MACH_R2 value greather than 1

[Daniel John Panyard]

Hi,

I'm working with some PLINK results from a GWAS, and I see there's a MACH_R2 output column described in the documentation as " MaCH imputation quality metric". Looking at those values, I see a number of variants (about 2.5 million out of 9 million tested variants) with a MACH_R2 > 1.0.

I'm not too familiar with this column or how it might be calculated during a GWAS analysis (instead of during imputation itself). Two questions:

1) Is it indeed the case that PLINK is calculating this metric itself from the input genotyping data?

2) What would cause an imputation value to be more than 1.0 (my understanding is that the imputation R2 is a value between 0 and 1.0)? Do such values indicate some problem with that variant such that its GWAS results wouldn't be valid?

Thanks for your help!

-Danny

[Christopher Chang]

1. Yes.

2. The MaCH-r2 metric is defined to be in [0, 2], not [0, 1]; see page 17 of the MaCH paper for its formula.  It can exceed 1 when e.g. population stratification results in fewer heterozygotes (and consequently higher genotype variance) than expected under Hardy-Weinberg equilibrium.  This should not invalidate the GWAS result.