In this seminar, Dr Pardo will present his group’s AI algorithm DrugSynthMC2.0 which uses Monte Carlo Tree Search to design drug molecules for bespoke targets. The algorithm uses prior knowledge of atom combinations and chemical bond type frequencies gathered from the FDA-approved drugs to generate 3D-viable chemical molecules that adhere to Lipinski rules. These are then docked to the 3D structure of the target protein to determine predicted binding affinities. This information is then fed back to the algorithm which then learns and produces a next cycle of molecules with improved properties. The process is repeated until affinities are not further improved. Applying this method to identifying compounds targeting the androgen receptor (AR) and the epidermal growth factor receptor family member, HER2, generates compounds with improved predicted binding affinity and pharmacological properties as compared to existing clinical compounds. They achieve this by both involving interactions with the targets that are common with clinical compounds as well as novel interaction types. Importantly, the entire process only takes two weeks using 4 GPUs. The algorithm can be applied to any target types (protein, RNA,…) as long as a 3D structure can be identified.