Fast and Accurate Machine Learning Strategy for Calculating Partial Atomic Charges in Metal-Organic Frameworks

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• High-throughput screening of metal-organic frameworks (MOFs) is crucial for gas storage and separation.
• Accurate partial atomic charges are essential for modeling adsorbate-MOF interactions, especially for polar molecules like CO2 and water.
• Ab initio methods for calculating charges are computationally expensive for large-scale screening.

Purpose of the Study:

• To develop a computationally efficient machine learning model for predicting partial atomic charges in MOFs.
• To enable faster screening of MOFs for gas storage and separation applications.

Main Methods:

• A random forest machine learning model was developed.
• The model utilizes features representing elemental properties and local atomic environments.
• Trained and tested on density-derived electrostatic and chemical (DDEC) and charge model 5 (CM5) atomic charges from the CoRE MOF-2019 database.

Main Results:

• The machine learning model accurately predicts partial atomic charges in MOFs.
• The model significantly reduces computational cost compared to periodic density functional theory (DFT).
• The model demonstrates transferability to other porous materials like molecular crystals and zeolites.

Conclusions:

• Machine learning offers a computationally efficient alternative for calculating partial atomic charges in MOFs.
• This approach accelerates the identification of promising MOFs for gas storage and separation.
• Partial atomic charge correlates strongly with the average electronegativity difference of bonded atoms.