Table 2.

Performances of MTBAN and other predictors on the test dataset consisting of 1709 deleterious and 1709 benign variants.

Predictor	ROC-AUC	PR-AUC	Accuracy	MCC	Precision	Specificity	Sensitivity	F-score	NPV
MTBAN	0.883	0.878	0.787	0.585	0.739	0.686	0.887	0.806	0.859
mutationTCN	0.873	0.87	0.763	0.548	0.706	0.624	0.902	0.792	0.865
SIFT	0.856	0.861	0.77	0.55	0.728	0.671	0.868	0.792	0.833
MutationAssessor	0.855	0.849	0.763	0.535	0.722	0.686	0.843	0.778	0.819
PolyPhen-2	0.853	0.856	0.759	0.537	0.703	0.637	0.885	0.783	0.851
fathmm-MKL	0.844	0.812	0.743	0.518	0.681	0.567	0.918	0.782	0.873
phyloPa	0.836	0.838	0.753	0.532	0.865	0.905	0.602	0.71	0.693
DANN	0.814	0.775	0.753	0.51	0.722	0.68	0.825	0.77	0.794
phastConsb	0.789	0.829	0.749	0.506	0.711	0.657	0.84	0.77	0.803
GERP++	0.778	0.74	0.714	0.435	0.757	0.795	0.635	0.69	0.684
MPC	0.772	0.762	0.68	0.369	0.73	0.772	0.591	0.653	0.644
GenoCanyon	0.742	0.748	0.657	0.323	0.626	0.53	0.783	0.696	0.708

Since the score cutoffs for phyloP, DANN, phastCons, GERP++, MPC, and GenoCanyon were not provided by dbNSFP, we computed the cutoffs for each predictor using the Humsavar database (release 03/2021) as described in “Methods” section. The highest values for each evaluation metric are indicated in bold.

ROC-AUC, Receiver Operating Characteristic Area Under Curve; PR-AUC, Precision-Recall Area Under Curve; MCC, Matthews Correlation Coefficient; NPV, Negative Predictive Value.

^aPhyloP100way_vertebrate from dbNSFP.

^bPhastCons100way_vertebrate from dbNSFP.