GLIMPSE2_concordance: Some question about concordance

[Truongphikt]

Dear @srubinacci,

Thank you for a wonderful imputation toolkit. After doing an imputation by GLIMPSE2, I'm excited to evaluate the result using the GLIMPSE2_concordance tool. I did 2 tiny experiments to check GLIMPSE2_concordance tools. However, I got confused about the results, could you please explain them to me?

Test files

Experiment 1: Evaluate a VCF file vs itself

Same issue: #154

I evaluate a VCF file (Axiom_JAPONICA_EUR_chr12_split.vcf.gz) with itself. I expect Aggregated $r^2$ to be equal to 1 at all MAF bins.

This is my command:

echo -e " 12\tchr12_position.vcf.gz\tAxiom_JAPONICA_EUR_chr12_split.vcf.gz\tAxiom_JAPONICA_EUR_chr12_split.vcf.gz" > EUR-Axiom_JAPONICA_chr12_concordance.lst

GLIMPSE2_concordance  --input EUR-Axiom_JAPONICA_chr12_concordance.lst  \
                      --gt-val                    \
                      --af-tag MAF                   \
                      --samples EUR-Axiom_JAPONICA_12_listsample.txt

Result: Evaluate result doesn't show matched (Aggregated $r^2$ isn't equal to 1 at any bins)

Experiment 2: Evaluate without INFO/MAF information

My Question: Where do MAF values for binning come from?

I remove all INFO/MAF and INFO/AF values in the position file (chr12_position_nomaf_noaf.vcf.gz), true file (EUR_chr12_extract_nomaf_noaf.vcf.gz), imputation file (Axiom_JAPONICA_EUR_chr12_split_nomaf_noaf.vcf.gz) and do evaluation.

My command:

echo -e "12	 chr12_position_nomaf_noaf.vcf.gz   EUR_chr12_extract_nomaf_noaf.vcf.gz  Axiom_JAPONICA_EUR_chr12_split_nomaf_noaf.vcf.gz" > EUR-Axiom_JAPONICA_chr12_concordance.lst

GLIMPSE2_concordance     --input EUR-Axiom_JAPONICA_chr12_concordance.lst   \
                         --gt-val                    \
                         --af-tag MAF                    \
                         --samples ../EUR-Axiom_JAPONICA_12_listsample.txt                     \
                         --log ./concordance.log                     \
                         --output EUR-Axiom_JAPONICA_chr12_output                     \
                         --bins 0.0 0.016 0.03 0.05 0.06 0.083 0.1 0.116 0.13 0.15 0.16 0.183 0.2 0.216 0.23 0.25 0.26 0.283 0.3 0.316 0.33 0.35 0.36 0.38 0.4 0.416 0.43 0.45 0.46 0.48 0.5                     \
                         --out-r2-per-site

My expectation: Got an error because of missing INFO/MAF

Result: The command still runs successfully. I have no idea where MAF bins come from.

Summary

1. When concordance a VCF file with itself why does the Aggregated $r^2$ not equal to 1?
2. Where does MAF for binning (when using --af-tag MAF) come from (position file, true file, or imputed file)?