Accelerating Polyester Intelligence: Machine-Learning-Assisted Prediction of Glass Transition Temperature and Virtual Molecules Screening | JoVE Visualize

Area of Science:

Materials Science
Computational Chemistry
Polymer Science

Background:

Economic and societal development necessitates polyesters with tailored performance characteristics.
Current polyester material innovation relies heavily on experience and intuition, limiting rapid advancement.
Accurate prediction of thermal properties like glass transition temperature (Tg) is crucial for material design.

Purpose of the Study:

To develop interpretable quantitative-structure-property relationship (QSPR) models using machine learning to predict polyester glass transition temperatures (Tg).
To facilitate the exploration and discovery of novel polyester materials with specific performance requirements.
To provide insights into the relationship between polyester chemical structure and thermal properties.

Main Methods:

Collected data for 695 polyesters with Tg values to build QSPR models.
Employed three different machine learning algorithms, including deep neural networks (DNN), for model development.
Utilized Morgan fingerprint with frequency (MFF) descriptors and Shapley Additive Explanations (SHAP) for interpretability and trend analysis.
Constructed a virtual polyester library and employed high-throughput screening and molecular dynamics (MD) simulations for validation.

Main Results:

The best DNN model achieved high accuracy with R² values of 0.9588 (training) and 0.9314 (testing).
SHAP analysis revealed novel physical trends linking Tg to substructure variations.
Identified 20 novel polyesters with low synthetic complexity, validated by MD simulations with an average absolute error of 9.42 °C.
Demonstrated the effectiveness of machine learning in improving material discovery efficiency.

Conclusions:

Machine learning-assisted QSPR models accurately predict polyester Tg, significantly enhancing material discovery efficiency.
The approach offers a powerful tool for understanding the microscopic origins of thermal properties in polyesters.
This work provides a promising perspective for accelerating the innovation of advanced polyester materials.