(169bt) Combining Forward and Inverse Design of Covalent-Organic Frameworks for Methane Storage Via Data-Driven Discovery

This study introduces a new strategy for the forward and inverse design of Covalent-Organic Frameworks (COFs) aimed for optimal methane (CH₄) storage. By computationally synthesizing CH₄ deliverable capacities in COF databases, our approach distinguishes itself through a dual featurization model that captures both framework and pore chemistries. Sparse chemical features, such as the presence of specific bond types, are adeptly managed through factorization machines, while continuous variables like density and pore sizes benefit from deep learning methodologies. This fusion significantly advances prediction accuracy for COFs with high CH₄ storage potential.

Inverse design poses a unique challenge for COFs, given their varied connection types and similar chemical identities of secondary building blocks (SBUs). To address this, we employed a Crystal Diffusion Variational Autoencoder (CDVAE), adopting a Structure is All You Need philosophy. This approach ensures that every component, including SBUs and topologies, remains flexible while navigating the chemical space, relying entirely on the encoded space. With a focused training regime on top-performing COFs guided by property optimizations, we aim to facilitate the efficient design of high-performing COFs for CH₄ storage.

Our work transcends traditional boundaries in COF research, offering synergistic forward and inverse design to fast-track the discovery of new COF for high CH₄ storage. We believe this work shall lay a solid groundwork for future advancement in gas storage.