A random-permutations-based approach to fast read alignment

BMC Bioinformatics

Table 6 Simulated paired-end reads: search time and percent of correct alignments.

		Low indel probability			High indel probability
Mismatch probability		1%	2%	5%	1%	2%	5%
Software	Time/%Correct
Bowtie -v 3	time (s) %correct	1766 91.8	2274 69.3	3068 6.1	1849 83.2	2375 63.8	3059 5.7
Bowtie -v 3 -k 10	time (s) %correct	6161 93.6	5283 70.5	3260 6.1	5793 84.9	5151 64.9	3246 5.8
Bowtie -v 2	time (s) %correct	241 79.7	333 42.2	353 1.3	256 73.2	337 39.6	352 1.2
Bowtie -n 2	time (s) %correct	483 86.4	593 64.8	645 50.4	486 78.4	618 59.6	664 46.1
Bowtie -n 2 -k 10	time (s) %correct	1308 87.8	1293 65.7	1054 51.1	1283 79.6	1199 60.5	1036 46.8
Bowtie2 -very-fast	time (s) %correct	757 98	715 94.6	506 67.0	734 98.0	670 94.6	484 67.0
Bowtie2 -fast	time (s) %correct	834 98.1	800 94.8	571 67.5	804 98.1	750 94.8	551 67.5
Bowtie2 -sensitive	time (s) %correct	1104 98.4	1079 97.5	893 83.8	1062 98.4	1014 97.5	857 83.8
Bowtie2 -very-sensitive	time (s) %correct	2070 98.5	2045 98.3	1671 96.5	2080 98.5	1968 98.3	1648 96.5
BWA -o 0	time (s) %correct	1026 99.1	1404 98.7	1670 78.1	1162 98.6	1379 98.0	1676 76.3
BWA -o 1	time (s) %correct	1252 99.1	1956 98.8	3062 78.4	1499 99.1	2274 98.8	3180 78.0
Permutations-based (report one)	time (s) %correct	155 98.2	208 97.9	300 95.0	166 97.6	223 97.2	313 93.6
Permutations-based (report more)	time (s) %correct	227 99.5	328 99.2	530 96.5	268 98.9	365 98.5	553 95.2

The "low indel probability" datasets were generated with mutation rate: 0.1%, and indel ratio: 15%. The "high indel probability" datasets were generated with mutation rate: 0.1% and indel ratio: 100%. Each of the datasets contains 10⁶ pairs of 100 character-long reads.
For each dataset and each program, we report the search time (for BWA: overall run time) and the percent of the reads which were aligned to the correct position in the genome.
In some cases and in some settings, the programs may report several possible alignments for some reads. When needed, additional filtering can be added to aligners in order to eliminate some of the results, as appropriate for specific applications. In the experiment, the programs reported < 3 alignments/read (in average, in sensitive modes), with 0-1 alignments for the majority of reads. In this table, when the program produces multiple possible alignments, it is enough that one of the reported alignments corresponds to the correct location in order to consider the alignment correct.

ISSN: 1471-2105