Wrong variant calling

Healthcare & Life Sciences Parabricks
1

I have been comparing Parabricks DeepVariant (v4.4.0-1) with the standard DeepVariant (v1.6.1-gpu) and noticed some discrepancies in variant classification. Specifically, I observed that Parabricks DeepVariant tags some variants as RefCall when they seemingly should not be.

For example, consider the following variant at position chr7:72628972:

Output from Parabricks DeepVariant:

chr7    72628972        .       A       G       0.6     RefCall .       GT:GQ:DP:AD:VAF:PL      ./.:9:387:201,186:0.48062:0,8,17

Output from DeepVariant 1.6.1-gpu:

chr7    72628972        .       A       G       36.2    PASS    .       GT:GQ:DP:AD:VAF:PL      0/1:32:387:201,186:0.48062:36,0,33

Additionally, I checked this variant using Parabricks GATK, which produced the following record:

chr7    72628972        .       A       G       3775.64 .       AC=1;AF=0.500;AN=2;BaseQRankSum=-2.856;DP=381;ExcessHet=0.0000;FS=20.149;MLEAC=1;MLEAF=0.500;MQ=52.77;MQRankSum=-14.382;QD=9.99;ReadPosRankSum=1.416;SOR=1.209  GT:AD:DP:GQ:PL:SB    0/1:199,179:378:99:3783,0,4733:95,104,110,69

As shown, this variant has excellent coverage and quality metrics. I am puzzled as to why Parabricks DeepVariant categorizes this as RefCall, as we typically filter out such variants during our pipeline.

If it would be helpful, I am happy to share the FASTQ files for further analysis. Please let me know if any additional details are needed.

Looking forward to your insights.

2

Hi Amin,

Thanks for sharing this with us. It would be so helpful if you can share the fastq files, and the command that you used.

3

Hi,

Thanks a lot, to share the FASTQ files, could you please send me an email address?

For the code, I use the same code as in the documentation page:

https://docs.nvidia.com/clara/parabricks/latest/documentation/tooldocs/man_deepvariant.html

Thanks.

4

Sure thing!

Please share it with this one: tongz@nvidia.com

And I just want to double check that you used the command in our documentation page for both Parabricks and Google deepvariant, Is that right?

5

Also which GPU are you using?

6

Just sent you an email, with all the details.

Thanks a lot.

7

Hi @amin_davani

Based on the email I believe all of the mismatches have been solved by passing the correct arguments. Please let us know if you have more questions.

Thanks.