Leveraging Cross-Diversity Machine Learning to Unveil Metal-Organic Frameworks with Open Copper Sites for Biogas Upgrading.

Abstract

Biogas, primarily composed of methane (CH₄) and carbon dioxide (CO₂), is considered an alternative renewable energy resource. Efficient CO₂/CH₄ separation is essential for biogas upgrading to increase energy density, and in this context, metal-organic frameworks (MOFs) have demonstrated significant potential. Here, we integrate multiscale modeling with cross-diversity machine learning (ML) to unveil MOFs with open copper sites (OCS-MOFs) that exhibit exceptional CO₂/CH₄ separation performance. Our focus on diversity-adaptable ML guarantees that ML models trained in one chemical space are rigorously transferable to unseen MOFs from distinct chemical spaces, assuring their robustness in real-world applications. By leveraging a meticulously curated data set of 27592 OCS-MOFs, we develop ML models with high predictive accuracy, capable of identifying top-performing OCS-MOFs across diverse chemical environments. This work not only elucidates the reticular chemistry that governs optimal CO₂/CH₄ separation performance in OCS-MOFs but also establishes a new benchmark for scalable and resilient digital MOF discovery, with cross-diversity accuracy as the key determinant of model transferability.