What is the best approach for computing in-sample LD for a larger cohort (+100.000), without it being very computationally intensive?
Replying to “What is the best approach for computing in-sample …”:
Great question - the biggest driver of how computationally expensive it is to calculate the LD is the region size - if you shrink the window [say 1 megabase rather than say 3 megabases] will have a big impact on the computational time - relatedly, I would encourage the use of covariate-adjusted LD calculation if there is a relatively standard association method and secondarily for the big cohorts such as those doing this centrally and sharing with others is very much encouraged
Replying to “What is the best approach for computing in-sample …”:
Apropos what Benjamin just said, would an in-sample LD of “just” 5000 people, randomly taken from the >100.000 cohort, be useable? Or could you just as well do the LD for the whole cohort, in that case?