rsq Filter in Michigan imputation Server

Question

[Shicheng Guo]

Hi All,

Who knows why rsq filter in Michigan imputation Server only have 0.1, 0.2, and 0.3 three options to select? Usually, we will use R2>0.6 or R2>0.8 as threshold right?

Thanks.

Answer 1

[chenhaitao0919]

In this post, rsq = 0.3 is recomended: https://genome.sph.umich.edu/wiki/MaCH_FAQ

Answer 2

[jtull]

Hi Shicheng Guo,

When you select one of those cuttoffs, the MIS filters the SNPs BELOW that threshold. So, for example, if you select 0.3, you get rid of the SNPs with Rsq < 0.3 (which means btw, choosing to remove > 70% of poorly imputed SNPs). You could think of it as keeping the SNPs with Rsq >= 0.7. When you say "Usually, we will use R2>0.6 or R2>0.8 as threshold", maybe you refer to what you keep?

Best,

Answer 3

[AMARU]

Hi,

is R2 impute accuracy = rsq filter?, I am using Michigan Imputation server thanks

[yuanma0427]

Yes. According to here: https://topmedimpute.readthedocs.io/en/latest/getting-started.html#rsq-filter

To minimize the file size, Michigan Imputation Server includes a $r^2$ filter option, excluding all imputed SNPs with a $r^2$-value (= imputation quality) smaller than the specified value.