Separation of formaldehyde, N, and O in MOFs: Crystal graph convolutional neural network, machine learning, and experimental.

BACKGROUND

Formaldehyde, a hazardous indoor pollutant, poses serious health risks even at low concentrations. Metal-organic frameworks (MOFs), due to their high porosity and tunable structures, offer promising potential for selective gas adsorption and separation. However, identifying high-performance MOFs from thousands of candidates using traditional experiments is time-consuming and inefficient. This study addressed the need for efficient MOF screening strategies by integrating high-throughput atomistic simulations, machine learning, and crystal graph convolutional neural networks (CGCNN) to evaluate the adsorption and separation performance of MOFs for formaldehyde, nitrogen (N), and oxygen (O).

RESULTS

We recalculated charges for 4400 MOFs from the CoreMOF2019 database using PACMAN and performed GCMC simulations to determine Henry's constants and adsorption isotherms. Based on hydrophobicity screening, 440 MOFs were selected for detailed simulation and machine learning modeling. Gradient Boosted Regression (GBR) and CGCNN models were trained to predict adsorption behavior with R values over 0.98, significantly reducing computational costs. Breakthrough and Ideal Adsorbed Solution Theory (IAST) simulations were used to assess multicomponent separation performance, identifying LEVLEF as the top MOF with high selectivity for formaldehyde. Ideal Vacuum Swing Adsorption (IVSA) simulations further confirmed the dynamic separation behavior. Experimental validation with MIL-101 and HKUST-1 confirmed the prediction accuracy of ML models, highlighting the structural factors influencing gas uptake.

SIGNIFICANCE

This work demonstrates the power of machine learning and CGCNN in accurately predicting gas adsorption behavior in MOFs, drastically accelerating the screening process. By integrating simulation, modeling, and experimental validation, the study offers a comprehensive pipeline for the rational design and selection of MOFs for efficient indoor air purification and multi-component gas separation, with strong potential for environmental and industrial applications.