Research Interests: As global CO₂ emissions rise, effective carbon capture technologies are critical for climate mitigation. Adsorption-based methods, particularly those using metal–organic frameworks (MOFs), offer energy-efficient solutions due to the tunable porosity and chemical diversity of MOFs. However, the presence of water vapor in real-world flue gas poses challenges for performance and material stability. In this work, we used active learning to develop a dataset and build a machine learning model trained on simulated adsorption data for ~10,000 MOFs to predict the uptake of CO₂, N₂, and H₂O at 80% relative humidity and 15% CO₂ by mole fraction in the gas stream. By incorporating energy-based descriptors—quantifying Henry's coefficients, host–adsorbate and adsorbate–adsorbate interactions—we substantially improved the model's predictive accuracy.
We integrated the model into a high-throughput screening workflow and applied it to the ARC-MOF database, which contains more than 240,000 hypothetical and experimental structures, to identify top candidates and shared structural properties. Guided by these results, we designed new MOFs and summarized design rules that enable more efficient CO₂ capture under humid conditions. Experimental efforts to validate these findings are currently underway, offering a data-driven pathway to accelerate the discovery and deployment of next-generation adsorbents for industrial carbon capture applications.
