Protocol for leveraging local ancestry and cross-ancestry genetic architecture to improve polygenic prediction in admixed populations

Polygenic prediction is often less accurate in admixed populations due to limited representation in training data. Here, we present a protocol for building a polygenic prediction model using admixed populations from the All of Us program. We describe steps for variant extraction, phasing, local ancestry inference, and construction of polygenic risk score. This protocol may also serve as a useful reference for researchers aiming to conduct large-scale analysis in the All of Us Research Workbench cost-effectively. For complete details on the use and execution of the protocol, please refer to Zhou et al. 1