If you generate, or use MAGs, you should really read the work we have done benchmarking their limitations. Using complex mock communities (70 species), we assessed how deeply you need to sequence to accurately conduct common metaG analyses. Webwere shocked to see even HQ MAGs are often chimeric! - ThreadSky

tcahitch.bsky.social • 39 days ago

If you generate, or use MAGs, you should really read the work we have done benchmarking their limitations. Using complex mock communities (70 species), we assessed how deeply you need to sequence to accurately conduct common metaG analyses. Webwere shocked to see even HQ MAGs are often chimeric!

Reposted from bioRxivpreprint

Benchmarking of shotgun sequencing depth highlights strain-level limitations of metagenomic analysis https://www.biorxiv.org/content/10.1101/2025.03.27.645659v1

Comments

drxenogene.bsky.social•39 days ago

My lab cultivated a number of microbes from the murine gut and compared the genomes of the isolates to genomes from metagenomic assemblies. It wasn't encouraging. MAGs for some species were pretty good but most MAGs were chimeric.

stevenjrobbins.bsky.social•39 days ago

Can I ask what you mean by “chimeric?” Are you saying that the MAGs are mixes of strains or mixes between higher orders—eg species? The former I would say is a known limitation of short read de novo assembly, the 2nd would be more surprising. Good to define at what level you’re asserting chimerism.

stevenjrobbins.bsky.social•39 days ago

I would personally not call a MAG “chimeric” for representing multiple strains. One of the reasons we call them population genomes, not isolate genomes.

stevenjrobbins.bsky.social•39 days ago

Also curious how many samples were used here for differential coverage binning?

treichelnicole.bsky.social•39 days ago

The current data set is not suitable to test the effect of multi-coverage binning, but we are discussing an extension of the project with focus on MAG analysis.

stevenjrobbins.bsky.social•36 days ago

Thanks Nicole! See my comment on SingleM/Sandpiper for picking public metagenomes that could be used for differential coverage binning, without having to generate new data. Just a suggestion for extending the study. 🙂

treichelnicole.bsky.social•36 days ago

Thank you for drawing my attention to this tool. I will definitely have a look at it :)

treichelnicole.bsky.social•39 days ago

We focussed on benchmarking shallow metagenomics in general and used bacterial DNA Mock communities with a defined composition. MAGs were constructed by single-coverage binning at different sequencing depth.

treichelnicole.bsky.social•39 days ago

Thanks for the interest in our work and your feedback. We chose “chimeric” to describe fragments of higher order incorporated in a MAG. To investigate the strain-level we had added 4 strains of each E. coli and P. vulgatus in our Mock community, which our analysis wasn’t able to resolve.

stevenjrobbins.bsky.social•39 days ago

Got it, and thanks. :-) Would be good to have some breakdown of what sorts of contigs are being misbinned? From different species? Different genera? And I like studies like these for showing some pitfalls of MAG generation, but would be good to pitch how these these results apply a bit more narrowly

tcahitch.bsky.social•38 days ago

This is a nice idea, thanks! Making a taxonomic summery of the strains would definitely help us highlight how diverse they are

stevenjrobbins.bsky.social•39 days ago

than saying they apply to "MAGs". We know that single sample binning will result in some level of misbinning. For that reason though, single sample binning should probably be considered a sub-optimal practice, not the standard.

https://www.nature.com/articles/s41592-023-01934-8

stevenjrobbins.bsky.social•39 days ago

Reason being, I don't think you'd see this level of chimerism with best practice multi-sample binning--eg >8-10 samples. And long reads would make this less of a problem still. It's an interesting study i'd be happy to cite, just want to frame in the right context--single sample binned MAGs.

tcahitch.bsky.social•38 days ago

Out of interest, how does this work with the SeqCode where you need to state which samples your MAG originated from?

stevenjrobbins.bsky.social•39 days ago

Don't want people thinking this level of chimerism is a wide-spread problem with everybody's MAGs.

taylorpriest.bsky.social•39 days ago

Along similar lines to @stevenjrobbins.bsky.social - differential coverage is essential for reducing binning errors. As you didn't have many samples to generate differential coverage info, I imagine that played a big role in the results you obtained.

stevenjrobbins.bsky.social•39 days ago

Right. I couldn’t quite tell how many sample were used for differential coverage. If less than 7 or 8, misbinning with short read contigs is somewhat expected.

https://www.nature.com/articles/s41592-023-01934-8

stevenjrobbins.bsky.social•39 days ago

Also reminds me of Titus Brown’s old preprint comparing strain level assembly and binning. They also showed that similar strains essentially fail or coalesce into one genome during de novo genome assembly.

https://www.biorxiv.org/content/10.1101/155358v3.full.pdf

taylorpriest.bsky.social•39 days ago

In addition, it would be good to see whether you get the same results using Spades, instead of megahit, as it often can be a more accurate assembler

drbfx.bsky.social•39 days ago

Would be interesting to see if you can overcome these limitations with long reads

Comments

Posting Rules

Reply