Challenge questions on imputation

Imputation $r^2$

The R2 field is a measure of how well the imputation software THINKS it is doing - it is an interal estimate of what it thinks the correlation between the imputed and true genotype should be.

Question. What is the distribution of R2 across all variants? How does it vary with MAF?

Imputation ER2

The ER2 field measures the actual "leave-one-out" R2 for this variant (i.e. the correlation between what the model would have predicted for this variant and what it actually observed).

Question. How do R2 and ER2 compare? Is the software's guess at its accuracy (R2) close to the emperical accuracy (ER2) well? How does this vary with MAF?

Calling secretor status

The SNP rs601338 at chr19:48703417 in FUT2 determines whether an individual secrets the Lewis antigen. An individual who is homozygous AA doesn't secret Lewis antigen ("secretors"), and everything else does ("non-secretors"). We have provided a file (secretor/GGVP_secretor_status.txt) that includes the real secretor status for these individuals.

Question. How accurate is the secretor status called from imputation, in comparison to the real data provided?