Prediction of Structure and Physical Properties of Aqueous Surfactant Solutions by Combining Molecular Simulation and Machine Learning

Abstract

Surfactants, as a class of soft materials, exhibit a wide variety of self-assembled structures and physical properties due to their flexible molecular design. However, exploring their full potential remains difficult due to the complexity of their behavior in multicomponent systems. This study introduces an integrated approach that combines molecular simulations and machine learning to predict the structure and performance of aqueous surfactant formulations. Using dissipative particle dynamics, we reproduced and analyzed the self-assembly behavior of over 200 formulations of cleansing agents. While the overall morphology of aggregates showed limited correlation with cleansing performance, detailed analysis revealed that the spatial distribution of weakly hydrophobic components significantly influenced the result. Based on this insight, we constructed machine learning models using molecular descriptors to predict cleansing efficiency, achieving high accuracy with an Extra Tree Regressor. Furthermore, we performed in silico formulation screening to identify promising ingredient combinations. Additionally, a separate model was developed to predict the critical packing parameter from molecular structure using deep learning, achieving high accuracy. These results demonstrate that combining simulation and machine learning offers a powerful method for understanding and designing functional surfactant systems.

graphical abstract