Developing a Suitable Model for Water Uptake for Biodegradable Polymers Using Small Training Sets.

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Modeling

Background:

• Accelerating polymer selection for specific applications requires accurate prediction of dynamic properties like water uptake.
• Traditional methods of polymer characterization can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop and validate predictive models for polymer water uptake.
• To identify suitable polymers for applications based on their water absorption characteristics.
• To demonstrate the utility of surrogate modeling in reducing experimental synthesis and characterization efforts.

Main Methods:

• Construction of semiempirical models using Artificial Neural Networks (ANNs).
• Utilizing all water uptake data versus consolidated parameters (equilibrium water uptake) as model inputs.
• Cross-validation and testing of model accuracy on independent datasets.
• Application of the validated model to a library of L-tyrosine-derived polyarylates.

Main Results:

• ANN models using individual water uptake data showed high correlation (R² > 0.78) but low cross-validation accuracy (<19%).
• Models incorporating equilibrium water uptake achieved good correlation (R² = 0.78) and accurately predicted properties for 50% of tested polymers.
• The study successfully identified polymer groups within a 56-polymer library likely to meet water uptake design criteria.

Conclusions:

• Consolidated parameters like equilibrium water uptake provide more accurate and reliable predictive models for polymer water absorption.
• Surrogate modeling significantly reduces the number of polymers requiring synthesis and experimental validation.
• This approach enhances efficiency in selecting polymers for targeted applications based on water uptake performance.