One of the most stubborn obstacles in computational drug discovery is the cold-start problem: how do you predict whether a new drug will interact with a protein target when the model has never seen a ...