Prediction of Protein Structure Classes and Secondary Structures by Means of Hidden Markov Models

摘要

This study deals with structure class/secondary struc - ture prediction of proteins using hidden Markov models (HMMs). With the proposed method, prediction is per- formed using HMMs designed so as to represent hierarchi - cality and periodicity of protein structural features. Secondary structures (partial tertiary structures) are formed from amino acid sequences while tertiary structures are formed through packing of secondary structures; thus, hier archicality of protein structure is represented by means of hierarchical combination of multiple HMMs. In so doing, the tertiary HMM is built from a sequence of secondary structure segments while the secondary HMM is built from amino acid sequences. In addition, periodicity is introduced into the HMM network topology so that Periodical struc- tural features can be represented. Transition probabilities and output probabilities are determined through learning of data related to known structures. HMMs designed as men- tioned were applied to the prediction of unknown struc- tures. Accuracy above 50 was achieved for structure class prediction. Besides, high accuracy prediction of secondary structure was obtained for α class and α/β class. Thus, the proposed method proved to offer faithful representation of protein structural features.