ww-poreCoV extension

[hoelzer]

I suggest using the poreCov pipeline as the backend for SARS-CoV-2 wastewater lineage deconvolution from nanopore long reads. You already added freyja ( #274 #270), which is great as the current community standard.

However, we are also interested in detecting new stuff, aka "cryptic lineages" or novel mutation profiles.

To do this, I would like to test/implement two recent approaches:

1. CONCOMPRA

• https://github.com/willem-stock/CONCOMPRA
• consensus approach for community profiling with nanopore amplicon sequencing data, focused on 16S rDNA
• I already tested this on example nanopore data from mixed patient samples (simulating wastewater) and it looked very promising
• should be easy to add as a single new process
• challenge is that the tool only works on one primer pair (Usage of more than one primer pair willem-stock/CONCOMPRA#1)

2. Floria

• https://github.com/bluenote-1577/floria | https://doi.org/10.1093/bioinformatics/btae252
• Strain-level haplotyping for metagenomes with short or long reads.
• I am curious how well this works on a) amplicon data and b) SARS-CoV-2
• input is VCF and mapped reads (BAM): so I think should be also easy to add

By this, we would get known lineage abundances from freyja plus potential new lineages from one or both of the other tools.

Finally, we could also write a little ww-poreCoV extension paper ;)

[hoelzer]

Another interesting method might be VirPool: https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-022-05100-3

They illustrated already the advantage that longer amplicons bring w/ their method in particular.

is able to use the entire length of reads instead of just the most informative positions, and can also capture haplotype dependencies within a single read. A crucial property of our model is its ability to capture long-range dependencies within reads, which is particularly relevant when coupled with use of long amplicons and nanopore sequencing.

They also tested the method on real data by sequencing a mixture of eight clinical samples using long amplicons (2kb).

(sidenote, attention was brought up and information was shared by Victor! thx!)

The question is, as usually, how up-to-date is the tool?

https://github.com/fmfi-compbio/virpool

They also provide scripts to create own profiles: https://github.com/fmfi-compbio/virpool?tab=readme-ov-file#creating-a-custom-variant-profile

Maybe covsonar could also create these profiles easily. Or Ashkans sc2mfc tool.

[replikation]

Yep, it's completely fine to extend Porecov now on the wastewater surveillance. We just need to make sure the normal "user experience" is not convoluted.

[MarieLataretu]

ad Floria:

You would use the VCF + BAM form ARTIC, right? Just to keep in mind: mixed indels might be tricky.
I saw an overlapping x nt deletion and y nt deletion, and neither of both was called by medaka, and/or a frameshift introducing indel was called

edit: corrected tool name; Florida would be also a fun name

[hoelzer]

ad Florida:

You would use the VCF + BAM form ARTIC, right? Just to keep in mind: mixed indels might be tricky. I saw an overlapping x nt deletion and y nt deletion, and neither of both was called by medaka, and/or a frameshift introducing indel was called

Yes, I would like to use the output porecov anyway produces to change as little as possible.

However, good point. I would live with such issues for now. Deconvoluting lineages from wastewater is anyway wild west :) but of course, important to keep such situations in mind.

Ps: Floria ;)

[replikation]

Not sure if it is possible to implement a subcommand or a second "main.nf" solely for the wastewater part? (thinking samtools subcommand for instance)

[MarieLataretu]

I think subcommands are rather unusual - afaik, this is mainly handled by a parameter.
E.g. viralrecon has a platform parameter for either Illumina, or nanopore:
https://github.com/nf-core/viralrecon/blob/3731dd3a32a67a2648ea22c2bd980c224abdaee2/main.nf#L62-L76

Also, it could make trouble with the execution from GitHub nextflow run replikation/poreCov ... (or it needs some extra configuration)

[hoelzer]

Started working on this in a branch ww-porecov. First step was a container for CONCOMPRA which was already a bit pain, but made it finally

rkimf1/concompra:v0.0.1--f6c273d

[hoelzer]

The authors of CONCOMPRA provide a docker now: willemstock/concompra:version0.0.2

However, we need to see if that works w/ nextflow

[hoelzer]

