Graph-Based Machine Learning Identifies Oxygenated Block Polymer Replacements for Conventional Plastics and Elastics

Area of Science:

• Materials Science
• Polymer Chemistry
• Machine Learning

Background:

• Oxygenated block polymers (esters, carbonates) are key for a circular plastics economy, aiming to match petrochemical polymer properties.
• Developing sustainable alternatives requires predicting polymer structures with specific thermomechanical performance.

Purpose of the Study:

• To introduce PolyReco, a machine learning approach for predicting oxygenated block polymer structures.
• To identify new block polymer combinations and polymerization degrees meeting target mechanical properties.

Main Methods:

• Representing triblock oxygenated polymers as graphs for feature extraction.
• Utilizing a link prediction algorithm within the PolyReco framework.
• Pairing the predictive model with a visualization tool for material selection.

Main Results:

• PolyReco successfully predicted novel oxygenated block polymers with high-performance mechanical properties.
• Experimental validation confirmed the predicted tensile mechanical properties, matching those of high-impact polystyrene, poly(dimethylsiloxane), and styrenic elastomers.
• Three case studies demonstrated the tool's predictive accuracy and utility.

Conclusions:

• The PolyReco machine learning approach can accelerate the discovery of sustainable oxygenated block polymers.
• This methodology aids in reducing reliance on fossil-based polymers by identifying viable, high-performance alternatives.
• The predicted polymers show potential to fill the property gaps left by current sustainable material options.