PREDICTION OF PROTEIN SECONDARY STRUCTURAL CLASSES FOR NON-HOMOLOGOUS PROTEIN DATABASE
从非同源蛋白质的一级序列预测其结构类

For the non-homologous protein database suggested here, the comparison of the predictive methods of the amino-acid composition-based approach, the auto-correlation function-based approach and the auto-covariance function-based approach are presented. The prediction by combining the above three features is investigated. It is found that the predictive accuracy could be remarkably improved by the methods of combining the amino-acid composition with the auto-correlation functions and the amino-acid composition with the auto-covariance functions. In the amino-acid composition with auto-correlation function-added approach, the overall resubstitution accuracy is 95.34%, the overall accuracy of Jackknife test is 81.92% and the overall accuracy of the cross-validation test is 86.61% when Miyazawa and Jernigan's index is used. In the amino-acid composition with auto-covariance function-added approach, the overall resubstitution accuracy is 96.71%, the overall accuracy of Jackknife test is 82.19% and the overall accuracy of the cross-validation test is 86.88% when Wold's index is used. It is shown that how to extract more information from the primary protein sequence is the key to promote the classifying accuracy.