Varying the number of principal components for modeling sample structure

We examined the sensitivity of Coal-Map to the number of covariates used to model global and local sample structures. First, we represented the local sample structure using the top three covariates obtained after applying principal components analysis on the local partition Xl containing the test locus xj. Global sample structure was represented using the top two covariates after performing principal components analysis on the full alignment X excluding the local partition X`. This resulted in two models: one using the covariates W global j = (w1, w2) and the other using the covariates W glocal j = (w1, w2 . . . w5), respectively. We selected one of the aforementioned two models for each test locus xj using the heuristic approach described in the Methods section. In Figure S1, the performance of Coal-Map, using five covariates to represent sample structure (two for global and three for local), and EIGENSTRAT is shown using receiver operating characteristic (ROC) curves. Using Delong et al. test [1] with Benjamini-