AIchemy’s Monthly Webinar Series – February 2026

We are delighted to welcome you to our AIchemy Hub’s monthly webinar series.

This month’s talks:

Prof. Natalie Fey – University of Bristol

Talk Title: Organometallic Chemistry x Data Science

Computational studies of homogeneous catalysis play an increasingly important role in furthering (and changing) our understanding of catalytic cycles and can help to guide the discovery and evaluation of new organometallic catalysts. While a truly “rational design” process often remains out of reach, detailed mechanistic information from both experiment and computation can be combined successfully with suitable parameters characterising catalysts and substrates to predict outcomes and guide screening.

In this presentation, I will use examples drawn from our recent work, including the exploration of maps of chemical space and of a reactivity database, to illustrate how we are increasingly applying data science techniques for visualisation and prediction, with the goal of informing the discovery and design of suitable organometallic catalysts.

Hyunsoo Park – Imperial College London

Talk Title: Rethinking Generative AI for Materials Discovery

Generative artificial intelligence (AI) has emerged as a potent paradigm for inverse materials design, offering the potential to invert traditional discovery workflows by directly proposing structures that satisfy desired properties. However, a fundamental challenge persists regarding the standard training objectives used in generative AI versus the goals of materials discovery. A critical misalignment exists between the likelihood-based sampling typical of generative modelling and the targeted focus on underexplored regions required to identify novel compounds.

To address this challenge, this talk presents Chemeleon2, a framework that reformulates crystal generation as a reinforcement learning (RL) task. Through the integration of Group Relative Policy Optimization (GRPO) with latent diffusion models, the system optimizes multi-objective rewards to simultaneously achieve stability, diversity, and novelty. The presentation further details how this methodology facilitates property-guided design, ensuring chemical validity while isolating desired functionalities. Ultimately, this approach establishes a modular foundation for controllable, AI-driven inverse design, effectively addressing the novelty-validity trade-off inherent in scientific discovery applications.