The Rise of Generative AI for Metal-Organic Framework Design and Synthesis

Chenru Duan, Aditya Nandy, Shyam Chand Pal, Xin Yang, Wenhao Gao, Yuanqi Du, Hendrik Kraß, Yeonghun Kang, Varinia Bernales, Zuyang Ye, Tristan Pyle, Ray Yang, Zeqi Gu, Philippe Schwaller, Shengqian Ma, Shijing Sun, Alán Aspuru-Guzik, Seyed Mohamad Moosavi, Robert Wexler, Zhiling ZhengarXiv 2025 · DOI · arXiv

Introduction

Metal-Organic Frameworks (MOFs) represent a class of highly customizable, porous crystalline materials. By linking inorganic metal nodes with organic linkers, researchers can create "molecular sponges" with applications ranging from carbon capture to water harvesting. However, the near-infinite combinatorial space of these building blocks makes traditional discovery methods—relying on brute-force computational enumeration—increasingly inefficient. This paper explores the paradigm shift toward Generative AI, where models autonomously "dream up" and optimize novel, high-performance materials.

The Challenge of MOF Discovery

The primary difficulty in MOF research lies in navigating an intractable chemical space. Historically, researchers used enumerative design, stitching together known components to create databases of hypothetical structures. While useful, this approach is limited to existing templates and cannot easily explore truly de novo structures.

An example of an organic linker, a fundamental building block of MOFs.

Generative AI Methodologies

Recent advancements have adapted several AI architectures to handle the unique requirements of MOFs, such as periodic structures and large atom counts:

  • Variational Autoencoders (VAEs): Models like SmVAE encode MOF building blocks into a continuous latent space, allowing for the optimization of specific properties like CO2 uptake before decoding them back into 3D structures.
  • 3D Diffusion Models: Systems like MOFDiff use equivariant networks to assemble building blocks under realistic spatial constraints, while others generate 3D pore structures directly.
  • LLM Agents: Orchestrators like ChatMOF use Large Language Models to translate natural language prompts into executable discovery pipelines, including code generation, database querying, and simulation.

Data and Experimental Validation

The success of these models depends on high-quality datasets, including the Cambridge Structural Database (CSD) and hypothetical sets like hMOF. However, a significant "data gap" exists; many computation-ready datasets contain flawed metal oxidation states, which can negatively impact training.

Despite these challenges, the field has seen remarkable successes. The GHP-MOFassemble pipeline recently generated over 120,000 hypothetical MOFs optimized for carbon capture. More importantly, AI-generated candidates have been successfully synthesized in the lab, confirming the real-world viability of these computational predictions.

Future Directions: Toward Self-Driving Labs

The bottleneck in materials science is shifting from design to experimental validation. While AI can generate millions of candidates in hours, physical synthesis still takes months. Future work focuses on "self-driving labs"—closed-loop systems where AI agents not only design the MOF but also plan the synthesis and control robotic hardware to accelerate the discovery cycle.

Summary updated

Abstract

Advances in generative artificial intelligence are transforming how metal-organic frameworks (MOFs) are designed and discovered. This Perspective introduces the shift from laborious enumeration of MOF candidates to generative approaches that can autonomously propose and synthesize in the laboratory new porous reticular structures on demand. We outline the progress of employing deep learning models, such as variational autoencoders, diffusion models, and large language model-based agents, that are fueled by the growing amount of available data from the MOF community and suggest novel crystalline materials designs. These generative tools can be combined with high-throughput computational screening and even automated experiments to form accelerated, closed-loop discovery pipelines. The result is a new paradigm for reticular chemistry in which AI algorithms more efficiently direct the search for high-performance MOF materials for clean air and energy applications. Finally, we highlight remaining challenges such as synthetic feasibility, dataset diversity, and the need for further integration of domain knowledge.
BibTeX
@misc{duan2025rise,
  author = {Chenru Duan and Aditya Nandy and Shyam Chand Pal and Xin Yang and Wenhao Gao and Yuanqi Du and Hendrik Kraß and Yeonghun Kang and Varinia Bernales and Zuyang Ye and Tristan Pyle and Ray Yang and Zeqi Gu and Philippe Schwaller and Shengqian Ma and Shijing Sun and Alán Aspuru-Guzik and Seyed Mohamad Moosavi and Robert Wexler and Zhiling Zheng},
  title = {{The Rise of Generative AI for Metal-Organic Framework Design and Synthesis}},
  year = {2025},
  eprint = {2508.13197},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.13197},
}
Figure