Data-Driven Prediction of Janus/Core-Shell Morphology in Polymer Particles: A Machine-Learning Approach

Area of Science:

• Materials Science
• Polymer Science
• Computational Chemistry

Background:

• Traditional models for Janus particle morphology prediction rely on interfacial tension or free energy.
• Data-driven approaches offer a powerful alternative for identifying complex patterns in particle formation.

Purpose of the Study:

• To develop a predictive model for Janus and core-shell particle morphology using machine learning and explainable artificial intelligence (XAI).
• To identify key molecular and system parameters influencing particle morphology during solvent evaporation-induced phase separation.

Main Methods:

• Utilized a dataset of 200 instances to train machine-learning ensemble classifiers.
• Employed simplified molecular input line entry system (SMILES) syntax to extract features like cohesive energy density, molar volume, Flory-Huggins parameter, and solubility parameter.
• Applied XAI tools, including Shapley plots, to interpret model predictions and understand feature importance.

Main Results:

• Achieved up to 90% accuracy in predicting particle morphology (Janus vs. core-shell).
• Identified solvent solubility, polymer cohesive energy difference, and blend composition as critical factors affecting morphology.
• Determined thresholds for cohesive energy density, molar volume, and Flory-Huggins interaction parameter that favor Janus or core-shell structures.

Conclusions:

• The developed model provides accurate, data-driven predictions for particle morphology.
• XAI analysis reveals that kinetically stable morphologies can be achieved by tuning parameters to lower the driving force for phase separation.
• The findings offer novel strategies for designing Janus or core-shell particles by selecting specific feature values.