A generative artificial intelligence framework based on a molecular diffusion model for the design of metal-organic frameworks for carbon capture

Hyun Park,Xiaoli Yan,Ruijie Zhu,E. A. Huerta,Santanu Chaudhuri,Donny Cooper,Ian Foster,Emad Tajkhorshid

DOI: https://doi.org/10.1038/s42004-023-01090-2

2024-03-13

Abstract:Metal-organic frameworks (MOFs) exhibit great promise for CO2 capture. However, finding the best performing materials poses computational and experimental grand challenges in view of the vast chemical space of potential building blocks. Here, we introduce GHP-MOFassemble, a generative artificial intelligence (AI), high performance framework for the rational and accelerated design of MOFs with high CO2 adsorption capacity and synthesizable linkers. GHP-MOFassemble generates novel linkers, assembled with one of three pre-selected metal nodes (Cu paddlewheel, Zn paddlewheel, Zn tetramer) into MOFs in a primitive cubic topology. GHP-MOFassemble screens and validates AI-generated MOFs for uniqueness, synthesizability, structural validity, uses molecular dynamics simulations to study their stability and chemical consistency, and crystal graph neural networks and Grand Canonical Monte Carlo simulations to quantify their CO2 adsorption capacities. We present the top six AI-generated MOFs with CO2 capacities greater than 2 $m mol/g$, i.e., higher than 96.9% of structures in the hypothetical MOF dataset.

Materials Science,Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in carbon capture applications, how to design metal - organic frameworks (MOFs) with high CO₂ adsorption capacity and synthesizability. Specifically, the research aims to accelerate the discovery of new MOFs with specific topological structures and nodes through a generative artificial intelligence (AI) framework, thereby overcoming the computational and experimental challenges faced by traditional experimental methods in screening the best materials in the vast chemical space.

### Problem Background

Metal - organic frameworks (MOFs) have received extensive attention due to their diverse industrial application prospects, including gas adsorption and storage, catalysis, and drug delivery. However, although MOFs exhibit excellent CO₂ adsorption performance in an appropriate environment and can be recycled multiple times, their actual industrial applications still face some challenges, such as poor long - term recyclability and sensitivity to moisture. In addition, due to the diversity of the building blocks (nodes and linkers) of MOFs and their spatial arrangement, a vast chemical space is created, making it difficult for traditional experimental methods to conduct a comprehensive search.

### Research Objectives

To solve the above problems, the researchers developed a generative AI framework named GHP - MOFassemble. The main objectives of this framework are:

1. **Generate new linkers**: Generate new MOF linkers through the Diffusion model (DiffLinker), which can be assembled with pre - selected metal nodes into MOFs with specific topological structures (such as the primitive cubic topology).
2. **Screening and verification**: Screen the generated MOFs to ensure their uniqueness, synthesizability, and structural validity, and study their stability and chemical consistency through molecular dynamics simulations.
3. **Quantify CO₂ adsorption capacity**: Use crystal graph neural networks (CGCNN) and grand canonical ensemble Monte Carlo (GCMC) simulations to quantify the CO₂ adsorption capacity of the generated MOFs.

### Main Contributions

- **Efficient generation and screening**: GHP - MOFassemble can generate a large number of candidate MOFs in a short time and quickly identify MOFs with high CO₂ adsorption capacity through a multi - level screening process.
- **Innovative method**: Combine multiple technical means such as generative AI, molecular dynamics simulations, and machine learning to propose a novel high - throughput computational framework.
- **Experimental verification**: Through detailed MD and GCMC simulations of the generated MOFs, verify their stability and adsorption performance, and finally discover 6 AI - generated MOFs whose CO₂ adsorption capacity exceeds 2 m mol g⁻¹, which is better than 96.9% of the structures in the hypothetical MOF dataset.

Through these efforts, the researchers hope to accelerate the discovery of high - performance MOFs suitable for carbon capture and promote their further development in industrial applications.