SNPs detection by eBWT positional clustering

Table 2 Post-processing comparative results of ebwt2snp (i.e. building clusters from the eBWT and performing SNP calling) and DiscoSnp++ (i.e. running KisSNP2 and kissreads2 using the pre-computed de Bruijn graph)

Tool	Param.	Wall clock	RAM (MB)	TP	FP	FN	SEN (%)	PREC (%)	Non-isol. SNP
Individual HG00096 vs reference (chromosome 22, 50818468bp), coverage 29× per sample
DiscoSnp++	b = 0	5:07	101	32,773	3719	13,274	71.17	89.81	4707/8658
	b = 1	16:39	124	37,155	10,599	8892	80.69	77.80	5770/8658
	b = 2	20:42	551	40,177	58,227	5870	87.25	40.83	6325/8658
ebwt2snp	\(\hbox {cov}=4\)	35:56	314	42,309	1487	3738	91.88	96.60	7233/8658
ebwt2snp	\(\hbox {cov}=6\)	22:19	300	40,741	357	5306	88.47	99.13	6884/8658
Individual HG00100 vs reference (chromosome 16, 90338345bp), coverage 22× per sample
DiscoSnp++	b=0	6:20	200	48,119	10,226	18,001	72.78	82.47	6625/11,055
	b=1	31:57	208	53,456	24,696	12,664	80.85	68.40	7637/11,055
	b=2	51:45	1256	57,767	124,429	8353	87.37	31.71	8307/11,055
ebwt2snp	\(\hbox {cov}=4\)	33:24	418	59,668	898	6452	90.24	98.51	9287/11,055
ebwt2snp	\(\hbox {cov}=6\)	44:53	337	53,749	190	12,371	81.29	99.64	8169/11,055

Wall clock (mm:ss) is the elapsed time from start to completion of the instance, while RAM is the peak Resident Set Size (RSS). Both values were taken with /usr/bin/time command. We recall that DiscoSnp++ makes use of multiple cores while ebwt2snp is currently designed to use one core only, thus explaining the difference in speed