Interpreting ExcessHet INFO field

Hi team,
First thanks for the wonderful help and clarity of explanations on this website.

I am considering applying a filter on ExcessHet on my vcf files to have only markers that follow HWE. I understood that the ExcessHet value was the probability of getting the same or more heterozygotes as was observed unde HW conditions. But looking at my data, I get sites like this:

Scaffold_100 316384 . A C 1085470 PASS AC=55;AF=0.724;AN=76;BaseQRankSum=0.764;ClippingRankSum=0.00;DP=36981;ExcessHet=0.0000;FS=0.000;InbreedingCoeff=0.9342;MLEAC=55;MLEAF=0.724;MQ=57.15;MQRankSum=0.771;QD=29.50;ReadPosRankSum=0.118;SOR=1.517 GT:AD:DP:GQ:PL 1/1:1,120:121:99:4815,322,0 1/1:1,594:595:99:24888,1748,0 1/1:0,789:789:99:33158,2371,0 1/1:4,461:465:99:19229,1157,0 1/1:2,106:108:99:4322,245,0 1/1:9,279:288:99:11278,484,0 1/1:1,265:266:99:10808,754,0 1/1:8,246:254:99:10149,462,0 1/1:5,293:298:99:12072,726,0 1/1:1,734:735:99:30363,2167,0 1/1:9,302:311:99:12455,568,0

Where ExcessHet is 0 but the site is monomorphic for the alternate allele, so following the null hypothesis of HWE there should be no heterozygote, as observed, and I should get a high p-value?

Did I misinterpret something there?

As a result I am not so sure how to apply the filter...

Thank you

Joane

Best Answer

Answers

  • anapaulamachadoanapaulamachado LausanneMember

    Hi @mshand , I also trying to filter on ExcessHet. I've got the typical nagging question: how much is "quite high"? My values range from 0 to 174.
    Is there a easy way to convert phred-scaled to normal p-values? (not sure I understand your 10^0 explanation from before)

    Thanks,
    Ana

  • mshandmshand Member, Broadie, Dev
    edited July 20

    Hi @anapaulamachado, To convert the p-value back from phred-scale you'd use this formula:

    P=10^(-Q/10)

    So your value of 0 plugged into that formula is:

    P = 10^(0/10) = 10^0 = 1

    The value of 174 is:

    P = 10^(-174/10) = 3.98e-18

    As for choosing a cutoff, in our pipeline we filter anything more extreme than a z-score of -4.5 which is a p-value of 3.4e-06, which phred-scaled is 54.69. So we run VariantFiltration with the filterExpression "ExcessHet > 54.69".

    The more samples you have the more power you'll have to detect these kinds of artifacts so you'll filter more sites (and likewise the fewer samples you have, the fewer sites you'll be able to filter with this annotation). So you could start with 54.69 as a potential cutoff and adjust depending on your data as needed.

  • anapaulamachadoanapaulamachado LausanneMember
Sign In or Register to comment.