Mutect2 - using a primary tumour sample instead of a normal sample
Hello,
I am interested in a matched analysis not necessarily between normal samples vs. tumour samples - but rather primary tumour samples vs relapse tumour samples.
The following FAQ (point 6), suggests that this is possible with mutect2:
https://gatk.broadinstitute.org/hc/en-us/articles/360050722212-FAQ-for-Mutect2
As well as variants exclusive to the relapse sample, or exclusive to the primary tumour sample - I am also interested in variants which are shared between the two tumour types.
My understanding of mutect2 is that these types of shared variants will be labelled as 'normal_artifact' in the FILTER column: I think this is because this filter is applied to determine whether any (test/somatic) variants found in the reference sample are non-germline or not - and this is done by applying mutect2's somatic likelihood model to these variants. In the canonical use of Mutect2 a non-germline variant in the normal is assumed to be an artifact, hence the 'normal_artifact' label which is assigned.
My first question is whether it is valid for my use case to assume that variants labelled 'normal_aritfact' are actually somatic variants shared between the primary and relapse tumour samples?
Secondly, as a Bayesian model is used for assessing the existence of the variant in the reference, which uses the number of artifacts in the test/non-reference/somatic sample as the prior - wouldn't this then be problematic for this type of analysis. Would a relaxing of the '--normal-lod' parameter therefore be necessary to overcome the perhaps inappropriate prior for this use case? What would be an appropriate --normal-lod threshold to offset the prior in this case?
Apologies if I have misunderstood anything here. And I understand that the tool wasn't designed for this use case.
Thanks,
Thomas
-
Hi Thomas Bradley,
Your first question: Depending on the purity of your primary tumor sample and the allele fractions that you get, you might see the germline filter in addition to the normal_artifact filter.
Your second question: We wouldn't recommend messing around with the normal-lod parameter.
Hope this helps!
If other GATK users have tried this use case, feel free to chime in with what worked for you!
Genevieve
-
Hi Genevieve Brandt (she/her),
Thank you very much for your response.
Ah yes, I forgot about this - some of the variants in the primary tumour will also be likely classed as germline.
I agree, I don't want to touch the normal-lod parameter unless I am certain I know what the effects of that will be.
As I am still unsure, I am postponing this proposed type of analysis for now (i.e. using mutect2 to look at the same cancer pre- and post-treatment), but I am also interested to hear if others have attempted this type of analysis with mutect2 and what their findings are
Thank you,
Thomas -
Hi Thomas Bradley, I am doing a similar kind of analysis as what you described (pseudonormal = tumor state 1, tumor = tumor state 2). Did you manage to run successfully the analysis you were describing?
I find most of my variants (~84%) are categorized with "normal_artifact", I ran with default parameters, but I am also interested in seeing where the proportion of alternative allele varies significantly beween pseudonormal and tumor.
Please sign in to leave a comment.
3 comments