Machine Learning-Based Prediction of Polymer Chemical Resistance to Organic Solvents

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Materials Science

Background:

• Predicting polymer chemical resistance to organic solvents is a significant challenge impacting sustainable materials design.
• Existing solubility models often fall short in accurately predicting resistance outcomes.
• Understanding these interactions is vital for industrial applications and material longevity.

Purpose of the Study:

• To develop an interpretable, data-driven framework for predicting polymer chemical resistance.
• To extend predictive capabilities beyond traditional solubility models.
• To facilitate exploratory screening and hypothesis generation for novel polymer development.

Main Methods:

• Systematic analysis of a large dataset (2231 combinations) of polymer-solvent interactions.
• Utilized gradient boosting models trained on molecular dynamics (MD)-derived descriptors, force-field kernel mean descriptors, and COSMO-RS parameters.
• Employed polymer-level and solvent-cluster-level validation strategies.

Main Results:

• Observed associations between polymer crystallinity, density, solvent polarity, and resistance.
• Achieved high predictive performance with ROC-AUC values of 0.85 (in-dataset) and 0.91 (validation).
• Demonstrated the framework's effectiveness in predicting chemical resistance outcomes.

Conclusions:

• The developed data-driven framework offers a robust approach to predicting polymer chemical resistance.
• Findings support the role of polymer crystallinity, density, and solvent polarity in resistance.
• This work provides a foundation for accelerated materials discovery and design in polymer science.