Improvements to Mutect2's Permutect training data mode #8663
Conversation
…nt M3 dataset modes -- Illumina and Ultima
| public Mutect3DatasetMode mutect3DatasetMode = Mutect3DatasetMode.ILLUMINA; | ||
|
|
||
| public enum Mutect3DatasetMode { | ||
| ILLUMINA(11), |
There was a problem hiding this comment.
I get a little nervous when you hardcode the sequencing technology, especially when given the same value. Why not just make this a default? What is the justification for this? (We do not need to block this PR over this)
There was a problem hiding this comment.
Since Ultima uses unpaired reads it can't be encoded in the same way as Illumina data. It's only a coincidence that the dimension of read tensors is 11 in both cases.
There was a problem hiding this comment.
(And note that this is an Advanced argument)
| final Allele longestRef = allRefAlleles.stream().sorted(Comparator.comparingInt(Allele::length).reversed()).findFirst().get(); | ||
|
|
||
| final List<Allele> remappedAltAlleles = ReferenceConfidenceVariantContextMerger.remapAlleles(vc, longestRef).stream() | ||
| .skip(1).toList(); |
There was a problem hiding this comment.
Nit: Is skip(1) just because we need to skip the reference allele? If so, just add a comment that you are making this assumption.
|
@LeeTL1220 Back to you. Note one extra commit setting the default ratio of non-artifact training examples to artifact examples to 1 instead of 20. I checked and this has no effect on the quality of the trained model while vastly decreasing training time. |
@LeeTL1220 here are the changes I was talking about. The first commit contains a small change I have been using for a year, the second is the recent multiallelic indel representation bug fix.