Machine Learning-Assisted Design and Chemical Screening of Metal–Organic Frameworks for Enhanced Carbon Dioxide Capture and Electrochemical Energy Storage Applications

Abstract

Because of the rising concentration of CO₂ in the atmosphere and the increasing demand for energy technologies that can improve energy storage efficiency, there has been a renewed interest in developing advanced functional materials that can help address some of the world's energy and environmental challenges. Metal-organic frameworks (MOFs), as a class of newly developed crystalline porous coordination polymers (made up of metal ions/clusters covalently bonded together using organic ligands), are a promising candidate to potentially provide solutions to these challenges due to their high specific surface area, tunable pore size, chemical diversity, and structural diversity. Coordination chemistry, metal-node characteristics, and linker functionalities (i.e., coordination chemistry and qualities of the metal nodes and linker groups) will play a large role in influencing the MOFs' adsorptive ability, electrochemical performance, and stability. Traditional approaches to finding and optimizing new MOFs require large amounts of time, money, and labor. To streamline this process, a computational chemistry framework was developed to identify and assess new high performance MOFs for application to CO2 capture and electrochemical energy storage. The framework utilized a database containing 2,500 unique MOF structures curated from well-established databases and published literature. The structural descriptors of these MOFs (physical, chemical, and mechanical property values, such as surface area, pore volume, framework density, metal electronegativity, metal coordination, thermal stability, electrical conductivity, and ion diffusion) were used to aid model training. Machine learning models were created using common algorithms (Random Forest, Support Vector Machine, Gradient Boosting, and Artificial Neural Networks) to predict the maximum amount of CO2 that can be adsorbed and/or the maximum specific capacitance of a MOF per gram of MOF material. Random Forest was the best algorithm, with an R2 value >0.90 for both target values. The top five important determining factors in predicting the MOFs' adsorptive and capacitive properties were shown to be surface area, pore volume, electronegativity of the metal nodes, functionality of the linkers, and stability of the framework. A chemical explanation of MOF adsorptive and electrochemical properties was determined using Lewis acid-base interactions between the open metal sites of the MOFs and CO2 molecules, along with linker functionalization (i.e. adding specific functional groups to the organic linkers). The addition of redox-active transition metal ions and conjugated organic linkers allowed for increasing charge storage capacity and use of electroactive materials for conducting electrons. Through the use of AI-based virtual screening methods, it was possible to identify hypothetical MOFs with predicted CO2 capture capacities >9 mmol/g and specific capacitances approaching 500 F/g. The successful combination of coordination chemistry, computational chemistry techniques, and AI is shown to be an efficient means of in vitro design of next generation MOFs. This work will be utilized to develop advanced porous materials for sustainable technologies that capture CO2 and produce high-performance energy storage devices.