Explainable random forest predictions of polyester biodegradability using high-throughput biodegradation data

Area of Science:

• Polymer Science
• Biotechnology
• Computational Chemistry

Background:

• Growing demand for sustainable polymers necessitates efficient biodegradability assessment.
• Traditional biodegradability testing is time-consuming and costly, hindering rapid material development.
• High-throughput screening (HTS) offers a faster alternative for material biodegradability evaluation.

Purpose of the Study:

• To develop a high-throughput enzymatic biodegradation assay for polyesters.
• To create a machine learning model for predicting polyester biodegradability.
• To identify key structural features influencing polyester biodegradability.

Main Methods:

• Development of a high-throughput enzymatic biodegradation assay.
• Testing the biodegradability of 48 distinct polyester samples.
• Training and validating an explainable random forest model using assay data.
• Investigating transfer learning and model chaining for enhanced prediction accuracy.
• Utilizing SHAP analysis to interpret model predictions and identify influential structural features.

Main Results:

• The high-throughput assay successfully assessed the biodegradability of 48 polyesters.
• A predictive model achieved 71% accuracy in forecasting polyester biodegradability.
• Transfer learning and model chaining showed potential for improving predictive performance.
• SHAP analysis revealed specific structural characteristics that enhance polyester biodegradability.

Conclusions:

• The developed high-throughput assay and machine learning model significantly accelerate biodegradability testing.
• The predictive model provides a valuable tool for designing novel biodegradable polyesters.
• Understanding structure-biodegradability relationships aids in the rational design of sustainable polymers.