DIY a Molecule: Generative Models in Chemistry

Chemical Education Highlights

Abstract

This column introduces generative artificial intelligence and its application to molecular design. We contrast generative models with predictive models that most chemists have already encountered, build up the intuition behind conditional generation and explore how discrete diffusion models treat molecular design as a ‘fill-in-the-blank’ problem. Using a recently developed generative fragment-based drug discovery model, we provide a companion web application where chemists can interactively generate, evaluate and visualize novel molecules.