Multi-Modal AI for the Design of Functional Porous Materials

Porous materials are increasingly recognized for their potential in energy and environmental applications. Among these, metal-organic frameworks (MOFs) stand out due to their tunable molecular structures and diverse topologies, offering a vast chemical space for exploration. While theoretically, an unlimited number of porous materials can be synthesized, leveraging AI enables more effective navigation of this complex space. We introduce a multi-modal Transformer pre-trained with 1 million hypothetical MOFs, that enables accurate predictions of multiple properties and enhances the transferability of these predictions across various MOFs. This model acts as a robust surrogate, identifying optimal MOF candidates tailored for diverse photocatalytic applications. Our approach significantly accelerates the discovery and deployment of functional MOFs, showcasing the potential of AI in porous materials.