Abstract
The completion of Human Genome Project has triggered a wave of investigating various biological problems directly through the string of nucleotides and also its derived amino acid sequence. Therefore, the urgent need of predicting protein three-dimensional structure simply from the amino acid sequence propels us to develop a model-based method to predict the composition of the fundamental structural elements–that is, secondary structures–of any protein chain. To accomplish this goal, we first represent all the eligible secondary structure sequences as specific paths in a secondary structure trellis. Then we employ the method of dependency graphs and their expanded Bayesian networks to quantify the relationship between primary and secondary structures. Following the similar procedure as in the coding theory, we finally assign a secondary structure element to each amino acid through the use of two decoding algorithms: the Viterbi algorithm and the sum-product algorithm. The simulation results reveal that our proposed method achieves an accuracy that is indistinguishable from other existing sequence-only methods, and that a better outcome is reached when the target sequences are confined to a specific protein fold.