Evaluating CRISPR Spacer Detection Accuracy

Eight genomes containing between zero and seven CRISPR loci were chosen at random from CRISPRdb (20 (link)) (Table 1). Synthetic Illumina data sets (101 bp reads) representing ∼20× coverage were generated for each genome using Grinder 0.4.5 (21 (link)); command-line options: –cf 20 –rd 101 –md poly4. Crass 0.3.1 was run on each data set using a kmer length of 9 (all other parameters default). The spacers identified by Crass were mapped onto the reference genome using blastn 2.2.25+ (22 (link)) to determine whether they were correctly positioned. The spacer graphs for each data set were also analysed to determine whether the ordering of spacers generated by Crass accurately reflected the CRISPR loci found in the original genome assembly.
Table 1.

Specificity and sensitivity analysis of Crass on synthetic short read data sets

	Total spacers	Detected spacers	Missing edges	Erroneous edges	Specificity	Sensitivity
Bacteroides fragilis YCH46
CRISPR1	9	7	3	0	1	0.63
Acinetobacter sp. ADP1
CRISPR1	6	6	0	0	1	0.83
CRISPR2	21	21	0	0	1	1.00
CRISPR3	90	88	2	0	1	0.98
Sulfolobus solfataricus P2
CRISPR1	102	102	0	5	1	0.95
CRISPR2	94	94	0	0	1	0.96
CRISPR3	31	31	0	0	1	1.00
CRISPR4	95	95	0	3	1	0.97
CRISPR5	6	5	1	0	1	0.80
CRISPR6	22	22	0	0	1	1.00
CRISPR7	65	64	1	0	1	0.98
Natrialba magadii
CRISPR1	27	18	11	0	0.89	0.58
Helicobacter pylori B8	0	0	N/A	N/A	1	N/A
Magnetospirillum magneticum AMB-1	0	0	N/A	N/A	1	N/A
Tsukamurella paurometabola	0	0	N/A	N/A	1	N/A
Oligotropha carboxidovorans OM5	0	0	N/A	N/A	1	N/A
Overall	568	553	18	8	0.99	0.89

Crass was used to examine synthetic data sets constructed from four genomes that contained between one and seven CRISPR loci, in addition to four genomes that did not contain CRISPRs. The specificity of Crass was calculated by determining the number of detected spacers that did not originate from CRISPRs; the sensitivity was determined by comparing the reconstructed spacer ordering to the ordering found in the genome.