PacBio HiFi read mapping - scaling to larger dataset?

[royercj]

I have a eukaryote-derived long read metagenome set I am interested in mapping against ATB to isolate bacteria-only sequences. The reads are PacBio HiFi, mean length 17.5 kb and there are 5k reads in the set. I'm using AWS to mount and query the LexicMap ATB index so I'm trying to keep costs low. After reading more of the LexicMap documentation I wonder if these reads are too long or the overall set of 5k is too large to map with LexicMap in 24/48 hrs? Would it be better to remove host reads another way (like Deacon/Hostile) and then map a smaller set with LexicMap? I appreciate any advice you might have for me. Thanks

[shenwei356]

Thanks for considering LexicMap.

It would be slow for 5k reads with default parameters.

But if you only care about whether queries can be matched to any genomes in the database, you can add -N/--top-n-genomes to increase the speed.

  -n, --top-n-genomes int              ► Keep the top N genome matches for a query (0 for all) in the
                                       chaining phase. Value 1 is not recommended as the best chaining
                                       result does not always bring the best alignment, so it better be
                                       >= 100. (default 0)

And in the next version (pre-release is available here, a new option -N/--top-n-chains can help to reduce searching time, but mainly for big genomes like human.

  -N, --top-n-chains int               ► Keep the top N chains in a genome for the query (0 for all)
                                       in the chaining phase. Value 1 is not recommended as the best
                                       chaining result does not always bring the best alignment, so it
                                       better be >= 10. (default 0)

Additionally, you can also use bigger -p/--seed-min-prefix (default 15) and -P/--seed-min-single-prefix (default 17), e.g., -p 17 -P 19, to increase the search speed at the cost of decreased sensitivity for distant matches (similarity < 90%) or short queries. More tips to speedup searching.

Removing host reads is a good idea, but note that it might remove some reads similar to the host genomes.

For the reference database of LexicMap, in your scenario, a smaller dataset is recommended, such as GTDB representative (one genome per species). ATB is too redundant, and the number of genomes is uneven between species.

[shenwei356]

If the reads are all long enough, genome sketching based methods can also work, such as skani and sourmash.