Creation of Polymer Datasets with Targeted Backbones for Screening of High-Performance Membranes for Gas Separation

Area of Science:

• Computational chemistry
• Materials science
• Polymer science

Background:

• Developing novel polymers for gas separation is crucial for industrial applications.
• High-throughput screening methods are needed to accelerate the discovery of advanced materials.

Purpose of the Study:

• To computationally construct diverse polymer datasets.
• To screen these polymers for CO2/CH4 and CO2/N2 gas separation using machine learning.

Main Methods:

• Combined simplified molecular-input line-entry system (SMILES) strings of polymer backbones and molecular fragments.
• Created 14 polymer datasets using seven polymer backbones (including PDMS, PEO, PAGE, PPZ) and molecules from MOSES and QM9 datasets.
• Utilized machine learning models to predict polymer performance in gas separation.

Main Results:

• Identified several polymers of interest for CO2/CH4 and CO2/N2 separation.
• Machine learning models trained on polymer selectivities showed higher prediction accuracy than those using permeability ratios.
• Demonstrated the utility of generated datasets for cheminformatics tasks.

Conclusions:

• The developed computational approach efficiently generates polymer datasets for materials discovery.
• Machine learning models are effective tools for screening polymers for gas separation applications.
• Predicting polymer selectivity directly improves accuracy in identifying high-performance materials.