Machine learning-assisted polymer design

Polymers are some of the most important and widely used chemical products in industry and consumer markets, and they find applications in an increasingly wide range of fields from aerospace engineering to biomedical applications, drug delivery, electronics, tissue engineering, cosmetics, and packaging. Polymers are made up of chains of repeating subunits called monomers and display an incredibly wide range of physical-chemical properties. These properties depend on many factors including the polymer molecular weight, the monomer’s molecular properties, the conformation of the monomers within the polymer chain, the conformation of the chain itself, and its interaction with other chains.

Synthesizing polymers to have specific properties (density, conductivity, heat capacity etc.) is currently a time- and resource-intensive process. Researchers select monomers, target properties, a target molecular weight, and a manufacturing process according to their experience and a literature search. They then synthesise the polymer, and finally perform a wide range of analyses to determine its properties and verify if they match the initial target. This three-step process is repeated until the properties of the synthesised polymer match the desired requirements. This process is extremely time- and resource-consuming for the polymer industry and research because it is based on a high number of experimental variables, on the researcher’s “chemical intuition”, and on an inefficient trial and error strategy.

To mitigate this process, we propose to work at the interface between data science and polymer chemistry and use of Machine Learning (ML) algorithms to predict the bulk polymer properties from the molecular structure of the monomer. More specifically, we aim to address the following challenges of ML-assisted polymer design:
1. Identify which monomer properties (quantum mechanical, chemical, physical and structural) can best predict the properties their corresponding polymers.
2. Identify the best algorithms for predicting the polymer properties and their relationship to the polymer molecular weight.
3. Exploretheconfigurationalspaceofpolymerstotargetspecificphysical-chemicalpropertiesand ultimately decrease the costs of polymer research.