A New Encoding Scheme to Improve the Performance of Protein Structural Class Prediction.

Based on the concept of coarse-grained description, a new encoding scheme with grouped weight for protein sequence is presented in this paper. By integrating the new scheme with the component-coupled algorithm, the overall prediction accuracy of protein structural class is significantly improved. For the same training dataset consisting of 359 proteins, the overall prediction accuracy achieved by the new method is 7% higher than that based solely on the amino-acid composition for the jackknife test. Especially for α + β the increase of prediction accuracy can achieve 15%. For the jackknife test, the overall prediction accuracy by the proposed scheme can reach 91.09%, which implies that a significant improvement has been achieved by making full use of the information contained in the protein sequence. Furthermore, the experimental analysis shows that the improvement depends on the size of the training dataset and the number of groups. [ABSTRACT FROM AUTHOR]