search

omerwe / polyfun

PolyFun (POLYgenic FUNctionally-informed fine-mapping)
MIT License
96 stars 22 forks source link

Tuning of sample #142

Closed mommy003 closed 1 year ago

mommy003 commented 1 year ago

Hi Omer,

Thanks for developing this method for cross-ancestry genomic prediction. Did you use any tuning procedure either discovery or target population for PolyPred and/or PolyPred+? Thanks in advance.

Kindest regards

omerwe commented 1 year ago

Hi @mommy003, we didn't implement any tuning procedure, though it's technically possible to do so. We have many results showing that the underlying methods (PolyFun BOLT-LMM, PRS-CS) are quite insensitive to the hyperparameters values (both from our own experiments and those of others). Hence, I hypothesize that the benefits of tuning would be pretty minor compared to the time and effort involved.