Machine Learning–Driven Design of Sustainable Polymer Membranes: Integrated Prediction of Gas Permeability, Selectivity, and Biodegradability

Increasing CO₂ emissions are causing environmental pollution worldwide, and there is a growing need for the development of efficient CO₂ separation and recovery technologies. Many polymer materials, including polymer membranes, are difficult to decompose once released into the environment, leading to environmental pollution and adverse effects on ecosystems, and accordingly, biodegradable materials are required. In this study, we focused on the gas permeability and biodegradability of polymer materials and developed machine learning models to explore highly selective CO₂ separation polymer membrane materials and predict their biodegradability. For gas permeability prediction, we used a data set of polymer membranes with standardized synthesis and film-forming conditions, confirming that highly accurate gas permeability coefficient predictions were possible even with a small number of polymer samples. Furthermore, we predicted the gas permeability coefficients and gas selectivity of 3219 polymer material candidates and identified approximately 500 high-performance polymers that exceed the Robeson upper bounds. Then, the polymers predicted to have biodegradability were also included. The present machine-learning framework enables us to propose computational candidates for CO₂-separation membranes that are predicted to exhibit both high gas separation performance and biodegradability within the studied chemical space, providing hypothesis-generating guidance for future experimental studies.