(169dd) Cross-Database Discovery of Metal-Organic Frameworks with Open Cu(II) Sites for Biogas Upgrading through Machine Learning

This study presents a systematic approach to screen Metal-Organic Frameworks (MOFs) with open Cu(II) sites (OCS) for biogas upgrading, by integrating machine learning (ML) with a novel validation strategy. Distinguishing itself in both methodology and application scale, our work employs force fields tailored for the specific interactions between OCS and CO₂/CH₄ binary-component system, enhancing screening accuracy for biogas separation using unique OCS-MOFs meticulously sourced and curated from the Ab-initio REPEAT Charge (ARC) MOF database.

Our work further delves into the application of ML to this field. By tracing the origins of MOFs within the ARC-MOF database, which amalgamates various sub-databases, into focused subsets of OCS-MOFs, we not only broaden the scope of potential MOF candidates but also introduce a rigorous external validation step for the integration of data-science heuristics in materials discovery. This involves testing the ML model against a set of databases not used in conventional ML training/test sets, offering a stringent assessment of its predictive capability and generalizability across unseen chemical structures.

The methodological novelty of our work lies in this dual layer of innovation: the employment of OCS-specific force fields for more accurate description of interactions and the systematic external validation of ML predictions. Together, these advancements contribute significantly to the fields of materials science and chemical engineering, offering a reliable and nuanced approach to the discovery and evaluation of MOFs for environmental applications, particularly in the context of biogas upgrading.