Predicting Critical Micelle Concentrations for Surfactants Using Graph Convolutional Neural Networks

• DOI: 10.1021/acs.jpcb.1c05264
• Publication Date: 2021-09-09T14:51:35Z
• Authors: Shiyi Qin; Tianyi Jin; Reid C. Van Lehn; Victor M. Zavala
• Keywords: Cationic polymerization

Surfactants are amphiphilic molecules that widely used in consumer products, industrial processes, and biological applications. A critical property of a surfactant is the micelle concentration (CMC), which at undergo cooperative self-assembly solution. Notably, primary method to obtain CMCs experimentally-tensiometry-is laborious expensive. In this study, we show graph convolutional neural networks (GCNs) can predict directly from molecular structure. particular, developed GCN architecture encodes structure form trained it using experimental CMC data. We found with higher accuracy on more inclusive data set than previously proposed methods generalize anionic, cationic, zwitterionic, nonionic surfactants single model. Molecular saliency maps revealed how atom types substructures contribute behavior be agreement physical rules correlate constitutional topological information CMCs. Following such rules, small new for not available; these molecules, predicted our exhibited similar trends those obtained simulations. These results provide evidence GCNs enable high-throughput screening desired characteristics.