Yukio Kobayashi[1] (koba@iss.soka.ac.jp)
Nobuhiko Saito[2]
[1]Department of Information Systems Science, Faculty of Engineering, Soka
University, 1-236 Tangi-cho, Hachioji-shi, Tokyo 192, Japan
[2] Department of Applied Physics, Waseda University, Okubo 3-4-1, Shinjuku-ku,
Tokyo 169, Japan
Statistical mechanical method is proposed to predict the secondary structures of globular proteins. Three-state prediction which provides simultaneously the probabilities of alpha-helix, beta-strand and coil is performed with a recurrence method. The probabilities of the ith residue in alpha-helix or in beta-strand are calculated with statistical weights for amino acid pairs in alpha-helix or in beta-strand. We determine the statistical weights to yield the correct predictions for the proteins with known structures instead of calculating directly the interaction energies between residues. To do this, we introduce an objective function and estimate the weights so as to minimize this function by referring to the proteins for optimization. This method yields prediction accuracy of 67% for 13 proteins for accuracy estimation. This value does not exceed the best values obtained by the method based on homology. However, we have a hope to improve the accuracy, since we can analyze the reasons for poor accuracy in contrast to other methods.