG database. Our method was compared with ot...