Machine learning for the advancement of membrane science and technology: A critical review
Gergő Ignácz, Lana Bader, Áron Kristóf Beke, Yasir Ghunaim, Tejus Shastry, Hakkim Vovusha, Matthew R. Carbone, Bernard Ghanem, György Székely · Journal of Membrane Science · 2024
Machine learning (ML) has been rapidly transforming the landscape of natural sciences and has the potential to revolutionize the process of data analysis and hypothesis formulation as well as expand scientific knowledge. ML has been particularly instrumental in the advancement of cheminformatics and materials science, including membrane technology. In this review, we analyze the current state-of-the-art membrane-related ML applications from ML and membrane perspectives. We first discuss the ML foundations of different algorithms and design choices. Then, traditional and deep learning methods, including application examples from the membrane literature, are reported. We also discuss the importance of learning data and both molecular and membrane-system featurization. Moreover, we follow up on the discussion with examples of ML applications in membrane science and technology. We detail the literature using data-driven methods from property prediction to membrane fabrication. Various fields are also discussed, such as reverse osmosis, gas separation, and nanofiltration. We also differentiate between downstream predictive tasks and generative membrane design. Additionally, we formulate best practices and the minimum requirements for reporting reproducible ML studies in the field of membranes. This is the first systematic and comprehensive review of ML in membrane science.