Table 4.

RF prediction results

Feature set	Neg_1			Neg_2
	Optimal parameters	Se	Sp	Optimal parameters	Se	Sp
SEQ	ntreea = 2000, mtryb = 16	0.873	0.828	ntree = 1000, mtry = 66	0.821	0.885
STUR	ntree = 1500, mtry = 16	0.852	0.826	ntree = 1500, mtry = 7	0.807	0.808
POSI	ntree = 1000, mtry = 5	0.947	0.916	ntree = 1000, mtry = 4	0.917	0.949
Total	ntree = 2000, mtry = 6	0.971	0.918	ntree = 500, mtry = 37	0.870	0.922

Cross-validation was used to estimate the predictor performance of SEQ, STRU, POSI sets and the total feature set for two differet negative data sets. Neg_1 comprises all experimental samples and inferred negative samples and Neg_2 comprises all experimental samples and artificial negative samples from miRanda. Sensitivity (Se) was calculated as TP/(TP+FN) and specificity (Sp) as TN/(TN+FP), where TP is the number of true positives, TN is the number of true negatives, FP is the number of false positives and FN is the number of false negatives.

^a number of trees to grow.

^b number of variables randomly sampled as candidates at each split.