normal-lod documentation is misleading
I am varying the value of --normal-lod in Mutect2 in order to emit more Tumour-in-Normal variants (so that they may be rescued later with DeTiN for example). I find that I get more variants with I set the value to lower numbers, for example -6 instead of the default 2.2.
The documentation on the other hand says: "Increasing the parameter may increase the sensitivity of somatic calling, but may also increase calling false positive, i.e. germline, variants."
It looks like a discrepancy to me. I would also request that the documentation clarify what the parameter --genotype-germline-sites does on top of setting --normal-lod to -infinity.
REQUIRED for all errors and issues:
a) GATK version used: 4.6.1.0
b) Exact command used:
$gatk Mutect2 \
--tmp-dir $TMPDIR \
-R $reference_fasta \
"${input_bam_opts[@]}" \
-normal $specimen_name_normal \
-O $unfiltered \
--f1r2-tar-gz $f1r2 \
--germline-resource $gnomad_genome_af \
--native-pair-hmm-threads $SLURM_NTASKS \
--interval-padding 1000 \
--intervals $roi \
--normal-lod -6.0 \
--initial-tumor-lod 2.0 \
--tumor-lod-to-emit 3.0
c) Entire program log:
I think this is irrelevant to the question but I can add upon request.
-
Hi Pranav Garg
Do you get more Somatic Calls passing filter or more variant sites that indicate a contamination?
According to Mutect documentation below increasing normal-lod increases the need for more alt supporting reads to tag a site for somatic calling. However reducing the value will tag more artifact sites as somatic therefore you will get more somatic calls in expense of germline artifacts.
https://github.com/broadinstitute/gatk/blob/master/docs/mutect/mutect.pdf
Can you elaborate more about what you are observing?
Regards.
-
The number of emitted variants (not necessarily PASS variants) increases when normal-lod is lower (more negative). The number of PASS variants didn't change. The additional variants all have at least the "normal_artifact" filter tag.
However, I was testing in a relatively small ROI. There were a few instances where there was change to the number of PASS variants, but this was due to clustered_events.
According to your paraphrasing of the pdf documentation, reducing the nlod should increase the number of variants emitted because artifact sites will be considered somatic. This agrees with my observations but contradicts the online documentation that I quoted in my question.
-
Ah I see the issue now. Let me consult this with the main Mutect2 developer and see what he says.
-
-
Hi Pranav Garg
So the final verdict on this one from the developer is that wording is made wrong on the documentation therefore what you see is correct. We will fix the definition accordingly in the next release.
Thank you for pointing it out.
Regards.
-
Thanks. If you update the documentation, I would also appreciate if you could explain what the parameter --genotype-germline-sites does on top of setting --normal-lod to -infinity, because in the cases I tried, genotype-germline-sites always includes variants emitted with normal-lod set to a large negative value.
In the end, for my TiN problem I decided to simply set genotype-germline-sites to true instead of adjusting normal-lod. I am trying filters/rescues downstream, on variants marked as normal_artifact and/or germline.
-
Hi again.
The below 2 parameters lets Mutect2 to emit these sites even though they will be filtered out in the final FilterMutectCalls step. Consider a germline variant found heterozygous in the matched normal or a PoN site that has allele frequency of 0.5 in the population. These sites will ultimately be filtered by Mutect2 because of the evidence from the normal and PoN. Once evidence indicates filtration for those sites Mutect2 does not emit them in the VCF file. If these parameters are enabled Mutect2 will emit those sites.
--genotype-germline-sites <Boolean>
Call all apparent germline site even though they will ultimately be filtered. Default
value: false. Possible values: {true, false}
--genotype-pon-sites <Boolean>Call sites in the PoN even though they will ultimately be filtered. Default value: false.
Possible values: {true, false}Regards.
Please sign in to leave a comment.
7 comments