Accelerating the design of MOFs through a Machine Learning assisted High-throughput methodology

MOFs are highly crystalline porous hybrid materials with a highly tuneable structural and chemical diversity that show high promise in a variety of applications.

While there is still an urge to discover new materials and structures, the understanding behind the synthesis to properties remains as a happenstance where rarely `MOF design on target’ strategy works. As of today, machine learning (ML) tools and big data generative approach forge ahead as the most up-to-date tool to accelerate the speed of discovery of structures. However, most predictions result in materials that cannot be synthesized and, there are only a scant number of successes that can predict relations between the structure and chemical properties, thus restricting the databases to only structural properties of experimental and hypothetical structures. To accelerate the discovery of novel materials, one shall establish and develop databases that can act as training sets for ML approaches, in a systematic manner with key performance indicators such as physical and chemical properties. MOFs Learning is a first attempts to set-up a platform for the ML-driven high-throughput (HT) synthesis and characterization of MOFs. This is to establish a methodology and databases for the application of a ML model to accelerate the property prediction and to answer the question: “How to synthesize a selected MOF for a given application”.

MOFs Learning involves three partners (CNRS-IMAP, CNRS-IRCP and CEA/DES/ISEC/ICSM). CNRS-IMAP carries out the synthesis and characterization of MOFs, CNRS-IRCP will create enriched databases in terms of physical and chemical properties and training sets to identify key descriptors, and CEA will perform HT gas sorption studies of MOFs which will be later fed to the models prepared by CNRS-IRCP.

During the first year, IMAP has made high quality MOFs sent to CEA for first gas testing studies while IRCP has begun to set-up the standardisation of the data to enrich the databases. High-throughput instruments have been ordered by IMAP (synthesis robot, powder diffractometer…) to be installed end of 2024. Next plan is to synthesis other series of MOFs, collect extended gas sorption data and define the descriptors and increase the database enrichment.