I have a low-diversity metagenome (~11 bins > 80% completion). Out of the bins, 3 are of interest to me. None of the lower-completion bins that can be identified are from the group of interest. So let's call the bins of interest 1, 2 and 3.

The raw reads had been assembled separately with MEGAHIT and SPAdes (meta). Out of them, only the SPAdes assembly was binned / refined. There are 3 16S rRNA in the MEGAHIT assembly (x, y, z) and 3 16S rRNA in the SPAdes assembly (a, b, c).

Bin 1 is associated with sequence 'a', which is identical to 'x'.

Bin 2 is linked with SPAdes sequence 'b', which has no MEGAHIT equivalent. Bin 3 is not associated with any 16S rRNA (even with MarkerMAG).

Sequences 'c' and 'z' are very similar. Sequence 'y' is an outlier.

There are small hints from phylogenetics that bin 3 could be associated with sequences c/z. But then, could sequence 'y' just be a misassembly? Or is it that somehow my final bins are missing an entire genome of interest? If the 16S rRNA for 'y' is real, I would expect a genome divergent enough from the others that it wouldn't get binned together with them. This is what I am trying to understand. Would you happen to have any advice / suggestions on what I could to to test this?

Initial approach: assessing scaffold quality with the BBMap stats.sh . It's a bit difficult to tell, since SPAdes has about 10x more contigs. Worse N50/N90 but more long contigs and a better 'general assembly score'.

Additional thoughts: The new bins seem too divergent from known genomes to use a reference for misassembly estimation.

Thank you for your time!


1 Answer 1


This is easy @Laura - I'm a bit surprised by the results Spades meta didn't have a good reputation, but the criteria assessed was not what you are looking for.

Anyway take all outputs and place them in large multi-species alignment,

  • e.g. for each output perform a Blast for maximum genetic diversity for each sequence and sample across the large Blast output. Take note if there are multiple partial hits against diverse bacteria (online Blast has a diagrammatical summary).
  • You have a good understanding of your taxonomic group and intersperse that with other stuff, i.e. fill in the diversity.
  • Assemble it, e.g. muscle5 superfast option.
  • Perform a split decomposition phylogeny or equivalent which identifies phylogenetic networks.

SplitsTree is the historic package for this and it turns out that split decomposition was not the best, I think "neighornet" was superior. What this is does is identify mosaic gene signal in your data. It provides a visual representation. Thus if your data is a artifactual mix of two or more taxa this will cause a network effect between the two parents and your newly mis-assembled sequence will sit between the two in a two "diamond same". This is the antonym of a bifurcating phylogeny.

The final step is the assembly that produces sequences that minimises phylogenetic networks is the true sequence. If you've not encountered split-decomposition/networks before a verbal explanation might seem confusing but when you run through a few examples and try it on your data you'll get the grasp.

  • 1
    $\begingroup$ Much appreciated! Good idea. Tested. Couldn't see obvious signs of a chimera, but the alignment showed that most of the short 'outlier' sequence's distinguishing features were not present in any other (even distantly related), including a lot of weird Gs. Chimeric with something else maybe? Curious thing is, the depth of that MEGAHIT contig seemed to be pretty good (~120x). I had thought I'd done my homework by quality-controlling the reads with BBDuk, but it's a bit unsettling to see how software can be so certain and so wrong at the same time. Should look more into missassembly detection... $\endgroup$
    – Laura
    Aug 28, 2023 at 14:16
  • $\begingroup$ It tentatively sounds like you might have found a new species. Just keep in mind the diversity of bacteria in the analysis affects the ability to detect a possible chimera, i.e. if the "parents" are not present it could throw a false negative -------- Weird "Gs" high AG are a known feature of bacteria: particularly if its genetically distinct. Anyway if Splitstree if done correctly it's a sensitive method for chimeras: but you might have to shift the sample size - remove distant sequences which can cause a collapse in networks. Megahit is respected BTW. $\endgroup$
    – M__
    Aug 28, 2023 at 14:26
  • 1
    $\begingroup$ It's more like...The positions in questions are 100% conserved within related bacterial orders, and here's this short sequence with ~9 of these otherwise conserved positions being different. I guess I can't prove it's not real (part of my issue...), but it doesn't look...correct? $\endgroup$
    – Laura
    Aug 28, 2023 at 14:32
  • 2
    $\begingroup$ What you are saying sounds perfectly reasonable - keep in mind I can't see the data nor analysis. Proof ... it's the first stage requiring downstream analysis, so either leave it as a "putative result" or go into the complexity of metagenome assembly by assembling other single copy gene targets like atpD. $\endgroup$
    – M__
    Aug 28, 2023 at 14:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.