Data-Driven and Machine Learning to Screen Metal-Organic Frameworks for the Efficient Separation of Methane.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Growing economies increase reliance on energy, intensifying the energy crisis.
• Methane (CH4) is a crucial clean energy source, necessitating efficient purification methods.
• Metal-organic frameworks (MOFs) show promise as adsorbents for gas separation.

Purpose of the Study:

• To investigate the adsorption and diffusion properties of numerous MOFs for methane purification from binary gas mixtures.
• To identify high-performance MOFs and establish design principles for enhanced methane separation.
• To understand the microscopic mechanisms governing methane separation in MOFs.

Main Methods:

• Large-scale computational screening of thousands of MOFs.
• Machine learning models (including Light Gradient Boosting Machine - LGBM) to predict gas diffusivity and selectivity.
• SHapley Additive exPlanation (SHAP) technique to identify critical MOF descriptors.

Main Results:

• The LGBM model achieved high accuracy (R²=0.954 for diffusivity, R²=0.931 for selectivity).
• Pore limiting diameter (PLD) was identified as the most influential structural descriptor for molecular diffusivity.
• Three high-performance MOFs were identified for each CH4 mixture system, with common design principles proposed.

Conclusions:

• Computational screening and machine learning effectively identify optimal MOFs for methane purification.
• MOF structural design, particularly pore size, is critical for efficient gas separation.
• The study provides microscopic insights and design guidelines for developing advanced MOFs for methane enrichment.