Help! about GATK realignment speed

[wzhangvv]

Hi All,
I try to do SNP calling using GATK. This is my first time to do such things and I generated the work flow as follows. Everything went well till I was blocked by RealignerTargetCreator. It seemed to cost 15 days per sample! I don't know whether it was a normal speed with 300MB reference and 1GB bam file or not. Could anybody help me figure it out? I have 80 samples and obviously I don't have enough time to run this step.
Thanks for your time~!

My work flow (till RealignTargetCreator):
I used sga to do de novo assembly and I used the output file contigs.fa as reference.

bwa index -P contigs.fa -a bwtsw contigs.fa

bwa aln -t 4 contigs.fa R1.fq > R1.sai

bwa aln -t 4 contigs.fa R2.fq > R2.sai

bwa sampe contigs.fa R1.sai R2.sai R1.fq R2.fq > A.sam

samtools view -bST contigs.fa -o A_noRG.bam A.sam

java -Xmx20g -XX:PermSize=10g -XX:MaxPermSize=10g -jar /usr/share/picard/lib/AddOrReplaceReadGroups.jar INPUT=A_noRG.bam OUTPUT=A_std.bam SORT_ORDER=coordinate RGID=lib1_A RGLB=AA RGPL=illumina RGSM=lib1_A RGPU=none VALIDATION_STRINGENCY=LENIENT

java -Xmx20g -XX:PermSize=10g -XX:MaxPermSize=10g -jar /usr/share/picard/lib/MarkDuplicates.jar INPUT=A_std.bam OUTPUT=A_std_noduplicates.bam METRICS_FILE=A_std.duplicate_matrics REMOVE_DUPLICATES=true ASSUME_SORTED=true VALIDATION_STRINGENCY=LENIENT

java -Xmx20g -XX:PermSize=10g -XX:MaxPermSize=10g -jar /usr/share/picard/lib/BuildBamIndex.jar INPUT=A_std_noduplicates.bam VALIDATION_STRINGENCY=LENIENT

java -Xmx20g -XX:PermSize=10g -XX:MaxPermSize=10g -jar /usr/share/GenomeAnalysisTK-2.1-10-gdbc86ec/GenomeAnalysisTK.jar -T RealignerTargetCreator -nt 8 -I A_std_noduplicates.bam -R contigs.fa -o A_forIndelAligner.intervals

[adaptivegenome]

Try OpenGE for realignment:

https://www.github.com/adaptivegenome/OpenGE

[wzhangvv]

Hi,

Thanks for your reply!

I just read the menu of OpenGE. It can't help me because its localrealign step requires the intervals file which need to be generated by GATK RealignerTargetCreator, which was the very slow step I mentioned before.

Do you think my RealignerTargetCreator speed is normal (~10+ days per sample)? If it really is, I have to change my strategy.

Try OpenGE for realignment:

www.github.com/adaptivegenome/OpenGE

[qtrinh]

Hi,
What about running RealignerTargetCreator in parallel on each of the chromosomes ? This should speed things up for you.

Q

[wzhangvv]

Hi,
Thanks for your reply!
I just did de novo assembly and got millions of contigs without any chromosome info... The frog species I worked on doesn't have an assembled genome...
I tried to get more scaffolds, but it was difficult to my rad data.

Hi,
What about running RealignerTargetCreator in parallel on each of the chromosomes ? This should speed things up for you.

Q

[adaptivegenome]

I'm surprised that TargetCreator is the limiting step. Typically runs quite fast.

[wzhangvv]

Hi,
Do you think there is anything wrong in my workflow? Can you give me some advice? I really don't know how to improve and debug it because the previous steps finished smoothly.
Really appreciate your reply!

I'm surprised that TargetCreator is the limiting step. Typically runs quite fast.

[Zaag]

use targeted regions so it doesn't walk over the entire genome (unless it's whole genome data)

[wzhangvv]

Hi,
This is the whole genome data and I don't know the exact targeted regions...So I can't start with a known VCF file.
Thanks for your reply!

use targeted regions so it doesn't walk over the entire genome (unless it's whole genome data)