KolmogorovLab / hapdup

Pipeline to convert a haploid assembly into diploid
Other
85 stars 8 forks source link

Issue with selection of PEPPER model for new ONT chemistry #35

Closed thainsCEB closed 3 months ago

thainsCEB commented 9 months ago

Hi,

I was trying out PEPPER and HAPDUP at the same time for the same assembly and the same read set. I noticed that with PEPPER I could specifically pick the ont_r10_q20 model but with hapdup, it automatically chooses the ont_r9_guppy5 model. I just got my reads off a machine in July with the newest chemistry. Does this matter? Or can it be added to add a new flag to select the ONT model? I also noticed the same thing for margin as well.

mikolmogorov commented 9 months ago

Hi,

I think it should still work ok, because the SNP performance between R9 and R10 is somewhat similar. But once I have a chance I can check if choosing R10 profile in PEPPER makes an improvement.

sivico26 commented 8 months ago

Hi @fenderglass, any updates on this?

I agree that SNP performances should be roughly the same, but I am more concerned with mis-specifying the models. If PEPPER is expecting R9 signal data and I give him R10 data, it would try to correct the systematic errors typical of R9 data (and probably omit the ones of R10 data). In short, an increase in false positives and false negatives.

mikolmogorov commented 8 months ago

Hi @sivico26

I don't think this should be a concern since PEPPER SNP calls are only used for initial phasing and haplotagging of reads, which are then locally reassembled. So even if were are additional FP/FNs, it would not matter after reassembly (unless the error rate is very high and somehow affected phasing). I would give it a try first and see how it performs.

sivico26 commented 8 months ago

Dear @fenderglass,

Noted, I will give it a try. Although if that is the case, I already have my reads haplotagged. Is there a way to give HapDup these directly and skip that part?

Alternatively, I noticed that I can give the model myself with the --pepper-model option, but I did not manage to download the pepper models in the first place. pepper download models is not working for me, and I do not find them anywhere in the repo. Do you know how to get them?

mikolmogorov commented 8 months ago

@sivico26 Hapdup is using custom alignment preprocessing prior to variant calling and phasing, so it's not possible to skip haplotagging.

Not sure why pepper download models is not working - you may want to check with the PEPPER developers. You also should be able to extract it from a Docker package.