GATK licensing moves to direct-through-Broad model -- read about it on the GATK blog

Strange QUAL values with UnifiedGenotyper

mrubiocmrubioc Posts: 2Member
edited September 2013 in Ask the GATK team

I'm working with the UnifiedGenotyper walker and I have detected strange values for the QUAL field of some VCF entries in the output files.

Sometimes in the VCF output file, the QUAL value for different vcf entries it is repited, for example, the QUAL values 32729.73 or 2147483609.73 usually appear in the output and not only in my files, because when I have searched on the GATK forum, this value appears in other users posted vcf files related to other questions.

I have tested it with several GATK versions, and in the latests versions these QUAL numbers are extremely high and I have also detected that the value doesn't correspond with the relationship QUAL~QD*DP.

Another strange thing is that for other QUAL values in the VCF file, it is very common that decimal part begins with a seven. i.e : 32729.73

Have you detected this? Is some kind of bug?

I look forward to your response.

I copy some VCF entries with different GATK versions:

Last version (2.7-2):

chr13    32907535    .    C    CT    2147483609.73    .    AC=1;AF=0.500;AN=2;BaseQRankSum=-1.483;DP=1000;FS=0.000;MLEAC=1;MLEAF=0.500;MQ=37.88;MQ0=0;MQRankSum=13.620;QD=33.72;RPA=11,12;RU=T;ReadPosRankSum=-12.065;STR    GT:AD:DP:GQ:PL    0/1:1386,1037:2453:99:9301,0,13130
chr13    32907589    .    G    GT    2147483609.73    .    AC=1;AF=0.500;AN=2;BaseQRankSum=6.991;DP=999;FS=0.000;MLEAC=1;MLEAF=0.500;MQ=37.93;MQ0=0;MQRankSum=26.910;QD=28.67;RPA=7,8;RU=T;ReadPosRankSum=-2.595;STR    GT:AD:DP:GQ:PL    0/1:1306,1142:2469:99:14116,0,16944

V 2.6-5:

chr13    32907535    .    C    CT    2147483609.73    .    AC=1;AF=0.500;AN=2;BaseQRankSum=-2.106;DP=1000;FS=0.000;MLEAC=1;MLEAF=0.500;MQ=37.81;MQ0=0;MQRankSum=14.984;QD=29.49;RPA=11,12;RU=T;ReadPosRankSum=-9.803;STR    GT:AD:DP:GQ:PL    0/1:1261,1038:2453:99:7901,0,13152
chr13    32907589    .    G    GT    2147483609.73    .    AC=1;AF=0.500;AN=2;BaseQRankSum=6.976;DP=998;FS=0.000;MLEAC=1;MLEAF=0.500;MQ=37.74;MQ0=0;MQRankSum=25.865;QD=31.47;RPA=7,8;RU=T;ReadPosRankSum=-0.572;STR    GT:AD:DP:GQ:PL    0/1:1184,1142:2469:99:13365,0,16796

V 2.5-2:

chr13    32907535    .    C    CT    32729.73    .    AC=1;AF=0.500;AN=2;BaseQRankSum=0.023;DP=1000;FS=0.000;MLEAC=1;MLEAF=0.500;MQ=37.75;MQ0=0;MQRankSum=-3.054;QD=32.73;RPA=11,12;RU=T;ReadPosRankSum=3.137;STR    GT:AD:DP:GQ:PL    0/1:0,29:2453:99:7901,0,13152
chr13    32907589    .    G    GT    32729.73    .    AC=1;AF=0.500;AN=2;DP=999;FS=0.000;MLEAC=1;MLEAF=0.500;MQ=37.71;MQ0=0;QD=32.76;RPA=7,8;RU=T;STR    GT:AD:DP:GQ:PL    0/1:0,0:2469:99:13365,0,16796
Post edited by Geraldine_VdAuwera on


  • Geraldine_VdAuweraGeraldine_VdAuwera Posts: 7,781Administrator, GATK Dev admin

    Hi there,

    What you're seeing is related to limitations on the value we can represent in that field. So, not really a bug, but a sort of artifact from the format. You don't need to worry about it, just consider that those sites have very high QUALs.

    Geraldine Van der Auwera, PhD

  • mrubiocmrubioc Posts: 2Member

    Hi Geraldine,

    thank you for your answer.

    But I can't see exactly why the QUAL field has limitations on the value it can represent. If values are extremely high, an exponential format can be used, for example.

    In addition, if there is a sort of artifact in the representation, why doesn't it happen from some QUAL value onwards?, because in the entries of my previous example the QUAL should be ~ 33000 (QUAL~QD*DP), and in other tests I have seen values larger than this correctly calculated/represented.

    And also, there is another important issue, with these erroneous values in the QUAL field, the variants can't be correctly ranked by QUAL.

    Finally, have you found a reason for the tendency of QUAL's decimal part to begin with a 7?

    Thank you.

  • ebanksebanks Broad InstitutePosts: 684Member, Administrator, GATK Dev, Broadie, Moderator, DSDE Member, GP Member admin

    Hi there,

    You are right that we could use a different representation so that the QUAL values are not limited, but we choose not to (primarily for memory concerns, important when working with large whole genome projects). The 32729 cap from older versions arose when we capped the output to be represented as a "short" in Java; when we relaxed the constraint to allow full "int" representation it allowed the values to go much higher (2147483609, as you have noticed). Ultimately with QUAL values so astronomically high those variants will always be the highest confidence calls in your set, so at that point it's just not worth trying to eke out more resolution (at least not to us). And the decimal beginning with 7 that you are seeing also has to do with the max integer constraint we are imposing (and its interaction with the rounding to 2 significant digits in Java); it's not something that should be concerning.

    Eric Banks, PhD -- Senior Group Leader, MPG Analysis, Broad Institute of Harvard and MIT

Sign In or Register to comment.