Machine Learning Analysis of Liposome Stability in Polyol-Tween 85 Surfactant Systems.

Abstract

Introduction: Liposomes, bilayered vesicles capable of encapsulating both hydrophilic and hydrophobic compounds, are widely utilized in cosmetic formulations owing to their superior biocompatibility. However, their structural integrity can be compromised by surfactants and polyols, which are essential components in the formulation process. Therefore, it is imperative to evaluate how the concentrations of these additives influence the stability and release behavior of liposomes.

Method: To analyze the impact of Tween 85 and five different polyols on liposomal release, second-order multiple linear regression models were developed. Nonlinear interactions were visualized using 3D regression surfaces. Furthermore, K-Nearest Neighbors (KNN), logistic Regression (LR), and Stochastic Gradient Descent (SGD) algorithms were implemented to classify liposomal stability. To enhance the generalization performance of the models, Ridge and LASSO regularization techniques were incorporated.

Result: All regression models achieved high predictive accuracy. Notably, the overfitting issues observed in formulations containing 1,2-hexanediol (1,2-HD) and 1,2-octanediol (1,2-OD) were effectively mitigated through the application of Ridge and LASSO models. Among the classification models, the SGD classifier demonstrated the highest accuracy, followed by logistic regression and KNN.

Discussion: The results indicated that the concentration of the surfactant (Tween 85) had a more dominant effect on liposomal release behavior than that of polyols. This is consistent with previous findings that surfactants promote the disruption of the membrane structure by increasing the fluidity of the lipid bilayer. The developed models accurately capture the nonlinear interactions between additives, providing a practical predictive tool for formulation optimization.

Conclusion: This study systematically classified and predicted liposomal stability under complex conditions involving various concentrations of polyols and surfactants using machine learning algorithms. The developed models are expected to facilitate efficient formulation design in liposome-based systems, including cosmetic and drug delivery applications, by minimizing trial and error.