On a public oral ONT metagenome (from @ykiguchi.bsky.social), we assembled a lot more complete, similar (within species-level) genomes than previous methods.

So much to explore... for example, we compared 6 circular TM7 bacteria of > 93% ANI assembled from a single oral metagenome.

10 / N

For this gut sample, @mgmarin.bsky.social found two distinct ermF (erythromycin resistance) genes, with 98% similarity, spreading within Bacteroidota.

1. The distinct ermFs are spreading on two distinct MGEs.
2. There is even strain specificity, only 1/6 P. copri had it!

9 / N

With circular contigs, we can confidently analyze presence / absence of "stuff" within contigs _without worrying about binning issues_ (as much).

For example, mobile genetic elements, AMR genes that are hard to bin and assemble with short reads...?

8/N

Does myloasm offer better insights, not just good benchmarks?

It turns out myloasm can recover more near-complete contigs than other ONT methods.

For a gut sample, it could assemble _6 different Prevotella copri genomes_ into single contigs, whereas other methods struggled.

7 / N

My favourite result:

For jointly-sequenced gut samples (thanks to public data from @jjminich.bsky.social), ONT can assemble _more_ circular contigs than HiFi.

This is thanks to ~3-5x increases in circular contigs relative to previous methods.

5 / N

Myloasm seems to have nice results for metagenome-assembled genomes (MAGs).

For ONT R10.4 data, myloasm stands out, especially for retrieving circular, complete contigs.

For HiFi, metaMDBG is very competitive. hifiasm-meta is still strong for gut samples. 4 / N

We do a bunch more assembly graph cleaning using coverages (and even some _theory_!). See preprint for details.

Anyways, we show this leads to good assemblies, especially when similar genomes (within-species diversity) are present. See results for mock ONT R10.4 synthetic community:

3 / N

We designed a new string graph assembler using polymorphic within-sample k-mers.

We often use SNPs to disentangle similar genomic regions ("inexact repeats", e.g. haplotypes).

But in assembly, no reference exists, so we use a context-free k-mer representation instead ("SNPmers").

2 / N

Limitations of myloasm are that it takes slightly more memory and, like other assemblers, can occasionally produce errors.

We try to be upfront about this and discuss it here myloasm-docs.github.io/qc/. We provide additional info for plotting and curation too.

It seems we can assemble (reasonably simple populations) of co-existing strains with ONT data now.

We assembled 6 single-contig Prevotella copri genomes of > 97% ANI for one metagenome. 4 of them were circular.

(The largest metaFlye P. copri contig was 13.4% complete)

Preprint will come in a couple of months. For a brief algorithmic overview and preliminary results, see myloasm-docs.github.io/results/

The main strength of myloasm: it seems like it can assemble more circularized complete genomes than before, and on diverse metagenomes.

One aspect devider worked surprisingly well at: studying recombination of antimicrobial resistance (AMR) genes.

We applied devider to a bovine gut metagenome enriched for AMR... and we found lots of interesting mosaic haplotypes. See the recombination blocks in the image (found by GARD).

4/5

Bioinfo friends: devider uses a positional de Bruijn graph approach with sequence-to-graph alignments for path retrieval --- but restricted only to an alphabet of heterozygous SNPs.

Technique inspired by a great paper from Zhou et al. for diploid haplotyping (www.nature.com/articles/s41...)

3/5

In a nutshell, devider can cluster long reads into similar haplotypes (see the images).

Main points:

(1) Works best with sequence length ~= read length
(2) Does not require prior knowledge of # of haplotypes (can be many!)
(3) Extremely fast (> 20,000x coverage is ok!)
(4) Requires reference

2/5

We updated our metagenomic profiling + detection tool sylph to v0.5.1 (github.com/bluenote-157...).

Sensitivity for low-abund species is much improved for Illumina now (thanks @fplazaonate for showing issue). bioRxiv updated with a few new benchmarks too.