Developing a new drug currently costs around $2.3 billion and fails 96% of the time. This project builds a low-cost robotic system that uses artificial intelligence to design and synthesise small molecule fragments, aiming to slash that waste. The core problem is that existing AI models often propose molecules that are impossible to make in a lab, and the few automated systems that exist don't learn from their own experimental failures. ALGeNeM solves this by linking a generative AI—trained on real chemical reactions like Suzuki couplings—directly to an automated synthesis platform. The system will run experiments, record which designs actually worked, and feed that data back into the model to improve the next round of suggestions. If successful, the platform could cut years and hundreds of millions of pounds from the drug development pipeline. That would make it economically viable to pursue treatments for neglected diseases that currently lack commercial incentive. It would also reduce the environmental footprint of pharmaceutical R&D by replacing thousands of manual, wasteful experiments with targeted, automated ones. All code and data will be released openly, allowing other labs to build on the framework.
View original technical description
ALGeNeM: molecule design with active learning and generative modelling for experimental automation The current drug discovery process is inefficient, with each new drug costing approximately $2.3 billion, taking 10-15 years to develop, and having a 96% failure rate. This presents significant economic and moral challenges, especially given the substantial number of diseases that remain untreated. To address these issues, this project aims to develop Active Learning for GeNerative Models (ALGeNeM), a modular, low-cost automation system that integrates generative modelling and active learning for experimental drug discovery. This project focuses on the chemical synthesis of small molecule fragments designed using X-ray crystallographic structural data, specifically Suzuki couplings, Buchwald Hartwig reactions and amide couplings for neuraminidases and anti-malarial kinase targets. Through ALGeNeM we will generate technology that will improve the efficiency of the drug discovery process. Recent advancements in automation and generative algorithms have demonstrated potential to accelerate and improve the efficiency of drug discovery. However, current generative models often produce outputs that are not synthesisable and modern active learning approaches are underutilised and underexplored for experimental work. ALGeNeM seeks to overcome these limitations by integrating state-of-the-art generative models conditioned on experimental capabilities and exploring the impact of different active learning strategies in an experimental context. This project will also develop a database schema integrated with a model building and active learning framework, enabling the assessment, and recording of active learning strategies' impact on real-world optimization problems. All code and data will be openly released under permissive licenses. I will build ALGeNeM using pre-existing open-source frameworks for automated labs and through this will provide an extensible platform for further research and integration. The project is timely, leveraging recent innovations in low-cost automation and generative models to create a multi-purpose framework for efficient drug discovery. It leverages strategic investments at the University of Liverpool (Material Innovation Factory and Digital Innovation Facility), and at Diamond Light Source’s XChem facility. By leading to the generation of large and open and homogeneous experimental data points directly and through collaboration, ALGeNeM will aid researchers within the UK and globally. By reducing the cost and time of discovering new medicines, ALGeNeM will enhance the productivity of the UK pharmaceutical industry, reduce environmental impact from unnecessary experiments, and improve public health outcomes through more affordable medicines. This project will enable the growth of my independent research career by building my own groundwork for automated and AI driven design of small molecules. It will build a family of generative models and a unique software framework from which I can apply for substantial follow-on grant funding from industry, charity, and research councils. It will enable me to support my own independent fellowship and provide the training opportunities to enhance my personal development as an independent academic.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know