Mutect2 AF calculation discrepancies


I called some targeted DNA-seq using GATK 4.0 mutect2 :

java -jar $gatk_jar Mutect2 \
     -R $ref_fasta \
     -I $bam \
     -tumor $sample \
     -L $genomic_interval_bed \
     --dbsnp $dbSNP_vcf \
     -O $sample"_mutect2.vcf.gz"

Works well but I've a question regarding the generated vcf. Looking at it I found some variants with calculated AF that do not reflect the number of alternate reads vs reference reads.

Example :

chr10 43597975 . C T . . DP=395;ECNT=1;POP_AF=1.000e-03;P_GERMLINE=-3.512e-04;TLOD=6.21 GT:AD:AF:F1R2:F2R1:MBQ:MFRL:MMQ:MPOS:SA_MAP_AF:SA_POST_PROB 0/1:313,6:0.112:125,6:188,0:30:80,90:60:15:0.010,0.020,0.019:0.011,1.407e-03,0.988

If I understood this line, I've 395 reads in total. 313 reference and 6 alternate looking at the AD tag. AF=0.112

  • First question : 313+66 != 395. Did I misinterpret the DP and AD tag ?

  • Second question : 6/395 != 0.112 . How is the AF tag computed ?

    Did I miss something here ?

Thank you


  • NicoBxlNicoBxl Member
    edited March 2018

    I dig a little bit and plotted the Mutect2 computed AF versus a classic Alt/ (Alt+Ref). Mutect2 AF is always higher than the standard Alt/(Alt+Ref) for low AF variants (see boxplot) . So which AF to use for further filtering as some variants can have Mutect2 AF ~= 0.1 and Alt/(Alt+Ref) AF ~= 0.01 ..

    Thank you for the clarifications.

  • SheilaSheila Broad InstituteMember, Broadie, Moderator admin


    1) Have a look at this article.

    2) Have a look at this thread, specifically my response at the end.


Sign In or Register to comment.