Ratatosk: hybrid error correction of long reads enables accurate variant calling and assembly

Table 4 Small variant calls accuracy (in %) for the ONT reads from the Ashkenazim trio in the high confidence regions. Best results are highlighted

		SNPs			Indels
		Precision	Recall	F1	Precision	Recall	F1
(a) Variants with quality scores below a threshold automatically computed by rtg-tools are filtered out
HG002	Raw	98.87	97.93	98.25	81.83	39.26	53.06
	FMLRC	96.42	96.31	96.37	89.96	80.55	85.00
	Ratatosk	97.84	99.10	98.47	92.10	89.19	90.62
HG003	Raw	99.16	98.85	99.00	84.00	47.62	60.78
	FMLRC	97.73	97.56	97.64	92.78	85.38	88.93
	Ratatosk	98.94	99.42	99.18	94.04	92.14	93.08
HG004	Raw	99.19	98.73	98.96	83.67	45.50	58.94
	FMLRC	97.22	97.44	97.33	90.11	84.52	87.23
	Ratatosk	98.56	99.46	99.01	92.65	91.38	92.01
(b) All variants with the FILTERfield set to PASS in the VCF files are used
HG002	Raw	85.32	99.69	91.94	11.60	72.96	20.02
	FMLRC	78.08	99.57	87.52	60.20	93.47	73.23
	Ratatosk	90.22	99.82	94.78	86.07	96.09	90.80
HG003	Raw	95.15	99.74	97.39	15.44	77.19	25.73
	FMLRC	86.73	99.58	92.72	69.99	93.26	79.97
	Ratatosk	95.94	99.80	97.83	88.38	95.83	91.95
HG004	Raw	93.34	99.80	96.47	14.83	75.98	24.81
	FMLRC	84.08	99.68	91.21	66.87	94.60	78.35
	Ratatosk	93.05	99.87	96.34	87.77	96.69	92.02

ISSN: 1474-760X