Scaffolding best practices when chromosome number is known?

[nikostr]

Hi! I'm running this workflow with HiFi and Hi-C data. For both output haplotypes I see 10 large scaffolds, while my expectation is that I would have 9 chromosomes. Inspecting the cooler contact map file in R I see about 100k pixels each between scaffold_10 and the remaining large scaffolds, apart from scaffold_9 which has 300k pixels. To me, this seems to hint at these belonging to the same chromosome. Do you have any setting recommendations for YaHS to get a chromosome level assembly? Would increasing the number of YaHS rounds be helpful? Any other advice?

[prototaxites]

It's hard to give exact advice without being able to see your contact map (the PretextSnapshot PNG is a useful visualisation!), but in general all of our genomes go through manual curation by a team of dedicated curators. This typically involves manipulating the Pretext contact map in PretextView - (joining scaffolds, correcting mis-joins, etc.) (https://github.com/sanger-tol/PretextView), emitting an AGP, and applying this to the scaffolds to produce a chromosome-level assembly. We have documentation available for this process: https://curation.tol.sanger.ac.uk/

Otherwise, I'm not sure there's a best approach for tuning YaHS specifically - in my experience there are always scaffolding errors that will need to be manually corrected. But you could certainly play around with some of the YaHS parameters - sometimes increasing the number of rounds, adding more small resolutions, or even running manually running YaHS on the output of YaHS (not yet supported in the pipeline) can improve the scaffolding results.

[nikostr]

Thank you! Good information regarding manual curation being expected. Would hifiasm phasing be something that can be helpful in fixing this?

Also, do let me know if this type of question belongs somewhere else than in your github issues.

[prototaxites]

Would hifiasm phasing be something that can be helpful in fixing this?

Sometimes this can improve the quality of output scaffolds, but not always. In general it makes manual curation easier though!

Also, do let me know if this type of question belongs somewhere else than in your github issues.

You're more than welcome to ask these types of question here. There's also an Assembly/Curation Slack that I could invite you to where people ask such questions, if that would be of interest - if so, send me an email at jd42@sanger.ac.uk and I can forward you an invite.

[nikostr]

Are there any plans to integrate https://github.com/sanger-tol/curationpretext outputs into this workflow, e.g. adding gaps to the pretext file? Or is there a recommended way of running the curationpretext workflow on the outputs of this workflow?

[prototaxites]

Not at the moment - the contact maps are mostly created for visual inspection of the scaffolding results. We always run the scaffolded assemblies through TreeVal (https://github.com/sanger-tol/treeval/) to generate the full contact maps with embedded data for manual curation.