Protein secondary structure. Studies on the limits of prediction accuracy

Abstract

A secondary structure prediction technique is proposed which includes nucleation site determination through multiplication of conformational preference parameters as well as weighting factors to represent structurally stabilizing short range interactions. The prediction accuracy of the method is calculated using data bases categorized according to the four protein structural classes and with differing assignments of secondary structural regions. The results indicate that nearest neighbor prediction techniques (a) are insensitive to various assignment criteria for the secondary structural spans, (b) have nearly achieved their upper limit of prediction accuracy, and (c) can be somewhat improved through the use of stereochemical weighting factors and conformational parameters derived from the four structural groups.