During the session on estimating relatedness with PC-relate, it was highlighted that if the PCs were computed without excluding related samples, they would account (at least partly) for the genetic relatedness. Consequently, including these PCs as covariates in PC-relate would lead to deflated relatedness estimates.
Connecting this back to mixed-effects GWAS in data with related samples (e.g., using SAIGE), I had a couple of questions:
To better control for the genetic relatedness in the GWAS, would it be desirable to include as covariates the PCs computed using the entire dataset (instead of excluding related samples from PCA and then projecting them onto the PC space)?
Taking a step back, what additional utility do genetic PCs have as covariates in mixed-effects GWAS when we already have the GRM as a random effect?
Thank you so much for answering these lingering questions!
@bneale @matthew_keller @Wei_Zhou