MOFology: A Knowledge Graph for Engineering Direct Air Capture Materials

Matthew Hart, Jeremy Goumaz, Philippe Schwaller

Track: Track 1: Original Research/Position/Education/Attention Track

Keywords: Metal-Organic Frameworks, Direct Air Capture, Knowledge Graph, Graph Embeddings, Amine Functionalization, Concept Vectors, Ontology, CO2 Capture, Materials Discovery

TL;DR: MOFology is a MOF knowledge graph with 8.4M RDF triples used to demonstrate that learned graph embeddings can predict properties and suggest direct air capture candidates.

Abstract: Direct Air Capture (DAC) of CO₂ requires sorbents that are simultaneously CO₂-selective, water-tolerant, thermally regenerable, and synthesizable, with Metal-Organic Frameworks (MOFs) as the most promising candidates. Yet the data describing MOFs is scattered across multiple databases in disparate formats, leaving cross-domain opportunities in DAC materials engineering unrealized. We present MOFology, an ontology-grounded knowledge graph (KG) that integrates 250,000 MOFs from six databases into 8.4 million RDF triples. The KG supports a suite of DAC-relevant tasks: chemically interpretable SPARQL semantic queries, graph embedding, concept vectors recovered as linear directions encoding chemical properties, prediction of change in CO₂ binding energy upon amine functionalization, and a multi-criteria DAC screen ranking over 9,000 real MOFs. To our knowledge, MOFology is the largest MOF KG to date and the first work to featurize the parent–derivative functionalization relation for predicting property changes, enabling a comprehensive database and prediction framework for DAC materials engineering.

Submission Number: 127