Table 5. Binary classification performance of four different tools for single amino acid substitutions in human and non-human proteins.

		Human dataset				Non-human dataset
Tool	Threshold	Balanced accuracy	Sensitivity	Specificity	No prediction	Balanced accuracy	Sensitivity	Specificity	No prediction
PROVEAN	−2.282	78.75	78.39	79.11	0	77.75	80.22	75.27	0
Mutation Assessor	0.800	68.57	96.54	40.59	317 (0.55%)	69.15	93.17	45.13	732 (2.39%)
	1.900	78.15	85.29	71.02		74.23	81.30	67.16
SIFT	0.050	76.99	85.03	68.95	1147 (1.99%)	78.36	87.45	69.27	1539 (5.03%)
PolyPhen-2	0.432	75.56	88.68	62.45	2279 (3.95%)	76.79	87.77	65.81	1499 (4.90%)

“Balanced accuracy” is a simple average of sensitivity and specificity, that is, (sensitivity+specificity)/2. The “No prediction” column shows the number of variants for which the tool fails to provide a prediction.