This might be even more interesting to add as a process to ww-poreCov:

• https://www.medrxiv.org/content/10.1101/2024.08.27.24312690v1
• Unsupervised detection of SARS-CoV-2 mutations and lineages in Norwegian wastewater samples using long-read sequencing
• https://github.com/garcia-nacho/HERCULES

One big questions is probably: how do they handle the reference used for lineage assignments at the end? And: can they detect something new/cryptic

[hoelzer]

Looking into HERCULES

• they have a pre-build docker container but we might want our own to work w/ nextflow (its maybe a bit cumbersome to use otherwise)

• they provide option for own reference, based on aligned spike genes:

To update the set of references that HERCULES uses, you need a fasta file with all the Spike-genes of the reference sequences aligned. The names of the sequences inside the fasta file must have the following structure IDXX_LineageY where IDXX is the unique identifier for the sequence and the LineageY is the lineage assigned to the sequence, it can be pangolin lineage or any other nomenclature system (e.g. nextclade clade ID, WHO nomenclature, etc). The fasta file must be stored in a folder that must be mounted inside HERCULES with the following flag. -v Path/to/Folder:/Reference

This might be a good option to build an up-to-date reference based on "representative spike sequences" per lineage (https://github.com/rki-mf1/sc2-mutation-frequency-calculator)

[hoelzer]

re-tagged the provided image to mhoelzer/hercules:0.0.0 bc the provided name (ghcr.io/garcia-nacho/hercules) crashes some process in poreCov

[hoelzer]

Unfortunately, the container is a pain to use in Nextflow. Here is the Dockerfile and how then the tool is executed (a bash script):

https://github.com/garcia-nacho/HERCULES/blob/master/Dockerfile#L76C19-L76C57

Probably, we need first to re-build the container. I tried to just docker build the original Dockerfile but this fails early (on my MacBook).

[hoelzer]

garcia-nacho/HERCULES#2 (comment)

What I would recommend you is to run HERCULES several times, one per amplicon, and to compare/integrate the results. The larger the amplicons the better (i.e., Midnight and ViralAmp protocols)

Hm okay.. similar issue like with CONCOMPRA

[hoelzer]

Is this maybe interesting for our purpose of detecting smt new?

https://github.com/zhuangx15/ICAvar
https://www.medrxiv.org/content/10.1101/2024.04.18.24306052v1.full

[hoelzer]

Another idea:

devider; https://www.biorxiv.org/content/10.1101/2024.11.05.621838v1

[hoelzer]

Another package from colleagues in France (Hughes has contact) where they reimplemented ww tools and making them faster

Also

... indeed the haplotype callers should do the task that you need. If you want to make full use of the long reads, you will need to start from the BAM and count the mutation cooccurrences. The tool from El Hacene DJAOUT does that: https://pypi.org/project/VaRaPS/1.0.0

[hoelzer]

Another interesting method might be VirPool: https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-022-05100-3

They illustrated already the advantage that longer amplicons bring w/ their method in particular.

is able to use the entire length of reads instead of just the most informative positions, and can also capture haplotype dependencies within a single read. A crucial property of our model is its ability to capture long-range dependencies within reads, which is particularly relevant when coupled with use of long amplicons and nanopore sequencing.

They also tested the method on real data by sequencing a mixture of eight clinical samples using long amplicons (2kb).

(sidenote, attention was brought up and information was shared by Victor! thx!)

The question is, as usually, how up-to-date is the tool?

https://github.com/fmfi-compbio/virpool

They also provide scripts to create own profiles: https://github.com/fmfi-compbio/virpool?tab=readme-ov-file#creating-a-custom-variant-profile

Maybe covsonar could also create these profiles easily. Or Ashkans sc2mfc tool.

Virpool might be promising...

you have the choice between : Virpool (cooccurrence), Frejya (frequencies), and LCS (counts)

And the tool in the post before has a faster reimplementation of virpool

[hoelzer]

Thomas is testing: https://github.com/bluenote-1577/devider

[hoelzer]

That looks also promising, VILOCA:

https://academic.oup.com/nargab/article/6/4/lqae152/7912062