Mutect2 for ultrasensitive low coverage calling
Can you please provide
a) GATK version used
b) Exact GATK commands used
c) The entire error log if applicable.
I am trying to perform mutation calling on 50 6-8X whole genomes corresponding to subclones derived from a parental cell line with a 30X WGS. I am interested in identifying ALL variants across the cohort of lower coverage whole genomes that are not present in the parental line (including cases with alt count of 1 or where there is low depth in my normal).
What would be the best set of settings to use for this analysis? I could imaging performing joint calling with all 50 clones using the same matched normal with the following settings:
and then removing variants that are also in the normal, but suspect there is a more efficient way to implement the caller. Any further suggestions welcome!
We have this documentation, troubleshooting how to make Mutect2 and HaplotypeCaller more sensitive. For low coverage, we recommend to run the analysis with
There are many ways you can run Mutect2, examples are in the documentation. Please post your entire command and GATK version number so that others can see what you have tried, and what they might recommend you change.
The GATK support team is focused on resolving questions about errors and abnormal results, so for questions regarding best parameters like this, we ask the community to chime in with what has worked for them.
Please continue to post your questions because we will be mining them for improvements to documentation, resources, and tools.
For context, check out our support policy.
I would try running Mutect2 in multi-sample tumor-normal mode if I were you:
gatk Mutect2 -R ref.fasta -I tumor1.bam -I tumor2.bam . . . -I tumor50.bam \
-I normal.bam -normal normal_sample_name_in_bam_header \
-germline-resource af-only-gnomad.vcf.gz \
-pon pon.vcf \
gatk FilterMutectCalls -R ref.fasta -V unfiltered.vcf -O filtered.vcf
The knobs to adjust to increase sensitivity in Mutect2 (at the expense of runtime) are -pruning-lod-threshold and -tumor-lod-to-emit. At very low values these may actually harm sensitivity as the assembly graph will get clogged with errors.
To increase sensitivity in FilterMutectCalls at the expense of precision you may adjust the -f-score-beta argument.
-callable-depth, by the way, sets the threshold for counting callable sites in the .vcf.stats output in order to help FilterMutectCalls learn the prior probability of somatic variation. It is not a hard threshold for variant calling and we do not recommend adjusting it.
Please sign in to leave a comment.