Classification of Protein Sequences into Paralog and Ortholog Clusters Using Sequence Similarity Profiles of KEGG/SSDB

We are constructing KEGG/OC (Ortholog Clusters) from KEGG/SSDB (Sequence Similarity DataBase) [2]. KEGG/SSDB contains exhaustive protein sequence similarity scores of completed and nearly completed genomes calculated by the SSEARCH program [3]. KEGG/OC is constructed automatically from the graph analysis of searching cliques with an appropriate definition for the profiles of similarity scores. But there are two major problems to construct the current version of KEGG/OC. First, the current procedure leaves numerous singletons, which should in fact be included in the clusters of related proteins. Second, there are many clusters which contain evolutionarily unrelated proteins. These problems are mainly due to the so-called multi-domain problem. Here, we tried to overcome these problems by using the correlation coefficients for the profiles of KEGG/SSDB sequence similarity scores during classification of paralog genes, and compared our results to the current versions of KEGG/OC and COG [4]. Evaluation of clustering results was made by the degree of consistent motif structuress in the clusters.