IMPROVEMENTS IN THE PREDICTION OF PROTEIN BACKBONE TOPOGRAPHY BY REDUCTION OF STATISTICAL ERRORS

We have simplified our previously published method for predicting the occurrence of residues in one of five conformational states [Maxfield, F. R., & Scheraga, H. A. (1976) Biochemistry 15, 5138] without sacrificing accuracy. An increase in the size of the data set from 3681 to 5082 residues led to a slight (1-2%) increase in the accuracy. In order to overcome the limitations in the accuracy due to statistical errors, we tested the usefulness of averaging the predictions for several homologous proteins at each position in the aligned sequence. When this procedure was used on 15 cytochrome c sequences and 24 globins, the accuracy of the predictions increased by 8 and 6%, respectively. Averaging did not improve the accuracy when only a few homologous sequences were available or when there was only a slight variability in the amino acid sequence. The improved accuracy from the use of homologous proteins and the slight improvement from an increase in the size of the data set are consistent with the hypothesis that statistical errors place a significant limitation on the accuracy of predictions which incorporate pairwise interactions between neighboring residues. Since a large increase in the size of the data set will be required to reduce the statistical errors significantly, the use of homologous sequences appears to be the most promising way to improve the predictions. © 1979, American Chemical Society. All rights reserved.