Prediction of Gas Permeability in Polymeric Membranes for Blood Oxygenators using Machine Learning Models
De Mitri, Aline G.
Marmey, Emma
Ferreira, Tiago J.
Vinga, Susana
de Pinho, Maria N.
Faria, Mónica
Pdf

How to Cite

De Mitri A.G., Marmey E., Ferreira T.J., Vinga S., de Pinho M.N., Faria M., 2026, Prediction of Gas Permeability in Polymeric Membranes for Blood Oxygenators using Machine Learning Models, Chemical Engineering Transactions, 128, 127-132.
Pdf

Abstract

Extracorporeal membrane oxygenation (ECMO) supports patients with impaired respiratory function. In these devices, efficient gas exchange depends strongly on the permeability of polymeric membranes to oxygen (O2) and carbon dioxide (CO2). However, the experimental characterization of gas permeability is time-consuming and dependent on the trial-and-error approach. In this context, machine learning (ML) has emerged as a promising strategy to accelerate material discovery by enabling data-driven prediction of transport properties. This work aims to develop and evaluate ML models for predicting O2 and CO2 permeability in polymeric membranes from a set of gas permeability data. Several regression models were implemented and compared: Bayesian Linear Regression (BLR), Generalized Linear Models (GLM), Extremely Randomized Trees (ERT), and Extreme Gradient Boosting (XGBoost). Model performance was assessed using R2 correlation and the root mean square error (RMSE) between predicted and experimental values. ERT provided the most accurate predictions for O2 permeability, while XGBoost achieved superior performance for CO2 permeability. The ML-based screening identified substituted polyacetylenes with high gas transport potential, consistent with known structure–property relationships. The results demonstrate that ML can serve as an effective decision-support tool to significantly accelerate membrane development while reducing experimental costs.
Pdf