GATK 4 Mutect2 gives too much low depth variants

Question

[Zhizhong]

As gatk3 mutect2 has problem with AF calculation, GATK Forum recommends move to gatk4. however after filter the "PASS" filter, thousands of snvs or indels left for one wes tumor sample. most of the newly called variants are with very low alt depth(from 1 to 5). I also compared the result with vardict and strelka2. It is very weird to see this huge difference. Since I don't know how to ask the gatk forum or comment on a discussion, I post it here. May somebody where the problem is!

[WouterDeCoster]

Well, isn't that what you would expect, that you find variants in a tumor with a low to very low frequency? If that alt depth is too low for you, I'd suggest you just filter the obtained vcf for a minimal of supporting reads.

[Zhizhong]

Yes, I would like to set minimal alt depth to 5, it is very hard to tell variants with very low frequency really exists. The new version is too sensitive, I'm not sure about the false positive ratio.

[WouterDeCoster]

That's a common problem. When increasing the sensitivity, the probability of including false positives raises. I'd suggest to just filter the obtained vcf.

[Kevin Blighe]

Maybe these are down-sampled 'high quality' reads? The actual position read depth may be higher. I note that GATK's tools don't always report the true read-depth.

Answer 1

[Charles Warden]

If you calculate the allele frequencies separately (such as in a .pileup file, or an allele counter that was probably meant more for RNA-Seq but could be used for DNA-Seq like in GATK or allelecounter or phASER, etc), you can potentially use that is a second measure to filter variants. With the exception of some potential indel formatting differences, you could also try looking for overlap with VarScan somatic variants.

You might also want to consider making joint germline variant calls, rather than with a formal somatic variant caller.