Using hierarchical hidden Markov models to perform sequence-based classification of protein structure

In the post-genome era, as an essential filternative of experimental method, the computational method is becoming popular. The prediction of protein structural class from protein sequence becomes one of research´s concerns because the knowledge of protein structural class can simplify and accelerate in the computational determination of the spatial structure of a newly identified protein. As one of sequence-based approaches, hidden Markov model(HMM) provides a convenient and effective tool to analyze and classify protein sequence. In this paper, we firstly present the 6-state HMM which holds fewer states, clear transition groups and fewer model parameters. Then, by considering the knowledge of hierarchical structure of protein based on the 6-state HMM, we further propose the hierarchical hidden Markov model (HHMM) which has not only clear biological meaning, but also fewer number of transitions. Finally, the experimental comparison of various methods demonstrates that both the HHMM and the 6-state HMM outperform other method.