miTAR: a hybrid deep learning-based approach for predicting miRNA targets

BMC Bioinformatics

Table 5 Performance evaluation metrics for the unified model, miTAR, using MirTarRAW test datasets and two independent test datasets

Model: dataset	Accuracy	Sensitivity	Specificity	F-score	PPV	NPV	Brier score
miTAR: MirTarRAW Test set	0.9627	0.9591	0.9663	0.9627	0.9664	0.9589	0.0321
miTAR: MirTarRAW (30 times)^a [95% CI]	0.9549 [0.9496–0.9610]	0.9538 [0.9418–0.9629]	0.9560 [0.9428–0.9657]	0.9548 [0.9489–0.9610]	0.9559 [0.9443–0.9662]	0.9540 [0.9424–0.9623]	0.0393 [0.0340–0.0438]
miTAR: DeepMirTarLeft	0.9770	0.9706	0.9844	0.9783	0.9862	0.9668	0.0200
miTAR: miRAWLeft	0.9476	0.9500	0.9452	0.9485	0.9471	0.9482	0.0440
miTAR: DeepMirTarIn	0.9375	0.9375	NA^b	NA^b	NA^b	NA^b	0.9254
miTAR: miRAWIn	0.9505	0.9494	0.9517	0.9514	0.9535	0.9474	0.0416

^aEvaluation was done by randomly running on the MirTarRAW datasets 30 times. The average value and the 95% confidence interval (given in []) are reported here. Details are in “Results” section
^bBecause DeepMirTarIn only contains positive miRNA:target pairs, the specificity, F-Score, PPV and NPV cannot be calculated. NA represents the value is not available

ISSN: 1471-2105