Experiment- and Human-Guided Representation Learning for Accelerated Chemical Discovery (Liverpool–Manchester)

About the Project

This PhD will develop fundamental AI methods that help chemists explore and navigate complex chemical spaces when data are scarce, high-dimensional, and continuously updated by new experiments. The core aim is to design representation learning approaches that can extract chemically meaningful structure from limited experimental labels, update as new measurements arrive, and remain aligned with how chemists actually make decisions.

A central challenge is that many chemical discovery problems have only a small number of experimental measurements, yet the underlying molecular spaces are vast and complex. Dimensionality reduction and representation learning can reveal hidden structure, but naïve compression risks discarding crucial chemical information and producing misleading insights. This project will address that by creating modelling strategies that prioritise what matters for downstream chemical objectives under sparse supervision.

You will develop AI modelling and analysis pipelines that:

• learn task-relevant molecular representations from limited experimental measurements,
• support incremental/online learning as new data arrives,
• incorporate human-in-the-loop guidance so domain experts can steer which patterns should be preserved for the chemical task,
• and evaluate performance in realistic chemical discovery workflows with close collaboration with chemists.

Training environment

This is a joint collaboration between two centres:

• Centre for AI Fundamentals, University of Manchester
• AI Hub in Chemistry (AIchemy), University of Liverpool

You will be supported by an interdisciplinary supervisory team spanning fundamental AI research in representation learning (Wood, D., et al., JMLR, 24(359):1−49, 2023) and human-in-the-loop discovery (Nahal, Y., et al., J Cheminform 16, 138 (2024)), as well as applied AI (Cissé, A., et al., IJCAI, 2025) and automation (Dai, T., et al., Nature 635, 890–897 (2024)) for chemistry. You will work across both sites and communities, but be primarily based at the University of Liverpool, with monthly in-person visits to the Centre for AI Fundamentals (University of Manchester) to engage with the broader research community and meet with the joint supervisory team.

Candidate requirements

We welcome applicants with a strong background and a Master’s degree in one or more of:

• Machine learning / data science / computer science / applied mathematics
• Physics / chemical informatics / related quantitative disciplines

Essential:

• Strong AI and math background (representation learning, uncertainty, or continual/online learning)
• Evidence of Python programming experience

Desirable:

• Enthusiasm for interdisciplinary research
• Interest in collaborating with experimental scientists and working with real discovery data

Supervisors:

Dr Vladimir Gusev

Dr Xenofon Evangelopoulos

Prof. Andy Cooper

Prof. Samuel Kaski

Dr. Tingting Mu

Please submit:

1. CV
2. Brief cover letter outlining your interest and relevant experience

to Dr. Xenofon Evangelopoulos (Xenofon.Evangelopoulos@liverpool.ac.uk)

Funding Notes

UK Tuition fees, stipend funded for 3.5 years. Funded from AI for Chemistry Hub based at University of Liverpool.