The different latent vectors obtained for each run, {{{{bf{z}}}}}_{r}}_{r = 1}^{R} are then tested in a GWAS component by component, for association with genetic variants. The best association, with P value {P}_{{rm{best}}}^{(ell ,r)} , is found for each run r and region ell (each region is of around 2 Mb in length).
