I did a metaSpades assembly of a big diverse metaG bacterial community and I pulled out the 16S and single copy genes that were assembled.
For comparison to a smaller metaG, I subsampled the community to the sample sequencing depth multiple times and assembled the reads. In each of the subsamples, I got different 16S and single copy genes than in the assembly of all the reads. Why might this be? Does metaSpades get overloaded and is not able to assemble everything with all the reads?
It depends on how the subsampling is done. If it was random, it is not surprise at all that you get different assemblies as low-abundance groups may be mostly or completely missing from subsamples. If you do a digital normalization subsampling (see khmer) which preserves DNA sequences with low reads coverage, then you can expect to get a similar assembly.
Depending on the average depth in your full sample, it is possible that sequencing errors overwhelm the assembler's ability to pull DNA fragments together. It is not a matter of overloading, but simply a signal being drowned enough by noise. If the average assembly is >500-1000x, and especially if it is in many thousands, that will create an artificial fragmentation of resulting DNA contigs. Subsampling the reads down to 20-100x will potentially solve this problem, but not if the subsampling is random.
I think a more precise description of what you did will likely yield better answers.
I did random subsampling using reformat.sh from bbtools. The sequencing depth of the community is 5-fold the depth of the other community, so I wanted to randomly subsample so that it could be analogous to the other sample when comparing alpha-diversity.
I think I will try the khmer though so that we can target low abundance groups. Thank you for the suggestion!
Log in to answer this question.
