performance of indel calling by MuTect2
Answered
I'm interested in MuTect2 performance on indel calling. My experience in using MuTect2 in some research projects of mine is that the indel accuracy is a bit low, and a lot of the indels are false positives. I wonder what's your thoughts on that? Any recommendations about how to increase the indel calling accuracy?
-
The sensitivity and false positive rate you get with Mutect2 will depend on a lot of variables.
How deep is your sequencing, are you doing WGS or a panel? Are your samples FFPE? Are you using a tumor/normal pair, what is the allele fraction of your variants, etc.
Indels are more difficult than SNVs, so it is not unusual to have poorer performance for indels.
-
Our coverage for tumor normal is around 200X / 120X on WES data, using FFPE tumor with matching blood. AFs of false positive indels vary with in the range of 0.02 and 0.4 I believe. I would say over 50% of the indels which I manually reviewed in IGV turned out to be false positives due to poor mapping quality or other issues seen in IGV
-
Having 50% of indels be false positives is definitely a problem. FFPE can cause problems, but typically don't cause mapping artifacts.
I would expect that mapping artifacts would be filtered well due to your normals.
So I'm honestly at a bit of a loss as to why you are having so many false positives. Perhaps I could take a brief look at your data?
-
These data were generated using tumor/normal pairs but without the use of panel or normal. Perhaps that could be a cause? I'm happy to share one T/N pair with you. What would you recommend the best way of data sharing?
-
If your data is on Google Cloud, that is probably easiest. Just give me permissions to access, and let me know the gs:// URL.
Otherwise, dropbox, and other sharing services work.
Please sign in to leave a comment.
5 comments