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!


shaky_dingoshaky_dingo VancouverMember

I am doing exome sequencing in 700 individuals from a species with a large genome and I would like to use GATK to realign around indels. I am using a reduced reference, which is still about 3Gb. I tested out the target creator, but it is taking 5 days for 12 individuals when each is done individually and this time frame is not feasible for all 700 individuals. I tried to run more in parallel (~30 individuals), but there are RAM limitations on our 250G server. I am currently testing out the program by running all 12 test samples as input for the same run and the time estimate is very long (on the order of several hundred weeks). Based on a preliminary run I have also included a vcf file with likely indels to try and speed the process. Can you suggest another way in which I can make the time frame for all 700 individuals more reasonable? Otherwise we will not be able to use this tool.


  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    Hi there,

    Can you tell me the version you are using and post your command line?

    Are you passing in the list of target intervals for your exomes with -L? If not that should significantly speed up the process. And are you using the multithreading options?

  • shaky_dingoshaky_dingo VancouverMember

    I am using GenomeAnalysisTK-2.4-9

    This is the command for the individual target creator:
    java -Xmx15g -jar $GATK_PATH/GenomeAnalysisTK.jar -T RealignerTargetCreator -R $REF -nt 10 -o $OUT_PATH/$1.bam.list -I $OUT_PATH/$1.mdup.sort.bam --known $OUT_PATH/INDELS.vcf >> $OUT_PATH/$1.log 2>&1

    I am also trying it with multiple input bams and a great deal more memory.

    We have already reduced the genome to contigs with blast hits to our target sequences (the genome is highly fragmented), however I could try using the locations of the aligned target sequences as the target area is much smaller than the reduced genome. However, we are expecting to recover areas adjacent to the target sequences that might have indels.

    Thanks for your help!

  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    If you are concerned about losing information in the areas adjacent to the target sequences, you can pad the intervals. At Broad we typically pad the intervals by 50 bp to capture that information.

    How many contigs do you have in your reference genome? We have seen that genomes with hundreds or more contigs (such as draft genomes) cause a huge inflation in runtimes.

  • shaky_dingoshaky_dingo VancouverMember

    The genome is a draft and there are hundreds of thousands of contigs in the reduced reference that we are aligning to, so perhaps that is contributing to the long run times. I will try padding the intervals. Thanks again!

  • Geraldine_VdAuweraGeraldine_VdAuwera Cambridge, MAMember, Administrator, Broadie admin

    Ah yes, that explains it. GATK is not designed to process highly fragmented references.

    If using the intervals is not enough, you might try making pseudo-scaffolds of subsets of contigs to reduce the number of contigs the GATK has to handle. It will require that you keep track of the contigs within the scaffolds, but it will certainly help reduce runtimes.

Sign In or Register to comment.