search

statgen / demuxlet

Genetic multiplexing of barcoded single cell RNA-seq
Apache License 2.0
116 stars 25 forks source link

Demuxlet Output Questions #82

Open nh13 opened 3 years ago

nh13 commented 3 years ago

I am trying to understand the output of demuxlet, in particular the .best file.

  1. LLK1 is for DBL.1ST while LLK2 is for DBL.2ND, but for LLK12, what does the ID mean in this case? ID of what?
  2. I noticed that for almost all samples with the singlet as the best assignment, that LLK12 is exactly equal to SNG.LLK1. The code to pick the "best" assignment will pick singlet in this case, due to the pairLLK12 > singLLK1 + 2. But if the likelihoods are equal (or close), shouldn't the output be ambiguous (modulo some prior)? I don't see a way to control this case, as I'd like to tune my output to only have unambiguous assignments that have a minimum confidence.
  3. Is there an easy way of getting the posterior probability of a singlet like we have for PRB.DBL?
nh13 commented 3 years ago

I am using --alpha 0.0 --alpha 0.5 --alpha 1.0 for two sample mixture.

nh13 commented 3 years ago

Perhaps the --alpha 1.0 is the issue, as it's allowing a "doublet" with 100% one of the samples?