High-throughput experimentation for discovery of biodegradable polyesters.

The consistent rise of plastic pollution has stimulated interest in the development of biodegradable plastics. However, the study of polymer biodegradation has historically been limited to a small number of polymers due to costly and slow standard methods for measuring degradation, slowing new material innovation. High-throughput polymer synthesis and a high-throughput polymer biodegradation method are developed and applied to generate a biodegradation dataset for 642 chemically distinct polyesters and polycarbonates. The biodegradation assay was based on the clear-zone technique, using automation to optically observe the degradation of suspended polymer particles under the action of a single Pseudomonas lemoignei bacterial colony. Biodegradability was found to depend strongly on aliphatic repeat unit length, with chains less than 15 carbons and short side chains improving biodegradability. Aromatic backbone groups were generally detrimental to biodegradability; however, ortho- and para-substituted benzene rings in the backbone were more likely to be degradable than metasubstituted rings. Additionally, backbone ether groups improved biodegradability. While other heteroatoms did not show a clear improvement in biodegradability, they did demonstrate increases in biodegradation rates. Machine learning (ML) models were leveraged to predict biodegradability on this large dataset with accuracies over 82% using only chemical structure descriptors.

Random Forest Predictor for Diblock Copolymer Phase Behavior.

Physics-based models are the primary approach for modeling the phase behavior of block copolymers. However, the successful use of self-consistent field theory (SCFT) for designing new materials relies on the correct chemistry- and temperature-dependent Flory-Huggins interaction parameter χAB that quantifies the incompatibility between the two blocks A and B as well as accurate estimation of the ratio of Kuhn lengths (bA/bB) and block densities. This work uses machine learning to model the phase behavior of AB diblock copolymers by using the chemical identities of blocks directly, obviating the need for measurement of χAB and bA/bB. The random forest approach employed predicts the phase behavior with almost 90% accuracy after training on a data set of 4768 data points, almost twice the accuracy obtained using SCFT employing χAB from group contribution theory. The machine-learning model is notably sensitive toward the uncertainty in measuring molecular parameters; however, its accuracy still remains at least 60% even for highly uncertain experimental measurements. Accuracy is substantially reduced when extrapolating to chemistries outside the training set. This work demonstrates that a random forest phase predictor performs remarkably well in many scenarios, providing an opportunity to predict self-assembly without measurement of molecular parameters.