scores (for height, BMI, and schizophrenia) and the first 20 PCs (N = 2577; see Supplementary Fig. 5 for representative scatterplots and Methods for additional details). Two key conclusions can be drawn from these figures. It is important that multiple, sometimes non-consecutive, PCs are correlated with polygenic scores for each of these phenotypes. For example, for height polygenic scores constructed with GIANT summary statistics, it is primarily the first PCs (1–4) that are significantly correlated with polygenic scores, but non-consecutive and later PCs are also correlated (e.g., PCs 7 and 12, in this example). Second, results vary somewhat across a range of p-value thresholds (pT) used for constructing polygenic scores. These points underscore that using only a small (e.g., under 10) number of PCs may be inadequate for polygenic scoring analyses of mixed ancestry and admixed samples, and that careful inspection of data and plots is always needed.
