As part of the events organized by the Thomas Young Centre (TYC), we are excited to announce two upcoming talks on Thursday 19th March 4:00-6:00 PM:
The talks will be delivered both in-person and online, with a drinks reception for those attending in person. Please see the event website for more information, and see the talk abstracts below.
“Speak to a Protein: AI Co-Scientists for Interactive Drug Discovery”
In this talk, we introduce Speak to a Protein, an interactive multimodal AI co-scientist for drug discovery. The system brings together scientific literature, structural biology, ligand knowledge, molecular visualization, and code execution into a single conversational interface. It can answer questions grounded in a live 3D molecular scene, highlight and manipulate structural features, retrieve and synthesize evidence across sources, and generate analyses on demand, explaining results through words, graphics, and interactive views.
Rather than treating AI as a passive search or summarization tool, Speak to a Protein points to a new model of scientific interaction: one in which researchers collaborate with systems that help them think, interrogate evidence, and generate hypotheses in real time. We show how this capability can accelerate tasks such as identifying binding pockets, comparing conformational states, exploring structure-activity relationships, and moving rapidly from question to insight.
More broadly, this work suggests a future in which AI co-scientists lower the barrier to complex molecular reasoning, make advanced analysis more widely accessible, and help reshape how discovery science is done.
"Data-driven Interatomic potentials for computer-aided drug design"
Drawing on computational methods that are based around training to extensive condensed phase physical property and quantum mechanical datasets, I will describe some of our efforts to design accurate and transferable inter- and intra-molecular potentials, with a view to applications in condensed phase atomistic modelling and computer-aided drug design.
I will explain how recent collaborations with the Open Force Field Initiative enable the automated development of fast, accurate force field models. I will describe the development of a graph neural network based charge model targeting accurate electrostatic properties of organic molecules, and the use of Open Force Field infrastructure to rapidly train valence parameters on the GPU. Finally, I will describe MACE-OFF, a transferable force field for organic molecules created using state-of-the-art machine learning technology and first principles reference data.
