Accurate single nucleotide variant detection in viral populations by combining probabilistic clustering with a statistical test of strand bias

Table 3 SNV calling accuracy

Data set	Method	ShoRAH					VarScan					LoFreq
		TP	FP	FN	Rc.	Pr.	TP	FP	FN	Rc.	Pr.	TP	FP	FN	Rc.	Pr.
HCV1	Raw	38	29	0	1.000	0.567	38	206	0	1.000	0.156	37	1	1	0.974	0.974
	Fisher’s exact	32	6	6	0.842	0.842	36	179	2	0.947	0.167	29	0	9	0.763	1.000
	Bin. (σ = 0)	37	7	1	0.974	0.841	36	108	2	0.947	0.250
	B-bin., σ = 0.0004	37	7	1	0.974	0.841	37	108	1	0.947	0.255
	B-bin., σ = 0.0014	37	8	1	0.974	0.822	37	108	1	0.947	0.255
	B-bin., σ = 0.0111	38	8	0	1.000	0.826	38	108	0	1.000	0.261
HCV2	Raw	38	50	0	1.000	0.432	38	826	0	1.000	0.044	37	1	1	0.974	0.974
	Fisher’s exact	33	9	5	0.868	0.785	36	766	2	0.947	0.045	29	0	9	0.763	1.000
	Bin. (σ = 0)	34	8	4	0.895	0.810	35	577	3	0.921	0.057
	B-bin., σ = 0.0004	36	8	2	0.947	0.818	36	577	2	0.947	0.059
	B-bin., σ = 0.0014	36	8	2	0.947	0.818	36	577	2	0.947	0.059
	B-bin., σ = 0.0111	38	13	0	1.000	0.745	38	589	0	1.000	0.061
HCV2	Raw	153	2	35	0.814	0.987	175	732	13	0.931	0.193	121	0	67	0.644	1.000
	Fisher’s exact	87	0	101	0.462	1.000	125	671	63	0.665	0.157	41	0	147	0.218	1.000
	Bin. (σ = 0)	88	0	100	0.468	1.000	113	473	75	0.601	0.193
	B-bin., σ = 0.0004	101	1	87	0.537	0.990	121	474	67	0.644	0.203
	B-bin., σ = 0.0014	126	1	62	0.670	0.992	135	475	53	0.718	0.221
	B-bin., σ = 0.0111	151	2	37	0.803	0.987	162	490	26	0.862	0.248

SNV calling statistics for ShoRAH, VarScan, and LoFreq. For ShoRAH and VarScan, SNV calls without the strand bias test (Raw) and using different values of the beta-binomial dispersion parameter σ in the strand bias test are given, with σ = 0 corresponding to a binomial forward read distribution. For LoFreq, the results of applying our strand bias test are absent as this software does not report forward and reverse strand counts. For all SNV calling methods, the results of applying Fisher’s exact test to the raw output are also given (this was possible for LoFreq as it reports a strand bias Fisher’s exact test p-value for each variant). Reported statistics include true positives (TP), i.e., genomic sites with a variant matching a known true variant, false positives (FP), i.e., genomic sites with a variant that is not a known true variant, and false negatives (FN), i.e., known variants which are not identified by the relevant SNV calling method. Individual genomic sites may contribute to both true positives and false positives. Recall (TP/(TP + FN)) and precision (TP/(TP + FP)) are also reported.

ISSN: 1471-2164