Polymers for Extreme Conditions Designed Using Syntax-Directed Variational Autoencoders

Abstract

The design/discovery of new materials is highly nontrivial owing to the near-infinite possibilities of material candidates and multiple required property/performance objectives. Thus, machine learning tools are now commonly employed to virtually screen material candidates with desired properties by learning a theoretical mapping from material-to-property space, referred to as the forward problem. However, this approach is inefficient and severely constrained by the candidates that the human imagination can conceive. Thus, in this work on polymers, we tackle the materials discovery challenge by solving the inverse problem: directly generating candidates that satisfy desired property/performance objectives. We utilize syntax-directed variational autoencoders (VAE) in tandem with Gaussian process regression (GPR) models to discover polymers expected to be robust under three extreme conditions: (1) high temperatures, (2) high electric field, and (3) high temperature and high electric field, useful for critical structural, electrical, and energy storage applications. This approach to learn from (and augment) human ingenuity is general and can be extended to discover polymers with other targeted properties and performance measures.

Predicting Polymers’ Glass Transition Temperature by a Chemical Language Processing Model

“…For high-throughput screening tasks, the candidates with extreme properties are usually desired and of great value in material discovery [ 80 , 81 ]. As an example, twelve candidates in this unlabeled database with larger than 400 °C are quickly identified, as shown in Figure 6 c, although their values have not been reported before.…”

“…Particularly, we find the chemical structures of these identified polymers share similar features as other high-temperature polymers, such as polyaryletherketone and polyimide. For instance, saturated 4,5 member rings, bridged rings, benzene rings, oxolane groups, amine groups, and halogens had a higher occurrence rate for polymers with high [ 81 , 82 , 83 ]. For preliminary validation of ML predictions, we have performed all-atom molecular dynamics (MD) simulations on these polymers, with simulation protocols and detailed results given in the Supporting Information .…”

Machine Learning in Chemical Product Engineering: The State of the Art and a Guide for Newcomers

“…At the same time, it is also very common to combine different ML techniques, mostly in a sequential manner, within the framework of the same problem. Typical examples include the implementation of a dimensionality reduction technique prior to application of a regression or classification method, in order to reduce the feature space and select the most relevant inputs, thus, reducing the computational cost of the latter step [35,[65][66][67][68]. Although the characteristics and the objectives of these combinatorial modeling approaches are quite different compared with those of the aforementioned hybrid models, they are sometimes used interchangeably in reported studies [52,65].…”

“…The availability of data is also more or less dependent on the application area. For example, publicly available data are less abundant in the organic materials and polymer research domains, compared to inorganic materials and drug design [10,68,79,187]. Examples of databases can be found for organic materials [10], inorganic materials [11], chemicals [87], materials [188] and molecules and solid materials [5].…”

Featurization Strategies for Polymer Sequence or Composition Design by Machine Learning

“…41 Batra et al have also demonstrated the use of VAEs to translate a modified, polymer-based SMILES grammar into a suitable vector space for constructing Gaussian process regression models to predict glass-transition temperatures and bandgaps of homopolymers. 67 Featurization for sequence-defined polymer systems can be pursued in several ways. For example, feature extraction architectures may be used to learn relevant sequence and topological correlations during supervised ML.…”