Benchmark vs `bwa aln`

[RagnarGrootKoerkamp]

See #30

[RagnarGrootKoerkamp]

@tfenne @nh13

Can I replicate your results with plain bwa?

I did ./bwa index $d/chm13v2.0.fa (takes ~40 minutes) and now running

./bwa aln -l 23 -M 1 -O 0 -E 1 -R 1000000 -m 999999999 -n 5 -N $d/chm13v2.0.fa crispr.fa

Flags:

• -M1 for mismatch cost 1, -O0 for no gap open cost, -E1 for gap-extend of 1 (so it's edit dist)
• -R 1000000 for reporting many matches
• -m 999999999 for large queu size
• -n 5 for allowing up to 5 errors
• -N for exhaustive mode

crispr.fa: 10 patterns

>crispr1
AGTATCCATATCCGCATCGTNGG
>crispr2
AAATGTAGTTTTTGTAGACANGG
>crispr3
TTTTTTTGTTCCTCTTTTGGNGG
>crispr4
CCTCCCTCTGTCCCCTCCTCNGG
>crispr5
TTGCTGGTGTTATATTCTGANGG
>crispr6
TGGTGGTTGAGCCAGAAGCTNGG
>crispr7
CCTGCTGCCTGTGTGACTTTNGG
>crispr8
GGAGGAGGTCAGTAGGCAGGNGG
>crispr9
CAGGCTGCAGCCCTGGGCTGNGG
>crispr10
AGGCTGCAGCCCTGGGCTGGNGG

This runs in around 5 seconds, so relatively fast.

But still, the resulting .sai file converted back to .sam (./bwa samse $d/chm13v2.0.fa aln.sai crispr.fa > aln.sam) contains only 10 matches, so I assume a single best for each.

---

edit: also tried adding -k 5 but that doesn't seem to change anything, nor the output nor the runtime.

[nh13]

Consider YARA too, as that's what I use for exhaustive matching currently

[RagnarGrootKoerkamp]

This one right? Never heard of it:
https://github.com/seqan/dream_yara

[RagnarGrootKoerkamp]

Ah there's an earlier PhD thesis: https://refubium.fu-berlin.de/handle/fub188/11364

and the masai paper: https://academic.oup.com/nar/article/41/7/e78/1068067 (not sure how it relates to yara)

and the code: https://github.com/seqan/seqan/tree/main/apps/yara

But ok, looks like YARA is a mapper using an index. That means it'll surely be faster if you have sufficiently many queries (>1000 will probably do) to offset the time spend building an index.

Note that it's unlikely we'll run such benchmarks ourselves in the near future. Sassy kinda serves a different purpose, and if you throw enough queries at it, a well-designed index will always be faster. I haven't read how YARA works, but something like Columba (https://github.com/biointec/columba) should be a good candidate index to try.