Project: Atom-Level Generative Modeling of Metal-Organic Frameworks with Graph Transformers
Description
Metal-organic frameworks (MOFs) are crystalline, porous materials with modular architectures and vast structural diversity, making them ideal candidates for data-driven materials discovery. In recent years, generative machine learning models have been developed to explore the MOF design space by assembling frameworks from pre-defined building blocks—typically metal clusters and organic linkers. Notable examples include MOFFlow [1], a flow-based model for MOF structure prediction, and MOFDiff [2], a diffusion model trained to generate MOFs from a latent space. While these methods have enabled scalable and controllable MOF generation, they are constrained by the building block formalism, which limits the discovery of novel topologies and atomic arrangements.
Recent developments in geometric deep learning—particularly transformer-based architectures such as Erwin [3] or ADiT [4]—offer new possibilities for materials generation. These models are capable of handling large, complex graphs with long-range dependencies and hierarchies, making them suitable for modeling extended periodic systems like MOFs at the atom level. Atom-level generation could allow for more flexible and fine-grained exploration of the MOF design space, capturing chemical detail and discovering frameworks beyond known topologies and motifs.
This project investigates the potential of atom-level MOF generation using graph transformers. Instead of relying on predefined components, the model will learn directly from atomic graphs, leveraging the capacity of attention mechanisms to capture both local coordination and global structural features. By pushing beyond the building block paradigm, the project aims to contribute toward more general and expressive generative models for materials design.
[1] Kim, Nayoung, et al. "MOFFlow: Flow Matching for Structure Prediction of Metal-Organic Frameworks." arXiv preprint arXiv:2410.17270 (2024).
[2] Fu, Xiang, et al. "Mofdiff: Coarse-grained diffusion for metal-organic framework design." arXiv preprint arXiv:2310.10732 (2023).
[3] Zhdanov, Maksim, Max Welling, and Jan-Willem van de Meent. "Erwin: A Tree-based Hierarchical Transformer for Large-scale Physical Systems." arXiv preprint arXiv:2502.17019 (2025).
[4] Joshi, Chaitanya K., et al. "All-atom diffusion transformers: Unified generative modelling of molecules and materials." arXiv preprint arXiv:2503.03965 (2025).
Details
- Supervisor
-
Vlado Menkovski
- Secondary supervisor
-
Marko Petkovic
- Interested?
- Get in contact