Interpretable Machine-Learning and Big Data Mining to Predict Gas Diffusivity in Metal-Organic Frameworks.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Gas diffusion significantly impacts metal-organic framework (MOF) performance in separation and catalysis.
• Accurate prediction of molecular diffusion within MOFs is crucial for process optimization.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model for predicting molecular diffusivity and selectivity in MOFs.
• To identify key factors governing diffusion selectivity in MOFs.
• To create a user-friendly application for calculating molecular diffusion in porous materials.

Main Methods:

• Training a Light Gradient Boosting Machine (LGBM) model to predict diffusivity and selectivity for nine gases.
• Utilizing Shapley Additive Explanation (SHAP) for model interpretability.
• Developing an interactive desktop application based on the trained LGBM model.

Main Results:

• The LGBM model achieved high accuracy (average R² = 0.962) and demonstrated superior extrapolation for ethane (C2H6) diffusivity.
• The ML model's calculations were five orders of magnitude faster than molecular dynamics (MD) simulations.
• Molecular polarizability difference (ΔPol) was identified as the primary factor controlling diffusion selectivity.

Conclusions:

• Interpretable ML models can accurately and rapidly predict molecular diffusion in MOFs.
• The developed application facilitates quick calculations for researchers.
• This approach aids in exploring MOF structure-property relationships and optimizing gas separation and catalysis, including CO2 methanation.