An Effective Screening Method of CO ₂ Adsorption and Selectivity for Metal–Organic Frameworks by Machine Learning and High-Throughput Computation

Due to the large specific surface area, adjustable pore structure, diverse active sites, and flexible structural design, MOFs have significant advantages in gas adsorption and electrocatalysis, but the low utilization rate of catalytic sites and the poor reactant/product selectivity also hinder the completed application in industry. Therefore, based on the relationship between the structure properties (void fraction, gravimetric surface area, largest cavity diameter, pore limiting diameter, density and metal percentage) and the potential catalysis features (reactant adsorption, activation and reactant/product selectivity) in MOFs, the corresponding screening method by combining machine learning and high-throughput computing proposed in the present work will be helpful to put forward a rule of potential catalysis for the capture and conversion of CO2 as well as the release of products CH4. Determined from the experimental MOFs database ToBaCCo, 6200 MOFs were randomly selected to perform the machine learning training (ML) of the adsorption of CO2 in single adsorption NCO2, trade-off between adsorption and selectivity (TSN) in CO2/CH4 as well as the adsorption enthalpy of CO2 and CH4 in binary adsorption (HCO2 and HCH4)). The results revealed that the density (P) and gravimetric surface area (GSA) played an important role in the adsorption performance of MOFs for capturing CO2, separation, and conversion. After screening the four widely used databases (coremof2019, coremof2014, hmofs, ToBaCCo), the optimal MOFs due to the suitable P (0.5 g/cm3 > P > 0.8 g/cm3) and GSA (5500 m2/g > GSA > 1000 m2/g) generated the preferable CO2 adsorption capacity and selectivity.