Predicting how molecules bind

Whether a drug works depends on whether it finds its target, how tightly it binds and how long it stays. I study these questions with docking, molecular dynamics and, increasingly, machine learning.

On the machine-learning side, I build models that learn from representations of proteins and small molecules to predict how they interact, and I prefer models whose predictions can be explained over black boxes. On the simulation side, I use molecular modelling to understand binding mechanisms in detail, for example how a single compound can act on several targets at once.

I am also developing LangMol, an integrated platform for molecular modelling built around AI models. It brings these pieces together: molecule design driven by language models, protein modelling, docking, and binding and affinity prediction.