concordance between GATK and Samtools

[caswater]

Hi, I am just not sure about the concordance between GATK and Samtools mpileup.

I have one sample, and the variants called by GATK and Samtools showed 99% concordance for a total of ~3.1 million SNPs. is this number a typical one? The concordance was computed using GATK's SelectVariants module.

Thanks a lot for your advice

[adaptivegenome]

So I believe in the recently published HUGESEQ paper, they report greater than 90% concordance between GATK and samtools.

[lh3]

99% seems impossible

[caswater]

I tested the overall overlap between SNPs called by GATK and Samtools. In total, 4,115,639 was called by GATK (without imposing filters), and 3,836,479 were overlapped with samtool calls. Then the overlap rate is 93%.

After GATK filtering, 3,097,504 GATK calls among 3,111,272 total are overlapped with samtool calls, giving the overlap rate 99%

I am not sure if my procedure is correct? I used GATK SelectVarints module and get the concordance between the two sets.

Thanks a lot for your advice on this.

[Heisman]

I tested the overall overlap between SNPs called by GATK and Samtools. In total, 4,115,639 was called by GATK (without imposing filters), and 3,836,479 were overlapped with samtool calls. Then the overlap rate is 93%.

After GATK filtering, 3,097,504 GATK calls among 3,111,272 total are overlapped with samtool calls, giving the overlap rate 99%

I am not sure if my procedure is correct? I used GATK SelectVarints module and get the concordance between the two sets.

Thanks a lot for your advice on this.

When you calculate concordance like that you are ignoring the calls made by SAMtools that were not made by GATK. So, if SAMtools is calling a lot more SNPs (after GATK filtering) and the GATK calls are almost all in SAMtools, that implies with those parameters that SAMtools is some combination of more sensitive and less specific.

[ramirob]

Hi everyone,

In relationship to this, I am trying to assess concordance as well, but on the contrary, I am getting an overlap of around 30%! (as described by vcf-compare in vcftools). I am not sure what I am doing that leads to this result. My process is as follows:

bwaAlignment with hg19 -> remove duplicates | realign | recalibrate -> newAlignment

For samtools:
samtools mpileup -f ucsc.hg19.fasta -Q 30 -D -g -S newAlignment.bam | bcftools view -bvcg - > samtools.raw.vcf

for gatk:
java -Xmx4g -jar GenomeAnalysisTK.jar --min_base_quality_score 30 -I newAlignment.bam -R ucsc.hg19.fasta -T UnifiedGenotyper -o gatk.raw.vcf

Should this be giving more SNPs in common? Do I have any extra/missing parameters?

Thanks in advance,
Ramiro

[ramirob]

Hi everyone,

In relationship to this, I am trying to assess concordance as well, but on the contrary, I am getting an overlap of around 30%! (as described by vcf-compare in vcftools). I am not sure what I am doing that leads to this result. My process is as follows:

bwaAlignment with hg19 -> remove duplicates | realign | recalibrate -> newAlignment

For samtools:
samtools mpileup -f ucsc.hg19.fasta -Q 30 -D -g -S newAlignment.bam | bcftools view -bvcg - > samtools.raw.vcf

for gatk:
java -Xmx4g -jar GenomeAnalysisTK.jar --min_base_quality_score 30 -I newAlignment.bam -R ucsc.hg19.fasta -T UnifiedGenotyper -o gatk.raw.vcf

Should this be giving more SNPs in common? Do I have any extra/missing parameters?

Thanks in advance,
Ramiro