A community effort to collect a curated set of analysis pipelines built using Nextflow.

nf-core

Hello everyone!
 I have a quick question about *MMseqs2 taxonomy*. I’m using this module to perform taxonomic annotation of my metagenomic samples with the *nrdb* database to achieve maximum accuracy. However, each sample takes more than *7 hours* to process.
Is there any way to *reduce the runtime* without significantly *compromising sensitivity or accuracy*?

It sort of depends partly based on on the computing resources you've supplied to the module? How many CPUs and RAM?

But you've picked one of the largest databases, I'm not at all surprised that it takes a long time - 7h doesn't sound that unreasonable

The other factor is iirc MMSEQS has a few 1cpu steps tha result in bottlenecks where it's just writing files. So it also depends what the  log says...

Thank you! I’m using 16 CPUs and 72 GB of RAM per sample. Maybe it needs more CPUs? I’m also wondering if there are any parameters to reduce the runtime. I selected this db for maximum accuracy.

Is there an observed plateau when throwing more CPUs doesn't make any sense?

I don't know, I've not got that much experience with mmseqs2 other than loathing debugging it :sweat_smile:

&gt; Thank you! I’m using 16 CPUs and 72 GB of RAM per sample. Maybe it needs more CPUs? I’m also wondering if there are any parameters to reduce the runtime. I selected this db for maximum accuracy.
Yeah that's probably why, I would normally be running it on a 256GB RAM HPC node for speed. I understand you want maximum accuracy, but there is a tradeoff

Maybe you can look into the MMSeqs manual or general search to see if there are ways to speed it up, and you can insert those parameters/options via `ext.args`