Improvement of recognition speed protein tertiary structure prediction using hidden Markov model

Protein is a substance directly connected to the biological phenomena and its biological functions are said to be derived from its tertiary structure. Thus, clarification of its tertiary structure leads to an explanation of the biological phenomena process. It is rather easy to determine the sequences of the amino acids that the protein consists of. On the other hand, it is very difficult to determine the tertiary structure. The tertiary structure of the protein is believed to correspond to the conformation with the lowest residual potential energy. In this paper, we propose a non-heuristic fast decoding algorithm which is based on the Hidden Markov Model. The decoding algorithm reduces the repeated computations of the state sequences of amino acid HMMs by breaking up the computations into two stages: In the first stage, the unknown protein sequence is matched with each individual amino acid HMM and the likelihoods of all state sequences through the model are computed, and the best likelihoods are stored in separated arrays for further use. In the second stage, we concatenate the previous obtained arrays according to the composition of the protein sequences in the library (protein database) and the overall likelihood for the unknown protein sequence is computed. The simulation results show that the proposed decoding algorithm is faster than the conventional Viterbi algorithm for protein sequence recognition tasks, while maintaining the same recognition accuracy.