MIT researchers have developed a deep learning model that can rapidly predict the likely 3D shapes of a molecule given a 2D graph of its structure. This technique could accelerate drug discovery by narrowing down the number of molecules pharmaceutical researchers need to test in lab experiments.