Fast and robust ancestry prediction using principal component analysis

Population stratification (PS) is a major confounder in genome-wide association studies (GWAS) and can lead to false positive associations. To adjust for PS, principal component analysis (PCA)-based ancestry prediction has been widely used. Simple projection (SP) based on principal component loading and recently developed data augmentation-decomposition-transformation (ADP), such as LASER and TRACE, are popular methods for predicting PC scores. However, they are either biased or computationally expensive. The predicted PC scores from SP can be biased toward NULL. On the other hand, since ADP requires running PCA separately for each study sample on the augmented data set, its computational cost is high. To address these problems, we develop and propose two alternative approaches, bias-adjusted projection (AP) and online ADP (OADP). Using random matrix theory, AP asymptotically estimates and adjusts for the bias of SP. OADP uses computationally efficient online singular value decomposition, which can greatly reduce the computation cost of ADP. We carried out extensive simulation studies to show that these alternative approaches are unbiased and the computation times can be 10-100 times faster than ADP. We applied our approaches to UK-Biobank data of 488,366 study samples with 2,492 samples from the 1000 Genomes data as the reference. AP and OADP required 7 and 75 CPU hours, respectively, while the projected computation time of ADP is 2,534 CPU hours. Furthermore, when we only used the European reference samples in the 1000 Genomes to infer sub-European ancestry, SP clearly showed bias, unlike the proposed approaches. By using AP and OADP, we can infer ancestry and adjust for PS robustly and efficiently.