Machine learning to design metal-organic frameworks: progress and challenges from a data efficiency perspective.

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Metal-organic frameworks (MOFs) offer design flexibility but present a vast search space, making traditional screening methods resource-intensive.
• The growing availability of MOF data necessitates efficient computational approaches for accelerated discovery and design.

Purpose of the Study:

• To critically review machine learning (ML) applications at the intersection of ML and MOFs.
• To survey strategies for reducing data and resource burdens in MOF property prediction and design.
• To identify future opportunities for ML-empowered MOF design.

Main Methods:

• Feature engineering
• Model architecture selection
• Transfer learning
• Active learning
• Generative models

Main Results:

• ML methods offer a promising solution to accelerate MOF discovery by addressing the large design space.
• Various strategies, including feature engineering and generative models, can enhance data and resource efficiency.
• Current ML applications primarily focus on MOF adsorption properties, with potential for broader applications.

Conclusions:

• Efficient ML strategies are crucial for robust and accessible ML-aided MOF design.
• Addressing data quality and scalability challenges will further advance ML in MOF research.
• Future work should expand ML applications beyond adsorption properties to other MOF functionalities.