Figure 1.

Prediction performance of various SVM models. a) Prediction performance of models on a) training dataset evaluated using 5-fold crossvalidation; b) Test dataset evaluated using validation test and; c) independent dataset evaluated using validation test. Each bar represents MCC achieved by a model based on distinct input features. Hybrid-II based models achieved highest MCC both for the classification of secretion system proteins from nonsecretion system protein and at the sub-classification of secretion system proteins; d) Confusion matrix showing prediction results of SSPred in the subclassification of secretion system proteins from independent dataset. A relatively low MCC was observed for Type-II secretion systems due to many false positive predictions. This can be partly explained by the fact that many of Type-II secretion systems proteins are shared among Type-III and Type-IV secretion system proteins [21, 22].