AI-guided inverse design and discovery of recyclable vitrimeric polymers

TL;DR

This work couple molecular dynamics simulations and a novel graph variational autoencoder (VAE) machine learning model for inverse design of vitrimer chemistries with desired glass transition temperature (Tg) and synthesize a novel vitrimer polymer.

Abstract

Vitrimer is a new, exciting class of sustainable polymers with the ability to heal due to their dynamic covalent adaptive network that can go through associative rearrangement reactions. However, a limited choice of constituent molecules restricts their property space, prohibiting full realization of their potential applications. To overcome this challenge, we couple molecular dynamics (MD) simulations and a novel graph variational autoencoder (VAE) machine learning model for inverse design of vitrimer chemistries with desired glass transition temperature (Tg) and synthesize a novel vitrimer polymer. We build the first vitrimer dataset of one million chemistries and calculate Tg on 8,424 of them by high-throughput MD simulations calibrated by a Gaussian process model. The proposed novel VAE employs dual graph encoders and a latent dimension overlapping scheme which allows for individual representation of multi-component vitrimers. By constructing a continuous latent space containing necessary information of vitrimers, we demonstrate high accuracy and efficiency of our framework in discovering novel vitrimers with desirable Tg beyond the training regime. To validate the effectiveness of our framework in experiments, we generate novel vitrimer chemistries with a target Tg = 323 K. By incorporating chemical intuition, we synthesize a vitrimer with Tg of 311-317 K, and experimentally demonstrate healability and flowability. The proposed framework offers an exciting tool for polymer chemists to design and synthesize novel, sustainable vitrimer polymers for a facet of applications.

Figures (6)

• Figure 1: Schematic overview of this work.(a) A transesterification vitrimer comprises a carboxylic acid and an epoxide. The reversible covalent bond between acid and epoxide allows them to detach from and attach to each other, thus healing the polymer. The design space for vitrimers is defined as all possible combinations of 50,000 carboxylic acids and 50,000 epoxides and a vitrimer dataset is built by sampling from the design space. (b) We use calibrated MD simulations to calculate $T_\mathrm{g}$ on a subset of vitrimers. The vitrimer dataset and $T_\mathrm{g}$ are inputs to the VAE model. (c) By optimizing latent vectors according to desirable $T_\mathrm{g}$, novel vitrimers with $T_\mathrm{g}$ = 569 K and 248 K are discovered. (d) Synthesis of novel vitrimer chemistry proposed by the framework for target $T_\mathrm{g}$ of 323 K (50 ° C).
• Figure 2: Data generation by MD simulations and calibration by GP model.(a) The vitrimer dataset is obtained by randomly sampling one million combinations between 50,000 bifunctional carboxylic acids and 50,000 epoxides derived from the ZINC15 database. MD simulations are carried out to calculate $T_\mathrm{g}$ on a subset of 8,424 vitrimers. (b) We train a GP model to predict experiment-MD difference $\Delta T_\mathrm{g}$ with a training set of 295 polymers with experimental $T_\mathrm{g}$ in literature. (c) Using the trained GP model, we calibrate MD-calculated $T_\mathrm{g}$ of the vitrimer dataset. The calibrated $T_\mathrm{g}$, serving as a proxy of experimental $T_\mathrm{g}$, is the design target of this work.
• Figure 3: Illustration of the VAE model. The encoders convert acid and epoxide molecules into latent vectors $\boldsymbol{z}$ in a continuous latent space. The latent vectors $\boldsymbol{z}$ are further decoded into acid and epoxide molecules by the decoders. A property predictor is added to predict $T_\mathrm{g}$ from $\boldsymbol{z}$.
• Figure 4: Exploration in the latent space to discover novel vitrimers.(a)(b)(c) Starting with a known vitrimer as origin (adipic acid and bisphenol A diglycidyl ether), vitrimers are discovered by perturbing its latent vector in (a) acid-specific dimensions, (b) epoxide-specific dimensions and (c) all dimensions. (d) Novel vitrimers are identified along the interpolation path between two vitrimers in the training set. (e) The distribution of discovered vitrimers is visualized in the latent space by PCA. (f)$T_\mathrm{g}$ of discovered vitrimers. All presented $T_\mathrm{g}$ values of are validated by MD simulations and GP calibration.
• Figure 5: Inverse design of novel vitrimers by Bayesian optimization based on three targets of desirable $T_\mathrm{g}$:(a) Maximum $T_\mathrm{g}$, (b)$T_\mathrm{g} = 373~\mathrm{K}$ and (c)$T_\mathrm{g} = 248~\mathrm{K}$. All presented $T_\mathrm{g}$ values of proposed vitrimers are validated by MD simulations and GP calibration.