This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Beagle imputation results quality control

My region of interest is ~ 120 kb. I have 300 samples, all containing 14 SNPs in the region. I tried to impute using beagle to get more SNPs. I used CEU population (the closest one) as a reference. After filtering CEU samples I got about 350 SNPs per sample to use a reference.

After imputation I got the same number of SNPs in my population as in CEU. Is it legit to use them all? I have a gut feeling that I have to filter them according to the imputation quality or something. How do I do that? The output VCF looks something like this:

1       110187031       rs113581509     C       T       .       PASS    AR2=0.468;DR2=0.514;AF=0.06     GT:DS:GP        0|1:0.759:0.242,0.758,0 0|0:0.018:0.982,0.018,0 0|0:0.018:0.982,0.018,0 0|0:0.001:0.999,0.001,0

Can anyone give me a clue on how to filter the results? Or maybe I should use another software?

imputation beagle vcf
• 10,587 views • link
updated by zx8754 zx8754 • written by eyb eyb

Hi eyb,

I know it's been several years ago, but right now I'm facing the same problem that you had in that moment: I've just achieved to impute my data with Beagle, but now I would like to know how to filter out the bad quality SNPs.

I suspect that it is related with the DR2 field, but I'm not quite sure about it... Did you finally resolve your problem?? Thank you very much in advanced!

• 0 views • link
written by biosol biosol

1 answer

There is no consensus on how best to filter the post-imputation results. You can use a combination of AR2 (allelic R-squared), DR2 (dosage R-squared), and MAF. Take a look at this pre-print and subsequent publication, where they actually did not do any filtering post-imputation:

Very low depth whole genome sequencing in complex trait association studies (peer reviewed publication).

Variant-level QC

Beagle provides two position level imputation metrics, allelic R-squared and dosage R-squared. Both measures are highly correlated (Supplementary Fig. S8a). Values between 0.3 and 0.8 are typically used for filtering (Brian Browning, personal communication).

Kevin

• 0 views • link
written by Kevin Blighe Kevin Blighe

Great, thanks, I'll have that in mind!

• 0 views • link
written by biosol biosol

Update: Beagle 5.0 only have DR2.

• 0 views • link
written by Shicheng Guo Shicheng Guo

Hi Kevin, 

java -Djava.io.tmpdir=./temp/ -Xmx32g -jar beagle.16May19.351.jar impute=false gt=Exome.vcf out=Exome.vcf.phasing

Like above example. Suppose, we don't use map and reference parameters, what's the accuracy of the phasing for beagle?

Thanks.

• 0 views • link
written by Shicheng Guo Shicheng Guo

Log in to answer this question.

More posts like this