Improving polygenic prediction from summary data by learning patterns of effect sharing across multiple phenotypes.

scalable to extremely large data sets. Using simulations and real data analysis, we show that our method is competitive with other methods for polygenic prediction.