Carbohydrate binding site prediction accuracy benchmarks for 10-fold cross validations and independent tests.

The carbohydrate binding site predictions were carried out with the ANN, SVM and ANN_BAGGING algorithm on the proteins from the S497 dataset and the prediction accuracy of the ANN_BAGGING on the independent test set S108. Matthews correlation coefficient (MCC), F-score(Fsc), Accuracy(Acc), Precision(Pre), Sensitivity(Sen) and Specificity(Spe) are shown in Equations (4)∼(9). The benchmarks for the ten-fold cross validation results (i.e., S497/ANN, S497/SVM, and S497/ANN_BAGGING) are shown in two ways: The values enclosed in parenthesis are the averaged benchmark value and standard deviation calculated from the ten-fold results; the values above the average±standard-deviation pairs are the overall benchmark values calculated with the combined TP, TN, FP, FN cases from the test sets of the ten-fold cross-validation results.