. 2019 Feb 11;15(2):e1006481. doi: 10.1371/journal.pcbi.1006481

Table 1. Specificities of variant interpretation tools.

	All variants (n = 63,197)^a				Variants predicted by all tools (n = 7,268)^b
Tools	VUS^c	Benign	Harmful	Specificity	Benign	Harmful	Specificity
PON-P2^d	21,373	34,529	1,626	0.955	6655	613	0.916
VEST^d^,^e	1,168	22,614	4,480	0.835	5984	1284	0.823
FATHMM^d	5,531	43,005	6,766	0.864	6287	981	0.865
PROVEAN	3,908	45,868	13,421	0.774	5712	1556	0.786
PPH2^d^,^f	6,386	37,124	13,602	0.732	5404	1864	0.744
LRT	19,333	31,736	12,128	0.724	5465	1803	0.752
MA	8,044	39,493	15,660	0.716	5306	1962	0.730
CADD^g	0	40,659	22,538	0.643	4539	2729	0.625
SIFT	5,099	36,808	21,290	0.634	4868	2400	0.670
MT2^h	15,313	30,632	17,252	0.640	4764	2504	0.655

^aAll variants having AF> = 1% and <25% in at least one population and not present in the training dataset for the method. After excluding cases in the training datasets, the total number of variants was 57,528 for PON-P2, 28,262 for VEST, 55,302 for FATHMM, and 57,112 for PPH2.

^bVariants classified as benign or harmful. Variants present in training dataset of any of the tools were excluded. All variants that were automatically annotated without making predictions were excluded.

^cVariants for which the predictions were not available, were ambiguous, or were predicted to have unknown significance.

^dVariants present in the training datasets were excluded.

^eVariants were not classified into benign and harmful by the program. A cutoff of 0.5 was used so that variants with score greater than or equal to 0.5 were classified as harmful, otherwise benign.

^fHumVar version of PolyPhen-2 was used as the performance was higher than for HumDiv version.

^gVariants were not classified into benign and harmful by the program. A cutoff of 20 was used so that variants with score greater than or equal to 20 were grouped as harmful and otherwise benign. The authors have recommended a cutoff ranging from 10 to 20. The highest cutoff was used so that the highest possible specificity was obtained.

^hVariants that were automatically detected to be harmful or benign were not included in the classified cases as they are not real predictions by the tool, instead annotations based on known data.