The latest AIchemy-supported SMACT workshop at the Royal School of Mines brought together researchers pushing the boundaries of materials informatics. A fast-paced mix of lightning talks, live demos and collaborative hacking showcased both rule-based methods and AI-driven tools for materials chemistry.
Key Highlights
>_ SMACT roadmap – Anthony outlined development priorities and progress since the last SMACT workshop, focusing on strengthening the core library while opening clear hooks for data-driven methods.
>_ Oxidation-state filters – Kinga shared work on new filters with thresholds for commonality and consensus, improving how SMACT handles plausible oxidation states during screening.
>_ Element embeddings – Prof. Butler opened a discussion on using learned element embeddings for materials design, connecting traditional heuristics with representation-learning approaches.
>_ LLMs for chemistry tooling – Ry introduced progress on bringing large language models into the chemistry stack and integrating them via the Model Context Protocol (MCP) to make workflows more robust and composable.
Outcomes
- Property prediction module (WIP) – A new property prediction module is in development, integrating the ROOST deep-learning model to predict key properties (e.g. band gap, bulk modulus) directly from chemical composition — no crystal structure required. The design is extensible: additional models can be plugged in, with versioning for different fidelities (PBE, HSE06, experimental), enabling like-for-like benchmarking.
- Sustainability metrics expansion – SMACT’s decade-old HHI-style indicator is being broadened. Proposed additions include EU Critical Raw Materials listings, elemental toxicity, radioactivity, recyclability, and in silico stability proxies. The aim is a richer, transparent sustainability scorecard to support early-stage screening and decision-making.
- Mixed-valence validity checks – smact filters now optionally supports mixed-valence compounds (e.g. Fe₃O₄), improving validation for real-world systems where elements adopt multiple oxidation states.
Together, these additions help SMACT keep pace with the rapid growth of the computational materials design space, extending its scope to property prediction, sustainability screening and complex oxidation-state handling.
A huge thank you to everyone who joined us and made the day such a success. We look forward to continuing the conversation as part of the growing AIchemy Hub community.
If you would like to find out more about how AIchemy can support your workshop idea click the button below.