I am interested in running a GWAS for a binary phenotype in a diverse ancestry dataset using GENESIS. In the bioRxiv paper (https://doi.org/10.1101/409953) I had read that GENESIS was using the LMM OPS model. In the latest paper of Gogarten et al,., 2019 (10.1093/bioinformatics/btz567), it is mentioned that the fitNullModel uses the penalized quasi-likelihood (PQL) approximation to the GLMM (Breslow and Clayton). Which is the difference between the two? I want to make sure that I'm using the best model for an admixed population.
LMM-OPS refers to the method used for separating distant ancestry (population structure) from more recent relatedness (close kinship in families). In GENESIS this is accomplished with the functions pcair and pcrelate. If you run an association test using PCs calculated with pcair and a covariance matrix calculated with pcrelate, then you are using the LMM-OPS method. The PQL approximation to GLMM refers to the method used to adjust for the covariance matrix when the outcome phenotype is binary (as opposed to continuous). If you run fitNullModel with PCs from pcair as covariates, a covariance matrix from pcrelate, and family = 'binomial', then you are using both of these methods.
Many thanks for the quick reply! Indeed, we run two iterations of PC-AiR and PC-Relate and then use these outputs for the association testing. I thought that the two methods were for the same purpose, but it makes sense now! Many thanks!
Many thanks for the quick reply! Indeed, we run two iterations of PC-AiR and PC-Relate and then use these outputs for the association testing. I thought that the two methods were for the same purpose, but it makes sense now! Many thanks!