Secondary Structure Prediction Based on Statistical Mechanics

Recent genome science has been developing from two sides: one is biological evolution due to mutation of DNA sequence and the other is biological function of protein molecules formed according to amino acid sequences. We have devoted ourselves to elucidate the mechanism of protein folding, and to develop the method for protein structure prediction [1, 2]. Our method has successfully yielded the native-like tertiary structures of various proteins, by using secondary structure information. Secondary structure prediction with a high precision, however, is essential to proteins of unknown structures. Our method for three-state (α-helix, β-strand and coil) prediction have reached the prediction accuracy of only 68%, which accuracy may not be sufficient for packing secondary structures into correct tertiary structure. Some methods [3] for secondary structure prediction exceed a little the accuracy achieved by our method. Contrary to the other methods, we can analyze the reasons for the poor accuracy and thus are attempting to improve the present accuracy. In this article, we describe the way to improve the prediction accuracy presented in Genome Informatics Workshop IV (GIW ’93) [4].