Comparisons of Different Feature Sets for Predicting Carbohydrate-Binding Proteins from Amino Acid Sequences Using Support Vector Machine

Proteins which can interact with sugar chains but do not modify them are called as Carbohydrate-binding proteins. These proteins have several biological importance. To predict them computationally with SVM classifier, we have developed different feature sets based on secondary structures and selective physicochemical properties of the constituent amino acids. The feature set formed with combination of both the secondary structures and physicochemical properties gives better prediction accuracy (up to 89.19%). We have also prepared an up-to-date dataset of carbohydrate-binding proteins and non-carbohydrate-binding proteins in this work.