Machine Learning for Melting Temperature Predictions and Design in Polyhydroxyalkanoate-Based Biopolymers

Area of Science:

• Polymer science and sustainable materials development.
• Application of computational methods in materials discovery.
• Environmental science and plastic pollution mitigation.

Background:

• Growing demand for sustainable and biodegradable polymers due to fossil fuel depletion and plastic pollution.
• Polyhydroxyalkanoates (PHAs) offer eco-friendly and versatile alternatives but face design challenges.
• Optimizing PHA chemical composition for specific applications requires efficient structure-property mapping.

Purpose of the Study:

• To demonstrate the utility of machine learning (ML) for predicting polyhydroxyalkanoate (PHA) properties.
• To establish efficient structure-property relationships within the PHA chemical space.
• To facilitate the design and optimization of novel PHA materials.

Main Methods:

• Utilized a curated dataset of experimentally measured melting temperatures (Tm) for various PHA homo- and copolymers.
• Developed polymer descriptors based on topology, shape, and charge/polarity of backbone motifs.
• Employed machine learning models for predicting Tm and estimating prediction uncertainties.

Main Results:

• Successfully developed ML models capable of predicting Tm for diverse PHA copolymers.
• Demonstrated rapid property prediction for multicomponent PHA copolymers.
• Integrated Tm prediction with existing Tg models and evolutionary algorithms for multiobjective optimization.

Conclusions:

• Machine learning provides a powerful tool for accelerating the discovery and design of polyhydroxyalkanoates (PHAs).
• The developed structure-property mapping approach aids in identifying optimal PHA compositions for specific applications.
• This methodology addresses multiobjective optimization challenges in polymer design for sustainable materials.