Mofjson improves language models understanding metal organic frameworks
MOFjson: An Explicit Topological Representation of Metal-Organic Frameworks for Language Models
Machine LearningSoftware Engineering
Summary
Metal-organic frameworks (MOFs) are complex structures made of metal atoms and organic parts. Language models that analyze MOFs usually must choose between raw atomic coordinates or simple string identifiers, which lose important information. The authors created MOFjson, a new way to represent MOFs that keeps detailed atomic positions and clearly shows their building blocks and how they connect. This helps language models better understand and predict MOF properties, making it easier to work with these materials.
What this means in practice
- •For materials engineers: Improve computational analysis of metal-organic frameworks by using MOFjson to clearly represent structure and connectivity for property prediction and design.
- •For chemical software developers: Build software that interprets MOFjson to enable accurate language model assisted modelling and automated design of crystalline materials.
Authors
Anabel Yong, Matthew Hart, Philippe Schwaller
Abstract
Language models for metal-organic frameworks (MOFs) rely on formats that separate atomic geometry from structural topology, forcing models to choose between unstructured coordinate lists and non-spatial string identifiers[1]. We present MOFjson, the first representation separating structural data into a raw atomic layer with explicit bond shift vectors and a derived layer exposing secondary building units (metal nodes and organic linkers) and network topology. Additionally, we introduce a benchmark with crystal-derived ground truth across four question categories. Across four models, MOFjson beats all three baselines in every category: 100% vs. 18-47% on reaching a block's nearest periodic copy, 96% vs. 71% on crystal system characterisation, and 60% vs. 48% on metal node and organic linker coordination. Ablating shift labels drops accuracy to 12.7%, proving models read the periodic graph rather than the surface format. MOFjson gives language models structure to read rather than reconstruct, establishing a foundation for agentic tool workflows and constrained MOF generation.