Density plots post hard-filtering, good enough for VQSR resource?

sp580 · December 2018

Hello!!

I am trying to hard filter a variant call set using, as a first approach, the following thresholds, according to a related publication I found:
QD < 2.0 | FS > 60.0 | MQ < 40.0 | MQRankSum < -12.5 | ReadPosRankSum < -8.0

As I understand it, after going through the GATK documentation on hard filtering, there are no set rules to tell that the variant call set was (hard) filtered properly, other than visual inspection of the density plots, which requires experience.

For this reason, I would like to kindly require your inputs on the following plots, in order for me to see what are the next step I should undertake.

This is what I would venture:

QD: leave it as it is, enough variants filtered.

FS: according to log10 scale, leave it as it is, enough variants filtered.

MQ: may need higher stringency.

MQRankSum: according to log10 scale, leave it as it is, enough variants filtered.

ReadPosRankSum: according to log10 scale, leave it as it is, enough variants filtered.

Add SOR > 3.0?

I must mention that I want to use this variant call set as a resource for VQSR.

Thanks in advance!

bhanuGandham · December 2018

Hi @sp580

Here is a document that will help you in Understanding and adapting the generic hard-filtering recommendations.

bhanuGandham · December 2018

Hi @sp580

Here is a doc with suggested hard filtering steps:https://software.broadinstitute.org/gatk/documentation/article.php?id=2806

If you are going to apply VQSR then you do not need to hard filter the variants.

PS: You need atleast 30 exomes or 1 whole genome to do VQSR

sp580 · December 2018

Hi @bhanuGandham
thanks for getting back to me so quickly.

I actually followed the steps in the document you pointed out.

Regarding the amount of data I am working with, I have 60 whole genomes (~20x).

Also, I forgot to mention I am working with mouse and I am following the recomendations for when working with non-human models found here https://software.broadinstitute.org/gatk/documentation/article.php?id=1259, where it is stated that:

If you are working with non-human genomes, you will need to find or generate at least truth and training resource datasets with properties corresponding to those described below. To generate your own resource set, one idea is to first do an initial round of SNP calling and only use those SNPs which have the highest quality scores. These sites which have the most confidence are probably real and could be used as truth data to help disambiguate the rest of the variants in the call set.

In that sense, even though I will use VQSR, I should first define a costum resource by hard filtering the SNPs produced in an initial run. The plots shown in this post were produced from such data set.

I want to know whether applying these filters will produce an appropiate custom resource for VQSR.

Thanks in advance!

bhanuGandham · December 2018

Hi @sp580

Here is a document that will help you in Understanding and adapting the generic hard-filtering recommendations.

Test-drive the GATK tools and Best Practices pipelines on Terra

Density plots post hard-filtering, good enough for VQSR resource?

Best Answer

Answers