How to bechmark a human dataset ? (is it needed)

Question

[pinn]

Hi I'm working on exome dataset I generated variant calls from different combination of aligners vs variant callers. Can i benchmark my exome dataset with Genome-in-bottle (HG001, HG002,HG003, HG004 & HG005) to find the Truth positives (TP) and False positives (FP). Is this the way ? Benchmarking mean you can take any query dataset and do comparative analysis with confidence calls (HG001 ....HG005). Can any one explain, I'm very confused with benchmarking ?

Thanks for your comments.

[Ram]

Based on your past questions, IMO you’re obsessing over TP/FP a lot. Run preliminary QC, extract variants of significance/interest and invest effort in in-depth QCing them.

Answer 1

[Jeremy Leipzig]

You can now benchmark your own variant calls against HG002/HG003/HG004 on Truwl using 4 popular regions (more coming) and see how your results stack up against the top PrecisionFDA Truth Challenge competitors.

https://medium.com/truwl/accessible-and-uniform-benchmarking-22f598616ef5

[pinn]

I waited so long, thanks for developing.

[jacobhsu]

Jeremy Leipzig How about benchmarking the SV/CNV?

[Jeremy Leipzig]

Yes we have GATK-SV running and are currently working with a contact at Stanford to choose a subset of metrics to display in the benchmarking table. I would love to hear about your needs if you are working with CNV/SVs, so please sign up for a free trial and we'll reach out.