Best reference set

[Nicola Pirastu]

Hi,

I'm not sure that this is the right place to write this post on but it seemed more appropriate than the GenABEL forum:

I'm trying to select samples which will be sequenced whole genome to create my own reference set instead of HapMap or 1000G.
Since I have some isolated population my idea is to choose the best set based on kinship so that the imputations can be more acurate and I loose as less power as possible for association.
I've found out that minimac hides each marker and then compares the imputed marker with the original one. There is an output with three different measures of accuracy, the first is the canonical Rsq as in MACH called looRSQ , the second is empR which on the wiki is defined as: the empirical correlation between true and imputed genotypes for the SNP. If this is negative, the SNP is probably flipped.
The third is empRSQ defined as : the actual R2 value, comparing imputed and true genotypes.
From what I understand I should look at the third one in order to asses acuracy, however I'm not sure and I could not find any litterature on it.

Am I interpreting this correctly or should I try somthing different?

Best

Nicola

[yurii]

From reading your post - note I have no experience with minimac - I would agree that the last measure is probably the best (but it should also correlate almost perfectly with the first). And also #3 should be the same as the #2 - squared. I wonder if this is correct