Dear GATK Team,
We have noticed that Mutect2 ( 188.8.131.52) is missing some of the key (TP53 and KRAS) mutations from tumor-normal paired cfDNA somatic mutation calling. We have an analysis from another source where we know that these mutations are being picked up. On investigation, our raw data (aligned bams) shows that the reads are present but the bamout shows that these regions were not selected as active regions. Hence, mutations are not coming up.
We have a handful of examples as follows:
KRAS with 3% alternate bases: 40/1182 Alternate allele,
TP53 with 4% alternate bases: 21/556
TP53 with 3% alternate bases: 45/1310
We do have cases when similar mutations have been picked up with similar or even lower alternate counts. See the positive cases where these regions appear in the active regions at a very low frequency.
KRAS at 4%- 52/1166
ATM at 3% : 23/849
Both these types;
a) active regions not containing variants and,
b) active regions containing variants
have been run from the same pipeline. The mutect2 command is as follows:
--pairHMM AVX_LOGLESS_CACHING_OMP \
I have tried tunning a few parameters including
--max-reads-per-alignment-start 0 \
--f1r2-median-mq 5 \
--minimum-mapping-quality 10 \
--min-base-quality-score 10 \
--active-probability-threshold 0.001 (or 0.003)\
--initial-tumor-lod 1.0 \
--dont-trim-active-regions true \
But none of these helped to pick up these mutations. However when I used the force active mode then I picked up these mutations on their respective frequency.
KRAS - coming in active region now
TP53 - coming in active region now
Though this helps to solve the problem, this does not seem an appropriate solution because of
a). Some cases are being picked up at similar frequency while others are not
b). Computational time with this mode on is 10-12 times more compared to when you rely on mutect2 intrinsic behavior to choose active regions based on variations.
Based on these examples, can you please suggest what could fix this problem? I am happy to share bam files for investigations and testing if that's needed.
Looking forward to any solution to this problem.
Please sign in to leave a comment.