Dear GATK4 community,
I am carrying out a somatic variant calling analysis on amplicon data (custom panel) over FFPE tumor samples. This panel interrogates described SNVs for stomach adenocarcinoma. I've used GATK4 v126.96.36.199, following Best Practices for Mutect2 in tumor-only mode.
So far, I have analyzed 51 samples and obtained a total of 10,062 raw variants which have been reduced to 18 variants after filtering process. I considered all hard and probabilistic model filters included in FilterMutectCalls with default values except for 'Contamination' which was discarded based on this previous post (https://gatk.broadinstitute.org/hc/en-us/community/posts/4403802051227-Contamination-estimation-with-similar-error-magnitude-in-amplicon-data).
The most frequent filter instance is 'Orientation' from the LearnReadOrientationModel. In fact, it appears as 'solo' or in combination with other filters in 99% non-PASS variants (9,965 out of 10,062 raw variants). Exploring the results, I have seen that in every raw variant, both corresponding ALT and REF reads had the same orientation, specifically, F1R2. No read (ALT or REF) has been assigned to F2R1.
I was expecting to see a bias between F1R2 and F2R1 read counts for ALT to flag the Orientation filter but I was surprised to find all ALT and REF reads following same orientation. So, my question is:
Can we say there is an orientation bias if no REF/ALT reads are assigned to one of the orientations?
I wonder if this scenario is analog to the one described in this post (regarding the LearnReadOrientationModel) https://gatk.broadinstitute.org/hc/en-us/community/posts/360057770812-Artefactual-calls-using-Mutect2-on-RNAseq-data and discard the Orientation filtering.
Many thanks in advance for any advice on this!
Please sign in to leave a comment.