Figure 2

Cross training flow. Datasets are generated by cross training, where the taxonomy A (or B) had features as classes from taxonomy B (or A) respectively. In effect, classifier of PROSITE is trained using classes of SCOP as features and vice versa. SVM classifiers were created for both PROSITE and SCOP (Figure 1). Classes of PROSITE were used as features for SCOP and protein feature vector was updated. Similarly, classes of SCOP were used as features for classifier of PROSITE and protein feature vector was updated. Cross training was iterated till further gain in accuracy stops.