Hello everyone!
I had a question regarding the ass...
# metatdenovoa
Adrià Auladell
09/02/2025, 9:45 AMHello everyone!
I had a question regarding the assembly step. From what I am reading in the documentation it is not clear to me if during the assembly all the samples are merged together and a co-assembly is done, or if each sample is processed separately. How does that step work?
I am running with memory and disk usage problems and I am wondering if this could be the reason behind it. I am specifically assembling reads from Tara, and some of the samples have a high sequencing depth.
Thanks a ton
d
Daniel Lundin
09/02/2025, 10:02 AMBy design, the pipeline only does co-assembly. The reason is that this makes downstream analyses easier as there's just one set of contigs and features to quantify, which makes samples easily comparable. I've also seen lately that the amounts of sequences are outgrowing available memory for many users, like yours. The first option to try is to enable
--bbnormDaniel Lundin
09/02/2025, 10:07 AMI have also started to consider enabling multiple assemblies: per group or per sample (e.g. as the #CE9MS66BS pipeline does, with a separate column for assembly group). The assembly itself should not be difficult, but we would need to discuss how to best cluster the resulting contigs and/or features in order to make comparisons between samples possible. With this I mean both what clustring tool to use (Vsearch?), which sequence to target (my feeling is that features is best but that complicates quantification a bit) and whether to do this before or after quantification. An alternative would be to do an assembly of the resulting contig sets. If there's a good tool for this, I think that would be a good option.
a
Adrià Auladell
09/03/2025, 8:42 AMOk thanks a lot for the response.
Yes, I will run one version with --bbnorm to see how it compares.
• Either Vsearch or mmseqs perhaps for the merging?
• I do not know of a tool for assembling the resulting contig sets.
d
Daniel Lundin
09/03/2025, 9:03 AMI made an issue for this: https://github.com/nf-core/metatdenovo/issues/411. Please post any suggestions you come up with there. No timeline for this yet, I'm afraid. If you want to help out, you're more than welcome.
✅ 1
Daniel Lundin
09/03/2025, 9:04 AMAlso: Megahit uses much less resources than Spades. I run Megahit for all our projects. We're mostly interested in prokaryotes, but we do have a number of polyA-enriched projects.
✅ 1