Year: 2026
Type: preprint
Abstract: Porous materials, such as metal-organic frameworks (MOFs), covalent organic frameworks (COFs), and zeolites, hold great promise for energy and environmental applications, particularly in gas storage and separation technologies. However, their rapidly expanding design space, driven by experimental synthesis and computational modeling, challenges traditional exploration methods. Machine learning has emerged as a promising alternative, yet current approaches are typically developed for specific adsorption properties or material families. Consequently, each new property or material class requires training a new model from scratch using large amounts of labeled data, which are often not available. Inspired by the emerging foundation-model paradigm that has transformed computer vision and natural language processing, we introduce IntelliPore, a foundation model tailored for gas adsorption in porous materials. By pretraining on 12.5 million publicly available adsorption data from MOFs and COFs, IntelliPore learns transferable knowledge across adsorption tasks and material classes, enabling generalization to previously unseen materials such as zeolites and porous polymer networks. It demonstrates strong few-shot learning capabilities, substantially improving sample efficiency compared with the traditional approach of training from scratch. Furthermore, it achieves state-of-the-art performance in adsorption property prediction with exceptional parameter efficiency, requiring only 0.86 million parameters—two orders of magnitude fewer than existing leading models. To facilitate broader adoption, IntelliPore is made publicly available to the research community and accessible through a web application. Collectively, these advances establish IntelliPore as a practical foundation for data-driven adsorption modeling in porous materials and represent a step toward the emerging paradigm of foundation-style models in materials science.
Source: ChemRxiv
Authors: Antonios P. Sarikas, George E. Froudakis
Institution: University of Crete
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7212252546