Table 1.

The accuracy of DADA2, UPARSE, MED, mothur and QIIME on three mock community datasets

			Output Reads(%)	Output Sequences					Reference Strains
				Total	Reference	Exact	One Off	Other
Balanced	Forward	DADA2	99.2	93	59	33	1	0	57
		UPARSE	99.1	81	48	29	2	2	53
		MED	95.5	86	59	5	22	0	57
		Mothur	96.3	249	44	25	15	165	49
		QIIME	99.2	378	51	34	3	290	54
	Merged	DADA2	96.2	87	57	29	1	0	55
		UPARSE	94.2	76	45	27	2	2	50
		MED	91.1	64	56	6	2	0	54
		Mothur	94.1	108	42	27	11	28	47
		QIIME	94.1	170	45	28	4	93	50
HMP	Forward	DADA2	95.1	151	23	112	8	8	21
		UPARSE	96.7	161	20	123	10	8	21
		MED	80.9	83	23	2	58	0	21
		Mothur	95.4	849	20	177	47	605	21
		QIIME	97.4	1375	20	177	60	1118	21
	Merged	DADA2	92.3	67	23	40	2	2	21
		UPARSE	67.7	94	20	59	2	13	21
		MED	64.8	32	23	3	6	0	21
		Mothur	62.1	121	20	82	9	10	21
		QIIME	67.6	290	20	71	8	191	21
Extreme	Forward	DADA2	99.5	68	26	35	3	4	23
		UPARSE	99.5	74	21	40	0	13	21
		MED	86.4	95	16	0	79	0	13
		Mothur	*	*	*	*	*	*	*
		QIIME	99.5	3237	20	44	73	3100	20
	Merged	DADA2	97.6	25	24	1	0	0	21
		UPARSE	69.9	23	18	4	0	1	18
		MED	67.6	32	17	0	15	0	14
		Mothur	94.3	44	23	14	0	7	23
		QIIME	69.9	36	19	8	1	8	19

After a common filtering step, methods were applied to the forward reads, and the merged forward and reverse reads, of the Balanced, HMP and Extreme datasets (Methods). Output sequences were classified as Reference or Exact (true positives) and One Off or Other (false positives) by comparison to the known sequences of these mock communities (reference strains) and comparison to nt to identify contaminants (Methods). DADA2 detected the most reference strains and sequences, while outputting the fewest false positives. Mothur failed to complete on the Extreme forward reads.