Abstract:  The aim of the dissertation is to study small organic solute rejection in organic solvent nanofiltration using cheminformatics and machine learning. Chemically diverse datasets were curated using a novel medium high-throughput methodology and literature data extraction. These datasets contained thousands of datapoints of small organic solute rejections and process parameters.

Different chemical fingerprinting has been used for feature generation, such as molecular descriptors, Morgan fingerprints, and latent-vector representation. These features were used to train different machine learning models, such as graph neural networks, partial least squares regression, and boosting tree algorithms. The obtained models were used to predict small organic solute rejection for nanofiltration-related applications. The correlation between rejection and molecular descriptors has been studied to deepen the understanding of transport and rejection through polymeric membranes. An explainable artificial intelligence concept was used to study the effect of solute, solvent, and membrane structure on solute rejection. The conclusion of the dissertation highlights the importance of the chemical structure effect in nanofiltration and provides a future perspective on data-driven approaches for nanofiltration and organic solvent nanofiltration.