Can germline callers detect somatic variants? (RNAseq, cancer, no matched normal)

Question

[don]

Hi folks,

I have RNAseq data with tumor samples without matched normal samples. Unfortunately, I don't have any DNA or RNA samples, or DNA-seq data. My purpose is to check some cancer-specific mutations, very well-known mutations. Those mutations cannot be germline mutations. Actually, no matter if they are germline or somatic, I just want to know if those samples have the mutations. I just want to make sure that the mutations are not filtered out.

Question is,

1. Is it generally OK if I can understand the output of somatic callers is a subset of the output of germline callers, in case of tumor-only sample.
2. If so, can I use germline variant callers not to lose my mutation? When I look around, DeepVariant-RNAseq seems to be the best practice for RNAseq germline variant calling.
3. I don't know if the germline callers filter-out variants based on AF with high dispersion from 0.5 or 1.

My study is not for clinical purpose, and all the findings will be validated with wet experiments with cell lines. but I want to make sure how confident my data is for publication.

Thanks

[i.sudbery]

In general the output of somatic callers is not a subset of the output from germinline callers. Germline callers try to call variants that are present at frequencies that are compatible with them being hetrozygous (i.e. carried by approximately 50% of reads) or homozygous (carried by nearly all reads). Somatic variant callers try to call variants whose frequencies are NOT compatible with being hetrozygous or homozygous.

[Hyper_Odin]

did you check out this?

https://gatk.broadinstitute.org/hc/en-us/articles/360035531192-RNAseq-short-variant-discovery-SNPs-Indels

i believe this is what you are looking for.

Answer 1

[Istvan Albert]

I would recommend using a somatic mutation variant caller like Mutect2

https://gatk.broadinstitute.org/hc/en-us/articles/360037593851-Mutect2

Another way to do this, if you knew the mutations and all you wanted to check for their presence, would be to parse out the pileups at those position - though that process probably works well only for SNVs

[i.sudbery]

We've had a bit of a go with Mutect2, but without a germline control, we've struggle to seperate somatice from what looks very much like germline mutations.

[Istvan Albert]

If I read the OP's question, they don't mind if they cannot separate the two mutations - they just don't want to miss out on a variant because it is somatic hence far from looking like a valid heterozygous. Their concern is that the they variant caller will tacitly ignore some variants because the balance is off from 50%.

In general even a tool like bcftools when run with ploidy=1 will produce heterozygous calls, with high imbalance. So the variants will be present, only that the genotype likelhoods are lower. It is more of a tuning and parameter issue.

[i.sudbery]

The problem with Mutect we had was not so much that it was calling germline as somatic, but that it didn't really appear to be calling any somatic - once we filtered out obvious homozytous Alt, and things bang on 50% VAF, there was pretty much nothing left. From that point of view, we've actually had better luck with bcftools.

But I think you are actaully correct - the OP doesn't really look to be looking for things that would look like somatic mutaitons in the data - one would expect a driving mutation to be present in most/all reads, and therefore look like a germinline mutation to the caller.