King's College London

Dr Antonio Rago, Lecturer in Computer Science, in a partnership with Professor Alexei Lapkin from the University of Cambridge and Professor Francesca Toni from Imperial College London, has won an £800,000 grant from the Alchemy Frontier Fund to develop AI models to accelerate the development of pharmaceuticals. 

Supported by industry leaders AstraZeneca and a start-up Chemical Data Intelligence, the team will deploy specialised AI agents to support human experts in common computational chemistry approaches for drug development.  

The team believe that this would help get impactful medicines out of the lab and into clinics and hospitals faster, resolving the difficulty of bringing complex chemical manufacturing into practice. 

AI is increasingly being used to drive advances in discovery science, in engineering and in manufacture, with AI-for-chemistry being a key discipline.  This approach uses AI to predict structures and properties of molecules and conditions for chemical reactions to design new manufacturing processes with minimised physical experiments.

The field integrates many traditionally divided disciplines of chemistry and process development into a holistic approach that allows for potential new functional molecules and materials, such as medicines, to be developed and brought to market at lower cost and more sustainably.

Despite this promise, the field has come up against issues which stem from a lack of high-quality data and the complexity of development and manufacturing processes. While not unique to a particular field of chemistry in the case of medicines, where the creation of active ingredients can be a long and complex process and manufacturing has to adhere to strict quality and security criteria, there is a particular challenge of scaling research outputs into commercial practice.

To counter this, the team will develop argumentative agentic AI, i.e. intelligent agents that use computational argumentation to reach decisions in an explainable manner. These agents will be integrated into the multi-agent reinforcement learning algorithms to drive computational design in chemical process development. 

Learning from the expert knowledge which already exists to keep its outputs grounded in reality, the agentic AI will used a formalised framework, or ontology, of chemical processes while it is run through an automated simulation environment developed at the University of Cambridge.

Alongside this new human-in-the-loop approach, the next-generation tool is to also be benchmarked against a set of tests to assess its performance in a number of different tasks in chemical process development, including the scale up of processes, to improve confidence in its outcomes.

Dr Rago said “Computational approaches to drug discovery hold massive promise in accelerating impactful medicines, but problems around scale up present a significant challenge. Yet, we already hold the key to answering this, human knowledge gathered from decades of experimentation. 

“By integrating this human expertise through novel, explainable AI agents, we can create a system we know that industrial partners have been searching for – an easily explainable way to re-configure and optimise the synthesis of potentially lifesaving drugs.” 

Over the next two years of the project’s lifecycle, the team hope to build the pipeline for this approach in pharmaceuticals, as well as trial its usage in other areas of chemical science involved in the manufacture and processing of materials.