Hidden Markov Model to extract leucine zipper motif

Abstract

This paper describes a method to construct a stochastic motif represented by a Hidden Markov Model (HMM), which consists of probabilities representing the amino acid variations and the length flexibility of a motif. To express the stochastic motif with high accuracy, the optimal HMM topology must be obtained according to the motif's characteristic. To obtain the optimal HMM topology from unaligned sequences of a motif, the learning method, named the iterative duplication method, has been developed. It can automatically represent the subfamily branches and the periodic topology for α-helices. In this method, a small fully connected HMM is gradually expanded by splitting appropriate states one by one. The several techniques for splitting state selection are experimented for leucine zipper. The obtained HMMs contain the subfamily branches and the periodic topology relying on necessities. The experimental results show that the HMMs are more powerful than patterns in terms of prediction performance.