(567c) Design of Metal Organic Frameworks (MOFs) with Enhanced CO2 Adsorption Using Machine Learning

The rapidly increasing concentration of CO₂ in the atmosphere has resulted in a serious greenhouse effect. Porous materials like metal organic frameworks (MOFs) have shown potential for the carbon capture due to their high surface areas, tunable porosities and functionalities. By using trial-and-error experimental and brute-force computational methods, it is challenging to explore a large design space offered by MOFs to design a MOF with higher CO₂ uptake. Therefore, to accelerate the MOF design process, we have developed a novel computational framework that integrates our in-house MOF structure generation code with machine-learning (ML), and optimization algorithms. Initially, hypothetical structures of MOFs (HMOFs) with multiple functional groups were generated by using in-house structure generation code. These ~50,000 HMOFs were screened for CO₂ adsorption by performing grand canonical monte carlo (GCMC) simulations at 1 atm and 300 K. The primary structural features of HMOFs and CO₂ adsorption were used as input and output for the supervised ML model, respectively. These trained models were integrated with our in-house MOF generation code and genetic algorithms to discover new HMOFs with enhanced CO2 adsorption. The structures discovered by this ML-based framework were further validated by performing GCMC simulations on selected top ~1000 structures.