Hello,
I had another question about how the pipel...
# magj
Jaime Zolik
09/11/2026, 12:29 PMHello,
I had another question about how the pipeline filters contigs. Is there a minimum depth of contigs that the mag pipeline is looking for before assembling bins? Most of the mins assembles are shown to be about 10x (but went as low as 9), I was wondering if there was a parameter that I could use to reduce that? Maybe to 2x?
j
James Fellows Yates
09/12/2026, 4:38 AMNo, I don't think we do any filtering ourselves based on depth, only contig length filtering. Right @Diego Alvarez Saravia?
d
Diego Alvarez Saravia
09/12/2026, 1:05 PMAt pipeline level we don't have anything like that, only the contig length filter.
Diego Alvarez Saravia
09/12/2026, 1:08 PMBut it's possible that assemblers (and also binners) have something like that baked to reduce complexity, so you would have to look that on each tool options.
a
Alex Hübner
09/12/2026, 5:32 PMI think the reason for the coverage to be on average around 10-fold is that assemblers usually are not able to assemble anything below that. In experiments with simulated data we could push it to 5-fold coverage in certain cases but the rule of thumb is that need about 10-fold coverage to get contigs.
As @Diego Alvarez Saravia suggested, it might have something to do how assemblers work with the assembly graph to reduce complexity, but I am not sure.
In case you want to make use of the shallower data, I think you will need to take the bins that you get out of the pipeline and align your contigs or in worst case your raw sequences against them.
j
Jaime Zolik
09/19/2026, 12:46 PMOkay that makes sense to what I am seeing, thank you all for your help!
Jaime Zolik
09/19/2026, 1:01 PMI had another question, btu more looking for advice. I am trying to use mag to validate/further investigate some results I got from Krakenuniq. I have a hit with 180800 k-mers (well over 1000 uniq k-mers). and when I aligned I get > 360,000 uniq reads, > 80% 1x coverage and >50% 5x coverage. Yet, when I run that shotgun sample through mag, it creates no bin even close to that organism (not even a bin the the same kingdom). Do you have any advice on how I should proceed with this? I can try to get more reads from this sample, but is what I found in KrakenUniq be invalidated if I cannot make this bin?
j
James Fellows Yates
09/19/2026, 3:40 PMI would say it sort of depends on a lot of factors... one could be database characteristics. If you align your KU assigned reads to the reference genome of the taxa with a normal aligner (bwa, bowtie2, etc if short reads) - how clean is the alignment? Do yo usee a lot of SNPs etc? If so, maybe it is not the same organism as the reference. In which case maybe that bin does not exist in the same way. You could also try assembling the aligned reads and see what happens there.
Is the genome in the KU database actually contaminated would be another thing I would be highly suspicious of
j
Jaime Zolik
09/29/2026, 9:30 PMI have aligned the reads to the reference genome and see that it has even coverage with 80% coverage and more than 50% 5x depth. I was hoping that the de-novo assembly could be another authentication tool to the hit. And since it is ancient DNA, I think that a de-novo assembly would be needed in order to interpret it's phylogeny (in the case the aligned reads are from a different linage not existing in modern datasets). Are you suggesting extracting the aligned reads and putting only those through nf-core/mag to see if it will assemble it as bin?