A Self-Evolutionary Methodology for Reverse Design of Novel MOFs

A combination of high-throughput molecular simulation and machine learning (ML) algorithms has been widely adopted to seek promising metal-organic frameworks (MOFs) as energy gas carriers. However, the currently reported studies are mainly limited to extracting top performers from existing databases, not fully unleashing the ML capabilities for intelligently predicting novel structures with better performance. Herein, an efficient self-evolutionary methodology was proposed for searching high-performance MOFs that are unstructured in the origin database, in which a Tangent Adaptive Genetic Algorithm (TAGA) was newly put forward for structural evolution and the high-precision ML model of eXtreme Gradient Boosting (XGBoost) was employed as the fitness function. By taking CH₄ storage in MOFs at room temperature as a showcase and using the database of 51,163 hMOFs, the TAGA-XGBoost coupling strategy rapidly suggested a certain number of possible combinations of the building blocks to form new structures with gravimetric storage capacity (35 bar) and volumetric working capacity (65-5.8 bar) higher than the best materials in the original database. The structures of some promising MOFs successfully used the finally optimized material genes for the two application conditions, and their performances were also confirmed by subsequent molecular simulations. The best materials can respectively reach a storage amount of 580 cm³(STP)/g at 35 bar and a working capacity of 218 cm³(STP)/cm³ between 65 and 5.8 bar. An analysis of the top 100 materials predicted from our method revealed that the choice of organic linkers has a systematic effect on the storage performance of MOFs. It might be believed that the proposed methodology offers an opportunity to expedite the discovery of unprecedented materials for other practical applications.