Hi,
I have summary statistics from association studies with data imputed using different softwares, IMPUTEv2 and Minimac. I am having trouble understanding whether the two imputation quality metrics produced by IMPUTEv2 and by Minimac, called respectively the “info score” and minimac ȓ2, have comparable values and one can use the same threshold and obtain similar results (for example by filtering out SNPs with Rsq and INFO < 0.3), or whether these two are different measures and a different threshold has to be used.
I have found examples in the literature using the same threshold independently from the Software used for imputation, but I have also found references that state that the two measures are correlated but have different ranges and values, so a different threshold should be used (see for example blog http://blog.goldenhelix.com/?p=1911). The ranges fro the two quality scores are indeed very different for the results produced with the different softwares (Minimac is from 0.3 to 1 while IMPUTEv2 is from 0.3 to 2), so I am tempted to try to find a statistical way to compare them using a rank method in R. Would this be correct?
Note I only have summary statistics with imputation quality metric for each SNP (otherwise maybe the script r2hat available by Yun Li at http://www.sph.umich.edu/csg/yli/software.html could be used). However I do not have access to the dosage data.
If anybody has an idea of how to do this in R or experience on this it would be very helpful to get your opinion.
Thank you in advance!
infoscore for both Minimac and IMPUTEv2 should be in range of 0-1. I think you can safely cut off at 0.3, see here.thank you zx8754 for your reply. I had seen this paper. The odd part is that my range for info score from IMPUTEv2 is actually from 0.3 to 2.00, while the paper you cite reports all scores between 0-1. There are papers that state that occasionally the scores are greater than 1 (the Marchini paper). However this worries me because the metrics don't seem comparable anymore one to another... I have also found other references for using for example Rsq>0.3 for Minimac and INFO>0.4 for IMPUTEv2 (http://eurheartj.oxfordjournals.org.libproxy.ucl.ac.uk/content/35/8/524.full and the blog cited in the question where it is stated that these are highly correlated but different metrics with a different range), so it is unclear to me what the right thing to do is…Even if we have different ranges and the metrics are different, is it still correct to use the same cut-off?