To celebrate the release of GATK 4.0, we are giving away free credits for running the GATK4 Best Practices pipelines in FireCloud, our secure online analysis portal. It’s first come first serve, so sign up now to claim your free credits worth $250. Sponsored by Google Cloud. Learn more at

Why I can not reproduce my result on different hostcomputer?

Dear colleague,
I have encounter a strange issues with GATK-3.3/3.7:
I used HaplotypeCaller to call variants on the same realn.recal.bam file, and it gave me different results when i ran on different computers (calls or annotation were not exactly identical) . I finally found that only if i ran on the same computer, i can get identical results between different repeats.

This dose not make sense, I think even if i use down-sampling, i should able to reproduce the result in any condition.I have stuck in the problem for a few days, can you give some advices? how could this happened and how to reproduce my result on different computers?

here is my commmands:

java -Xmx5G -jar GenomeAnalysisTK.jar -T HaplotypeCaller -R hg19.fasta -I sample.realign.recal.bam -L chr1 --emitRefConfidence GVCF --variant_index_type LINEAR --variant_index_parameter 128000 -dt NONE -o sample.chr1.g.vcf.gz
java -Xmx2G -jar GenomeAnalysisTK.jar -T GenotypeGVCFs -R hg19.fasta --variant sample.chr1.g.vcf.gz -o sample.chr1.vcf.gz -stand_call_conf 30 -stand_emit_conf 10 -dt NONE

PS: I kept using same java version for all my tests.


Best Answer


  • shleeshlee CambridgeMember, Broadie, Moderator

    Hi @aaronico,

    I believe there were changes in v3.6 that allow for calls to be identical. Prior to this version, there was a random element to calling. Also, and I'm not sure exactly which version this pertains to, there were changes to the emit/call confidence defaults.

    I have encounter a strange issues with GATK-3.3/3.7:

    Which version are you getting the non-identical results for, 3.3 or 3.7?

  • aaronicoaaronico ChinaMember

    Thanks shlee and Geraldine. I tried 3.3 and 3.7, both of them output non-identical results.
    I took Geraldine's advice to using '-pairHMM LOGLESS_CACHING', and problem solved! Many thanks!!

Sign In or Register to comment.