Improving single end assignment/classification

[microbemarsh]

Hi medi devs,

I was wondering if there are any parameters or additional mapping arguments you would suggest when using single end nanopore reads? I've already applied the --single_end argument but was curious if you think changing the bracken lengths would also help assign a higher percentage of my reads to the database.

Thank you!
Austin

[cdiener]

It might not work great with Nanopore or PacBio simply because of the effective depth. Nanopore has great throughput overall but the longer reads mean you also see less of vewy low abundance DNA fragments. For instance assume you have throughput of 1MBps and a sample with one organism yielding 99% of the abundance and another 1%. If your read length is 100bps you would get 9.9M reads for one organism and 0.1M reads for the other (more or less). So you detect the low abundance organism quite easily. Now assume your reads are 100Kbps long. Now you only get 10 reads per sample and is likely that they all come from organism 1. The longer the reads, the lower the dynamic range (unless the throughput goes up). So it might be that you won't find any food reads at all. However, the ones you do find will be very well annotated as much more k-mers will be assembled into the final taxonomic classification.

So for Nanopore the following would have to be changed:

1. Switch Fastp to FastpLong
2. Make a long read Bracken DB (maybe 25% quantile of Nanopore read length)
3. Tune the architeuthis cutoffs for longer reads

The rest should work out of the box.

[microbemarsh]

Sounds great! I will definitely revisit this when I get back to some of the gut microbiome work. Thank you for the update and recently improving some of the DB craziness.