Hello everyone! I have been using the long-read assembly and binning pipeline for my analyses, and I really appreciate how streamlined and user-friendly the workflow is. While going through recent long-read metagenomic assembly studies, I noticed that many workflows include a polishing step, particularly hybrid polishing using short reads, to improve assembly accuracy and reduce sequencing errors. At present, the pipeline does not seem to include an integrated polishing module, so I was wondering whether there are any plans to incorporate this feature into future versions of the pipeline. I feel this could significantly improve overall assembly quality and make the workflow more streamlined by enabling smoother downstream processing within a single framework. Based on previous suggestions, I assembled the reads using the pipeline, performed polishing externally, and then reran the pipeline by providing the polished assembly as input. While this approach works, it becomes a bit cumbersome for larger analyses. Therefore, I wanted to ask whether there are any recommended strategies or best practices for integrating polishing more efficiently with the current workflow. If there are any plans to develop or integrate such a feature in future updates, I would be happy to cooperate or help at any stage from a user perspective. While I may not have extensive experience in pipeline development itself, I would still be glad to contribute in any way possible. Thank you again for maintaining such a useful resource!

That was discussed recently: https://nfcore.slack.com/archives/CE9MS66BS/p1773852968755209

Thank you for sharing the discussion and paper. That makes a lot of sense, especially regarding the improvements in newer Nanopore chemistries. My main motivation was actually more focused on hybrid polishing workflows using Illumina reads alongside long-read assemblies. Similar to your observation with polypolish and pypolca, I have also seen cases where some assembly quality metrics appeared to improve after short-read polishing, particularly for downstream analyses that may be sensitive to residual sequencing errors. I completely understand that this may not justify making polishing a default step in the pipeline. I was mostly wondering whether an optional module or a more streamlined integration strategy could be useful for users working with hybrid datasets at larger scales. In any case, thank you again for the explanation and for sharing your experience — it was very helpful.

Yes, I think that would be beneficial for hybrid datasets. I would vote for having polypolish & pypolca in the pipeline for long read assemblies. Both seem available as nf-core modules: https://nf-co.re/modules/polypolish_polish/ & https://nf-co.re/modules/pypolca_run/. There are some questions that will come up, for example I first mapped short reads to the bin of interest with strict settings, extracted those and mapped forward and reverse with settings recommended for Polypolish, but I am not sure whether polishing the metagenome itself before binning or the individual bins is better. So the general strategy might be debatable.

Yes, after reviewing the previous discussion on polishing and several recent papers related to long-read assembly and binning, I understand that if polishing is performed, short-read polishing is probably the more suitable option. At the same time, the recent Oxford Nanopore paper you shared suggests that with newer chemistries, polishing may not significantly improve assembly quality compared to earlier ONT datasets. However, many recent studies are still incorporating short-read polishing in their workflows. Earlier, I also performed short-read polishing using NextPolish, following methodologies similar to this paper: https://www.nature.com/articles/s41564-022-01270-1#Sec11 I did observe improvements in assembly quality metrics after polishing, although the improvements were not very large. So, currently, I feel the situation is somewhat debatable. While modern ONT data may reduce the necessity for polishing, many workflows still use conservative short-read polishing to improve residual base-level errors and downstream annotation quality. I also tried to find literature specifically discussing possible loss of strain variation after polishing, but so far, I have mostly come across studies reporting beneficial impacts of polishing. We could add an optional polishing module that could be considered for users who want to benchmark or apply polishing in their own datasets, while still allowing others to skip it depending on sequencing chemistry, coverage, and the biological question being addressed.

We could add an optional polishing module that could be considered for users who want to benchmark or apply polishing in their own datasets, while still allowing others to skip it depending on sequencing chemistry, coverage, and the biological question being addressed.

Sounds good to me.

Yes, I would be happy to cooperate if an optional polishing module is being considered. I may not have much experience in pipeline development itself, so I probably would not be able to implement it independently, but I would still be glad to help in any way possible.

Not a lot of experience is needed. Fork the repo in github, make a branch based on dev in your fork that you modify, open PR to dev again. All also documented in https://nf-co.re/docs/contributing/overview and especially for that task https://nf-co.re/docs/contributing/contribute-existing-pipelines. 😉

@Daniel Straub or @HARSHITA SRIVASTAVA, can you open a new issue to keep track of this? I also think this is valuable.

okay, @Daniel Straub and @Diego Alvarez Saravia