The design and functionalization of materials is a cornerstone of synthetic chemistry. Since its discovery, a quarter of a century ago, reticular synthesis — linking molecular building units by strong bonds into 2D and 3D extended crystalline porous metal-organic frameworks (MOFs) — has been added to the repertoire of synthetic chemists. Reticular chemistry operates in a vast space of composition, structure, property, and application. Although this field has flourished in the past two decades with unprecedented structural variety, we have only exploited a tiny fraction of the infinite chemical space. With an increasing amount of experimental and theoretical MOF structural databases, the design of MOFs using machine learning (ML) to identify target MOF structures in silico has become a driver in speeding up the MOF discovery cycle. Therefore, the bottle neck in MOF discovery shifted towards their experimental realization, which to date is still largely based on labour-intensive trial-and-error experiments and heuristics.
Angewandte Chemie 2022, DOI: 10.1002/anie.202200242
we set up a complete machine learning (ML) workflow to realize an inverse synthesis design of MOFs, i.e. the automated prediction of suitable synthesis conditions for a targeted arbitrary MOF structure (e.g. designed in silico). Our ML workflow for the inverse synthesis design of MOFs consists of three steps (see Figure): (1) automated data mining from literature; (2) training ML models; and (3) ML prediction and evaluation. In addition, the established SynMOF database will boost the natural language processing research within the MOF community and help them to significantly speed up the discovery cycle.
Importantly, the establish MOF Synthesis Prediction Tool can be use free of any charge:
in order to find fast the synthesis conditions for a novel MOF structure.
We hope to bring the attention of the reticular research community and beyond the importance of using artificial intelligence for gaining speed and sophistication in reticular chemistry and ultimately transforming the largely empirical science to a data-driven science.