Machine Learning Discovery of High-Temperature Polymers

Abstract

The design and development of high-temperature polymers have been an experimentally-driven and trial-and-error process guided by experience, intuition, and conceptual insights. However, such an Edisonian approach is often costly, slow, biased towards certain chemical space domains, and limited to relatively small-scale studies, which may easily miss promising compounds. To overcome this challenge, we formulate a data-driven machine-learning (ML) approach, integrated with high-fidelity molecular dynamics (MD) simulations, for quantitatively predicting polymer’s glass transition temperature (T_g) from its chemical structure and rapid screening of promising candidates for high-temperature polymers. Specifically, we collect a diverse set of nearly 13,000 real polymers from the largest polymer database, PoLyInfo. Among them, 6,923 experimental T_g values are available; while, the remaining 5,690 polymers do not have reported T_g values. We train the deep neural network (DNN) model with 6,923 experimental T_g values using Morgan fingerprint representations of chemical structures for these polymers. Interestingly, the trained DNN model can reasonably predict the unknown T_g values of polymers with distinct molecular structures, in comparison with MD simulations and experimental results. We find that this excellent transferability is attributed to the feature representation of Morgan fingerprints, which carry the chemical connectivity between neighboring repeating units in a polymer and the frequency of occurrence of different chemical substructures. With the validated ML model for high-throughput screening of nearly 1 million hypothetical polymers, we identify more than 65,000 promising candidates with T_g>200 oC, which is 30 times more than existing known high-temperature polymers (~2,000 from PoLyInfo). The discovery of this large number of promising candidates will be of significant interest in the development and design of high-temperature polymers. In short, our work demonstrates that ML is a powerful method for the prediction and rapid screening of high-temperature polymers, particularly with growing large sets of experimental and computational data for polymeric materials.