Closed dcgerard closed 10 months ago
@Cristianetaniguti, could you take a look at the f1 branch of {updog}
and see if this mostly resolves the issues that you discovered?
You can install that repo via:
remotes::install_github(repo = "dcgerard/updog", ref = "f1")
Hey @dcgerard,
Sorry for taking a long time to test it. But I finally did!
I also tested another idea that you gave me when we met in San Diego, about using a combination of the updog genotype probabilities with a global error of 5%. I included as an error rate in OneMap HMM: 1 - (1 - estimated genotype probability by updog)*(1-global error)
I built linkage maps for f1 populations using a subset of the data and the combinations:
My conclusions are:
I hope this answers your question. I think it is still worth using the main branch and filtering the non-informative before using updog.
Thanks for the work @Cristianetaniguti! This is very interesting (and counterintuitive, since the f1 branch should give better results when using model = "f1"
). Some questions:
model = "f1"
in both cases, right?Hey @dcgerard,
These issues are only details, overall updog performs pretty well and it usually improves low-depth data genotypes. Let me know if you have other ideas to test. I can do it easily now.
Thank you so much! I really appreciate you working on this!
Closing for now. Don't think there is anything more to do here at the moment.
{updog}
can sometimes have poor performance when all of the following occur:model = "f1"
ormodel = "s1"
,The issue is that
{updog}
can prefer to estimate a large sequencing error rate and place the parents to both have genotypes of 1.This should only affect the
model = "f1"
andmodel = "s1"
options.eaa231c869e8d2e4c1fd11dc28e53e6270c9d5c0 patches things a little bit, by placing an upper bound on the sequencing error rate.
However, full correction will need me to dive back into the code. Right now, I don't use the parental sequences when optimizing the sequencing error rate (just the offspring sequences). Adjusting the objective to include parental sequences should fix the problem.