Mining New Motifs from Cdna Sequence Data

General biological databases that store basic information on genome, transcriptome, and proteome are indispensable sequence discovery resources. However, they are not necessarily useful for inferring functions of proteins. To see this, we observe that SWISS-PROT —a protein knowledgebase containing curated protein sequences and functional information on domains and diseases—has grown a mere 26-fold in 15 years, from 3,939 entries in 1986 to 126,147 entries in 2003. Similarly, despite the human draft genome and the mouse draft genome and transcriptome, the number of human and mouse protein sequences with some functional information has remained low—7,471 (7.4%) for man and 4,816 (4.7%) for mouse—compared to an estimated proteome of 0.5–1.0 sequences. The majority of sequences in the TrEMBL database of SWISSPROT/TrEMBL, FANTOM, and other similar databases are hypothetical proteins, or are uninformative sequences described as “similar to DKFZ ...” or ”weakly similar to KIAA ....” These sequences have no informative homolog that had diverged from a common ancestor, and have matched to a non-informative homolog. Algorithms for identification of motifs are commonly used to classify these sequences, and to provide functional clues on binding sites, catalytic sites, and active sites, or structure/functions relations. For example 5,873 of 21,050 predicted FANTOM1 protein sequences contain InterPro motifs or domains. In fact, the InterPro name is the only functional description of 900 sequences. Extrapolations from current mouse cDNA data indicate that the proteome is significantly larger than the genome. This underlines the importance of exploring protein sequences, motifs, and modules, to derive potential functions and interactions for these sequences. Strictly defined new protein sequence motifs are either conserved sequences of common ancestry, or are convergence (functional motifs)