Speed of variant callers

[mboursnell]

My variant calling seems very slow. What do you think?

We have 10 BAM files, each about 2.5GB, covering a targetted region of about 15MB.

I am using the GATK HaplotypeCaller with 8 threads (-nct 8) and it is taking 31 hours.

When we start whole genome sequencing this will be impossible!

Any ideas on how to speed things up?

[dpryan]

You'll really want to follow Brad Chapman's blog. He writes a lot about various algorithm comparisons and also works on bcbio, which is probably what you're really looking for (see an associated blog post of his here).

[shi]

My variant calling seems very slow. What do you think?

We have 10 BAM files, each about 2.5GB, covering a targetted region of about 15MB.

I am using the GATK HaplotypeCaller with 8 threads (-nct 8) and it is taking 31 hours.

When we start whole genome sequencing this will be impossible!

Any ideas on how to speed things up?

The exactSNP is a lot faster. And it has at least comparable performance from our experience.

[mboursnell]

OK, thanks. I'll try exactSNP. How did you do the comparison between exactSNP and GATK?

[shi]

We are now preparing the manuscript for exactSNP. A preprint should be available early next year.

[adamyao]

If you have pair-end fastq files you can upload them to HiPipe directly (http://hipipe.ncgm.sinica.edu.tw/). HiPipe powered by 640 cores can do alignment (BWA) and variant calling (GATK) very fast. A human exome (70X-100X) can be processed in 2 hrs excluding upload time.