Machine-learning model reveals critical features needed for high-throughput screening of candidates for carbon-dioxide adsorption

Metal-organic frameworks (MOFs) are porous crystalline materials whose high porosity and surface area have been increasingly used for carbon dioxide (CO₂) adsorption. Because of their importance, the ability to quickly screen different MOFs for their potential efficiency is highly desirable. Teng and Shan developed a machine learning model that determined the key properties that should be used in screening MOFs.

Analyzing the effects of 23 structural and molecular features and 765 calculated features for their model, the duo ranked the importance of different features at different pressures. They found that, regardless of the pressure, molecular structure and pore size of MOFs were critical to improving the accuracy of their prediction model, up to 20 percent.

“High-throughput screening using our machine-learning model could reduce the demand for computational resources and minimize the need for extensive expert intervention,” author Guangcun Shan said. “Furthermore, the machine-learning model in this study could be used to provide theoretical support for other predicted results, so they can achieve higher prediction accuracy.”

To understand the effect of individual features on CO₂adsorption, the authors sequentially added structural, molecular, and calculated features to their model. Then, to understand why some features were ranked more highly than others, the team applied their domain knowledge on intermolecular forces, secondary bonds, and electric potential.

The team plans to continue exploring how MOFs adsorb CO₂at the atomic level using similar machine learning methods.

“By utilizing other advanced deep learning models, such as graph neural networks, we aim to further assist in screening MOFs with high adsorption capacities,” Shan said.

Source: “Interpretable machine learning for materials discovery: Predicting CO2 adsorption properties of metal-organic frameworks,” by Yukun Teng and Guangcun Shan, APL Materials (2024). The article can be accessed at https://doi.org/10.1063/5.0222154 .

This paper is part of the Emerging Leaders in Materials Science Collection, learn more here .