About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Graph Neural Network-Driven Multi-Objective Bayesian Optimization for Discovering Metal-Organic Frameworks with Optimal Separation Performance |
| Author(s) |
Lane Enrique Schultz, Nickolas Gantzler, Nathaniel Scott Bobbitt, Remi Dingreville, Dorina F. Sava Gallis |
| On-Site Speaker (Planned) |
Lane Enrique Schultz |
| Abstract Scope |
Metal-organic frameworks are porous crystalline materials used in gas capture, drug delivery, and separation. While high-throughput computational screening has traditionally identified suitable metalorganic frameworks for these applications, recent efforts focus on more efficient machine learning methods. Optimization techniques like Bayesian optimization and genetic algorithms have been used, but many overlook the critical structure-property relationships of metal-organic frameworks. This work presents a new framework that integrates graph neural networks with Bayesian optimization to enhance metal-organic framework discovery. By representing metal-organic frameworks as graphs, graph neural networks capture atomic-level properties and structural information, leading to more accurate predictions than those afforded by traditional methods. Coupled with multi-objective Bayesian optimization, this framework identifies Pareto-optimal metal-organic framework candidates, targeting improved separation properties. Our approach effectively finds materials for separating alkanes, alkenes, alcohols, and aromatics. SNL is managed by NTESS under DOE NNSA contract DE-NA0003525. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Machine Learning, |