Parallel Characteristic Extraction from Protein Sequence Database

An adaptive massively parallel system for flexible information processing has been investigated. This research requires a feedback from the real application. In this paper, a parallel characteristic extraction from the protein sequence database is described. Since the protein sequence database is huge and sequences have variety, an adaptive massively parallel system is mandatory. An HMM (hidden Markov model) is essentially suitable for the massively parallel system. An HMM also represents conditional probabilities to deal with the stochastic nature of the protein. However, finding the optimal HMM topology for protein is a hard problem. Thus, an iterative duplication method is developed for HMM topology learning. Using this method, a motif, one of the protein characteristics, is extracted. We obtained an HMM for a leucine zipper motif. Comparing to the accuracy of a symbolic representation which accuracy is 14.8 %, an HMM achieved 79.3 % in prediction. We demonstrated that this approach is applicable to the validation of the protein database; a constructed HMM has indicated that one protein sequence annotated as a “leucine-zipper like” in the database is quite different from other leucine-zipper sequences in terms of likelihood, and we found this discrimination is plausible. 2 HMMs