Protein function prediction using hidden Markov models and neural networks : Bioinformatics

We present a method for the prediction of protein function from its sequence information using hidden Markov models and neural networks. The hidden Markov models are used for representing the sequence order information and the neural networks are used for representing the amino acid composition information. For the hidden Markov models, the network topology is automatically designed using the iterative duplication method developed by our group. For the neural networks, 20 input units corresponding to 20 amino acid composition are used. We adopted this method to the problem of subcellular localization prediction. As a result, it shows high prediction performance. This implies that the system seems to acquire the biological features hidden in the sequences.