Table 3.

Spurious Arrays in the CRISPRCasFinder datasets (27). To detect arrays that are likely false positive candidates, we first filtered arrays that are not covered by the parameter setting of the different tools (i.e. ≤3 repeats, or repeat length <23 or >55). The remaining number of arrays is listed in the row standard arrays. Arrays that were labeled as likely negative (see Methods) if they had one of the following four defects: 1) damaged repeats, 2) high similarity between spacers, 3) uneven distribution of spacer lengths and 4) high similarity between repeats and spacers. The following rows list how many of these likely negative arrays had specific deficiencies (see text for details). Note that some arrays can have several problems, hence there are multiple entries in these rows. The FDR is calculated considering all likely negatives as true negatives, as we do not have other information. This FDR values are, however, likely overestimates. Nevertheless, they show the dimension of the problem

		CRT	CRISPRCasFinder	Detect
Arrays declared positive		1866	3263	764
Standard arrays		1358	1395	346
Likely negative arrays		602	628	110
of which are arrays	with damaged repeats	165	189	40
	with similar spacers	57	0	13
	with uneven spacers lengths	473	508	83
	with sim. repeats to spacers	23	4	11
FDR		44,3%	45,0%	31,8%
likely positive arrays		756	767	236