MOFFlow: Flow Matching for Structure Prediction of Metal-Organic Frameworks

Nayoung Kim,Seongsu Kim,Minsu Kim,Jinkyoo Park,Sungsoo Ahn

2024-10-07

Abstract:Metal-organic frameworks (MOFs) are a class of crystalline materials with promising applications in many areas such as carbon capture and drug delivery. In this work, we introduce MOFFlow, the first deep generative model tailored for MOF structure prediction. Existing approaches, including ab initio calculations and even deep generative models, struggle with the complexity of MOF structures due to the large number of atoms in the unit cells. To address this limitation, we propose a novel Riemannian flow matching framework that reduces the dimensionality of the problem by treating the metal nodes and organic linkers as rigid bodies, capitalizing on the inherent modularity of MOFs. By operating in the $SE(3)$ space, MOFFlow effectively captures the roto-translational dynamics of these rigid components in a scalable way. Our experiment demonstrates that MOFFlow accurately predicts MOF structures containing several hundred atoms, significantly outperforming conventional methods and state-of-the-art machine learning baselines while being much faster.

Biomolecules,Materials Science,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the complexity and computational cost in the structural prediction of metal - organic frameworks (MOFs). Specifically:

1. **Complexity problem**: Existing methods, including ab initio calculations and deep generative models, face challenges when dealing with MOF structures because the unit cells of MOFs contain a large number of atoms, which makes structural prediction very complex.

2. **Computational cost problem**: Traditional crystal structure prediction methods, such as those based on density functional theory (DFT), can provide accurate results but are very computationally expensive, especially when dealing with large and complex systems such as MOFs.

To solve these problems, the paper proposes a new deep generative model - MOFF LOW. This model solves the above problems in the following ways:

- **Dimensionality reduction**: MOFF LOW proposes a new Riemannian flow matching framework, which regards metal nodes and organic linkers as rigid bodies and utilizes the modular characteristics of MOFs, thereby reducing the dimensionality of the problem.
- **Efficient prediction**: By operating in SE(3) space, MOFF LOW can effectively capture the rotational and translational dynamics of these rigid components and achieve scalable prediction.
- **High precision**: Experimental results show that MOFF LOW can accurately predict MOF structures containing hundreds of atoms, significantly outperforming traditional methods and the state - of - the - art machine - learning baselines, and is also faster.

In conclusion, the goal of MOFF LOW is to provide an efficient and accurate method for MOF structure prediction to overcome the limitations of existing methods in dealing with large and complex MOFs.