Figure 1

SVM training of SCOP and PROSITE. 5751 protein instances common to both PROSITE and SCOP were taken to train the respective SVMs. 30 most populated classes in PROSITE and 37 most populated classes in SCOP were used. Randomly half of 5751 proteins were used to train the SVM classifier for PROSITE and other half for SCOP. Blocks, elastic 2-D profile, molecular mass, size, percentage of helices, β chain were used as orthogonal features for 1 vs rest SVM training for each class.