the difference between the positions called by somatic caller (MuTect2) and the position called by germline caller (UnifiedGenotyper)?

Question

[Zoey Zou]

Hi,

I am new to tumor phylogenetics. I want to call snvs from the tumor cells and using the snvs to build the phylogenetic trees. I found that currently there are two ways to call somatic mutations from tumor samples.

One is doing the somatic calling using somatic caller (MuTect2, Varscan ..).

Another way is doing the snvs calling using the germline calling software (GATK...)and then selected the positions which are 0/0 in normal tissue and 0/1 or 1/1 in tumor tissue.

Is there a huge difference between the results of these two methods? Which method is more accurate?

Appreciate for any help.

Answer 1

[Charles Warden]

I actually think "germline" joint variant calls (from HaplotypeCaller gVCF files) may be better than using MuTect2 (also, it is "MuTect2" not "Mutech2"), in order to reduce false positives from low frequency variants.

However, I think it is good to have as an option, along with VarScan (which can also be run with more conservative variant calling settings, if you want to help filter variants).

[Zoey Zou]

Hi,

Thanks for your reply ( Sorry for my mistake, I have corrected it). Do you mean that the joint calling are more likely to remove those low frequency variants when comparing to the somatic calling?

[Charles Warden]

I believe so:

Joint variant calling should increase the number of variants called (compared to separate GATK calls).

However, it is still intended for germ-line variants, so I think it is more likely to pick up variants that may have variant frequencies like 0.3 (compared to 0.05). Also, for MuTect2, I think you may want to use a .pileup file to check the variant frequencies (sometimes, what is reported by GATK can be noticeably different from what you would see from visualizing the alignment in IGV).

If you have dozen or less samples, I think the run-time will also be shorter for joint variant calls than MuTect2 variant calls. However, I think I've only tried it a couple times with ~6 total samples.