. 2016 Apr 4;11(4):e0152964. doi: 10.1371/journal.pone.0152964

Table 4. Accuracy obtained based on the cross validation on the training datasets TRAIN_ION and TRAIN_VLG and Q₄ based on the cross validation on the TRAIN_VGS dataset by different groups of input features.

Models	TRAIN_ION (accuracy)	TRAIN_VLG (accuracy)	TRAIN_VGS (Q₄)
Model based on the PSSM profile	89.6	95.6	81.8
Model based on the dipeptide composition	84.5	87.6	65.5
Model based on the predicted relative solvent accessibility	79.8	not used	not used
Model based on the predicted secondary structure	69.9	68.1	62.2
Model based on the predicted intrinsic disorder	60.3	not used	62.2
Model based on all features	91.6	96.3	88.5

We computed a single value of accuracy based the results that are combined over all test folds (entire test datasets)

Table 4. Accuracy obtained based on the cross validation on the training datasets TRAINION and TRAINVLG and Q4 based on the cross validation on the TRAINVGS dataset by different groups of input features.