What is the best way to generate VCF for vg-toolkit given raw sequences

[sumit-walia]

I am trying to generate variation graph (vg) from raw SARS-CoV2 sequences (~16k sequences). What would be the best way to generate VCF given these raw sequences? And, does the method works as the data scales up?

[adamnovak]

We used to support vg msga for making graphs from sequences, but it doesn't scale that high in sequence count.

You could try making a graph with Minigraph, or Minigraph and Cactus together, though I don't know if that would scale enough either. @glennhickey might be able to guess.

You could also throw @ekg's PGGB tool at the problem.

If you want to go sequences -> VCF -> graph... I don't know what tool you would use there. You might want your own tool based on individual pairwise alignments against whatever you are using for your reference.

I've never seen a VCF with genotypes for 16,000 samples in it, let alone tried to run vg on it. But the GBWT should store haplotypes in sub-linear space, so it might actually work.

If you aren't working with the genotypes but just the variable sites, vg probably ought to handle the VCF just fine as input.

[jeizenga]

If the sequences are mostly related by simple mutations (e.g. small indels, substitutions), you could also generate a multiple sequence alignment with an external tools and then use it to construct the graph with vg construct -M.

[glennhickey]

You could try making a graph with Minigraph, or Minigraph and Cactus together, though I don't know if that would scale enough either. @glennhickey might be able to guess.

minigraph-cactus will not scale to 16k sequences. If you have a guide tree, "regular" cactus should work fine but then you wouldn't get a nice vcf.