Predicting Synergistic and Antagonistic Micellization Effects in Surfactant Blends for Optimized Formulation Design

Surfactant mixtures exhibit non-ideal interactions that govern their macroscopic behavior, yet their prediction is limited by the need for experimentally derived interaction parameters. Within Rubingh's Regular Solution Approximation (RST), these interactions are described by the parameter β, which is typically obtained by fitting experimental mixture data, restricting its applicability in large-scale formulation design. In this work, a computational hybrid approach combining machine learning with thermodynamic modeling is introduced to enable end-to-end prediction and optimization of surfactant mixtures. A dataset of interaction parameters derived from experimental binary surfactant mixtures was constructed to train predictive models for β, together with models for pure-component properties. The predicted, composition-dependent β values are incorporated into the RST framework to estimate mixture behavior without requiring experimental mixture data, while treating the resulting activity coefficients as effective quantities. The overall framework was further validated against a comprehensive dataset of binary surfactant mixtures and further evaluated on ternary systems, showing good agreement with experimental values across diverse interaction regimes and compositions. In addition, a multi-objective optimization strategy is introduced to identify formulations that balance performance and molecular complexity, enabling efficient exploration of chemical space. A case study on representative systems further demonstrates the applicability of the framework for computational formulation design. Overall, the proposed approach provides a scalable computational tool for surfactant mixture design, reducing reliance on experimental parameterization while enabling rapid virtual screening of formulation space.