normal sample AF in Mutect2 paired mode

Follow

Fran

• May 26, 2023 13:13

Hi all,
I have a question regarding the variants' AF value reported in the normal sample when using Mutect2 in paired mode.

I have noticed that AF in the normal sample is never 0, even in those cases where AD says that there are 0 reads with ALT.

For example, for this deletion (first INFO column is for normal sample and second tumoral sample):

chr15    65150921    .    GAAAAGCCATGTTTGTTTTTTTAAAATAAAAAATGGAGGTAAACATGATCTTTAAAATCA    G    .    PASS    AS_FilterStatus=SITE;AS_SB_TABLE=24,75|5,36;DP=162;ECNT=1;GERMQ=66;MBQ=38,40;MFRL=182,215;MMQ=60,60;MPOS=35;NALOD=1.5;NLOD=8.39;POPAF=6;TLOD=160.96    GT:AD:AF:DP:F1R2:F2R1:FAD:SB    0/0:41,0:0.032:41:1,0:15,0:31,5:10,31,0,0    0/1:58,41:0.459:99:11,9:19,12:45,38:14,44,5,36

In the normal sample:

AD = 41,0

AF = 0.032

What I see in igv: 0

Another example for a SNV:

chr1    13391966    .    A    G    .    PASS    AS_FilterStatus=SITE;AS_SB_TABLE=21,13|7,6;DP=49;ECNT=1;GERMQ=66;MBQ=20,40;MFRL=147,256;MMQ=60,40;MPOS=19;NALOD=-1.687;NLOD=3.03;POPAF=6;TLOD=52.12GT:AD:AF:DP:F1R2:F2R1:FAD:SB    0/0:34,1:0.074:35:3,0:21,1:24,1:21,13,0,1    0/1:0,12:0.93:12:0,5:0,6:0,12:0,0,7,5

In the normal sample:

AD = 34,1

AF = 0.074

What I see in igv: 0.28

How is AF calculated? Is it calculated from the reallignment Mutect2 does?

Should I recalculate it form AD? Because I would expect AF to be 0 in the normal sample for somatic mutations.

Thanks in advance.

Francisco

REQUIRED for all errors and issues:
a) GATK version used: 4.3.0.0
b) Exact command used: 

gatk --java-options "-Xmx36g" Mutect2 \
        --input WES_FD_N_1-1.converted.cram --input WES_FD_T_1-1.converted.cram \
        --output WES_FD_T_1_vs_WES_FD_N_1.mutect2.vcf.gz \
        --reference GRCh38.d1.vd1.fa \
         \
         \
        --intervals chr1_12081-12251.bed \
        --tmp-dir . \
        --f1r2-tar-gz WES_FD_T_1_vs_WES_FD_N_1.mutect2.f1r2.tar.gz --normal-sample WES_FD_N_1 --genotype-germline-sites

1

• 

  David Benjamin

  • June 05, 2023 17:09

  The AFs emitted by Mutect2 are the means of the posterior distribution on cell fraction conditioned on a variant existing.  The AF of 0.032 you see, for example, means "if this variant is really present in the normal (not that Mutect2 thinks it is), then our best guess of its allele fraction in the normal sample's cells is 0.032".

  0

  Comment actions Permalink
• 

  Dmitrij Kazancev

  • August 22, 2024 15:22
  • Edited

  Hi, I'm glad I found this post because I had the same question regarding non-zero AF values.

  What would be the best method to filter out variants that have zero ALT reads? I'm thinking about counting ALT variants in reads and if there are zero assign "real" AF = 0. Maybe there are better methods?

  I'm using Mutect2 (GATK 4.5.0.0) to call somatic variants in whole exome tumor-normal pairs.

  The command in my scripts looks like this:

  docker run --rm -v $WDIR:$WDIR -v $REFDIR:$REFDIR -w $WDIR/bam \
      broadinstitute/gatk:$VERSION \
      gatk \
      Mutect2 \
      --native-pair-hmm-threads $THREADS \
      -R $REFG \
      --germline-resource "$REFDIR/gatk-BP-somatic-hg38/af-only-gnomad.hg38.vcf.gz" \
      -I $BAMFILE_NORMAL \
      -normal $SAMPLE_NORMAL \
      -I $BAMFILE_TUMOR \
      -O $WDIR/vcf/$ID.unfiltered.vcf

  Thanks for developing these tools!

  Dmitrij

  0

  Comment actions Permalink
• 

  Gökalp Çelik

  • August 23, 2024 11:15

  Hi Dmitrij Kazancev

  Germline AF calculation in Mutect2 has its own merits from population genetics and genotypic priors. But if you absolutely filter out any 0 ALT allele in normals you may use our VariantFiltration tool with to manually filter those sites.

  Regards. 

  0

  Comment actions Permalink

Please sign in to leave a comment.