I'm analyzing UKB imputed data with fastGWA have a question about generating the sparse GRM. Assuming I'm working with properly QC'd imputed data (i.e. MAF/INFO/etc.), do you recommend making the initial GRM using the GREML-LDMS approach (partitioning GRMs by LD; cnsgenomics.com/software/gcta/#GREMLinWGSorimputeddata)?
Or would it be preferable to use a GRM created from directly genotyped/non-imputed calls?
I'm assuming that creating a GRM from imputed data without accounting for LD not ideal, but please let me know if I'm mistaken!