How to calculate Imputation Accuracy Estimates like concordance with BEAGLE?

Question

[Shab86]

Hi all,

I have an imputed file output from BEAGLE IMPUTED FILE Rows are SNP's and columns are individuals. Now I would like to calculate some imputation accuracy estimates like concordance or Rsq and then plot them across MAF(minor allele frequency). How do I calculate them? Are there any tools which could generate such statistics?

Any help is highly appreciated.

Answer 1

[vskale135]

Hello All,

I would also like to determine imputation accuracy in our GBS dataset. Here is what I did:

1) randomly selected 1% SNPs: zcat all.vcf.gz | awk '$1~/^#/ || rand()<=0.01' | bgzip -c > eval.vcf.gz 2) exclude the evaluation sites from the original VCF : bcftools isec -C all.vcf.gz eval.vcf.gz -Oz > impute.vcf.gz 3) imputed the missing data using beagleV4 : java -Xmx100g -jar beagleV4.1.jar gt=impute.vcf.gz out=imputed window=100 overlap=30 niterations=10 when I compard the imputed.vcf.gz and eval.vcf.gz using vcf-compare, I got followign output:

SN Number of REF matches: 0 SN Number of ALT matches: 0 SN Number of REF mismatches: 0 SN Number of ALT mismatches: 0 SN Number of samples in GT comparison: 0

I request you to please help.

Thanking you with best regards

Sandip

[Gennady Khvorykh]

To estimate the quality of imputation, I think, imputed.vcf should be compared with all.vcf.gz and not with eval.vcf.gz :)

Answer 2

[Zev.Kronenberg]

Three step process.

1. Drop some % of your genotype calls.

2. Impute

3. Measure VCF concordance of original and imputed VCF file.

I've done this. You should play with the % of genotypes you remove and MAF.

[Shab86]

Thanks Zev for your reply. I have already done step 1 where I removed bad quality calls, and then masked the genotyped file which then I used in BEAGLE for imputation. Now I have the imputed file and the original one , and from these files I would like to get those accuracy estimates like concordance. The imputed file is the one I had attached in the original post. Any idea with how to get those estimates?

[Zev.Kronenberg]

To test the accuracy you remove high quality variant calls. Not low quality calls.

[Shab86]

I have done the steps 1 & 2 and my main query is about no. 3. Are there any tools etc which I can use to get step 3?

[vskale135]

You can use vcf-compare or bcftools stats to get stats which you can plot using plot-vcfstats. Can you please let me know, how you performed the step1 and step2. I don't have reference panel.

Thanks and regards

Sandip

[akang]

I'm trying to do the same thing..just that my data is multi allelic. How do we calculate imputation accuracy for multi allelic data? Ill appreciate any help on that.Thanks!