We've moved!
This site is now read-only. You can find our new documentation site and support forum for posting questions here.
Be sure to read our welcome blog!

Interpretation of LOD scores for CrosscheckFingerprints

FPBarthelFPBarthel HoustonMember ✭✭
edited July 2018 in Ask the GATK team


I am using CrosscheckFingerprints to compare sets of samples from the same individual (including tumor and non-tumor samples). My understanding is that LOD scores > 5 or so strongly indicate that the samples are from the same individual. From the GATK poster here I conclude that LOD scores typically range from 0 to 100. Nevertheless, I am getting what seems like LOD scores > 5000 across most of my samples and have not seen any LOD scores in the 0-100 range, ie.

  1. Should I take the log10 of these LOD scores to get the actual LOD score? If I do this however, the LOD scores are on the low end (3.5 - 4.5) though these samples are definitely from the same individual (those marked as expected match indicate multiple readgroups from the same sample) and a very high LOD score is expected, but perhaps not in the 5000+ range?
  2. What's the difference between LOD_SCORE_TUMOR_NORMAL and LOD_SCORE_NORMAL_TUMOR and when to use which? Depending on which sample is right and which is left in the comparison? What if both left and right side of the comparison are tumor samples?



  • FPBarthelFPBarthel HoustonMember ✭✭
    edited July 2018
    1. What is the correct definition for the LOD_THRESHOLD parameter for CrosscheckFingerprints (link) versus ClusterCrosscheckMetrics (link)? In the former an example of -5 is used and the explanation mentions that the -LOD_THRESHOLD parameter is used whereas in the latter an example of 3 is used and there is no mention of a negative LOD_THRESHOLD. This discrepancy is confusing and I am making the assumption that these parameters are different for both programs and that I should supply an inverse threshold (eg. -5) for CrosscheckFingerprints and an actual threshold for ClusterCrosscheckMetrics (eg. 5).
  • SheilaSheila Broad InstituteMember, Broadie ✭✭✭✭✭

    Hi Floris,

    Sorry for the delay, but I need to check with the team and get back to you.


Sign In or Register to comment.