How to speed up MMseqs2 taxonomy assignment with GTDB database on large contigs?

[lucianhu]

Hi nf-core/funcscan,

I am running the nf-core/funcscan pipeline to assign taxonomy to contigs using MMseqs2 with the GTDB database. My contigs are around 100-200 MB in size, and I am running the pipeline on a machine with the following specs:

• 36 cores
• 256 GB RAM

Despite utilizing all available resources, mmseq2 takes more than 4 hours per sample and does not finish. I am wondering if this runtime is normal or if there are ways to optimize the process to make it faster.

Questions:

1. What are the common bottlenecks when running MMseqs2 with the GTDB database, and how can I address them?
2. What is the expected runtime for MMseqs2 on contigs of this size?
3. Are there specific MMseqs2 settings (e.g., sensitivity, database partitioning) that could help speed up the analysis without compromising too much accuracy?

Any advice or insights from your experience with MMseqs2 and GTDB would be appreciated!

Thanks

[jfy133]

Being discussed on slack: https://nfcore.slack.com/archives/C02K5GX2W93/p1729236630834749?thread_ts=1729236630.834749&cid=C02K5GX2W93