Predicting protein structural class based on hidden Markov models

In this paper, we describe an application of hidden Markov model (HMM) to the classification of the protein structural class given the primary and secondary structure sequence of a protein. We proposed 3-state, 7-state and 8-state HMMs, and applied these HMMs to predict protein structural class, respectively. We evaluated their accuracy on two different datasets through the rigorous jackknife cross-validation test. The results show that the prediction ability of 7-state HMM is better than 3-state and 8-state HMMs. The major advantage of the proposed HMMs is that a small number of states is employed and the training algorithm is of low complexity and thus relatively fast.