Traditional design strategies for synthetic polymers and organic molecules are experiment-based, guided by experience and intuition, and driven by application requirements. Due to the ever-growing demand for new materials and the vast number of already existing organic molecules, traditional methods are facing significant challenges. With rapid advances in high-throughput computing, machining learning (ML), and artificial intelligence (AI) applications, polymer informatics is emerging as a promising tool to ensure breakthrough discoveries in the field of polymer science.
Image Credit: Zakharchuk/Shutterstock.com
Polymers are one of the most ubiquitous classes of materials in modern society. Their applications range from packaging, textiles, and consumer goods manufacturing to medicine, construction, and transportation. A polymer material consists of many repeating units, called monomers, assembled in long molecular chains. These polymer chains can form different structures resulting in polymer materials with highly diverse physical and chemical properties. Some polymer chains can include more than one type of monomer, thus creating even more complex topological structures on different length scales.
Traditional research methods based on intuition and trial-and-error have already revealed the great potential that polymer materials hold. However, new approaches are required to identify and develop novel applications given the macromolecular structural variety of specific polymer materials. The emerging field of polymer informatics aims to answer this demand by employing AI- and ML-based methods for data- and information-centric research.
How Can Polymer Informatics Help Polymer Research?
Building reliable empirical models based on fundamental physical and chemical properties can facilitate the prediction of polymer characteristics. Polymer Genome is an ML-based tool capable of rapidly predicting a variety of polymer properties. The prediction models used in Polymer Genome are trained on existing polymers databases, experimental polymer data, or first principles computations.
In polymer informatics, a large pool of chemically or synthetically feasible polymers can be used for high-throughput screening of potential candidates by applying predictive models relevant to the target material properties.
Polymer synthesis can be a costly and labor-intensive process; thus, experimental design algorithms can be an ideal solution to minimize the number of new experiments required to reach a design goal. Such algorithms exploit the knowledge embedded in existing data to suggest which candidates may best satisfy the design goal.
Key Developments in the Polymer Industry Using AI and ML
The polymer materials must meet several desired property requirements to be a good candidate for any specific application. With sufficiently large datasets to support the use of modern ML technology, polymer informatics can be a powerful tool for the discovery of novel polymers. Many research groups adopt data-driven approaches to polymer design and significantly improve their productivity of developing new functional polymers for the rapidly expanding market of polymer materials.
Polymer Dielectrics for Energy Storage
With the growing demand for polymer-based high-performance energy storage capacitors, polymer informatics can significantly facilitate discovery of novel polymer dielectrics.
Defining multiple desired properties, such as high glass transition temperature and dielectric strength, combined with computation and data-driven ML strategies, has resulted in developing and synthesizing novel dielectric films; these have shown excellent thermal stability at extreme temperatures.
Polymer Electrolytes for Li-Ion Batteries
Another example where AI-based property prediction models and design algorithms have proven successful is the development of safer solid polymer electrolyte materials for rechargeable Li-ion batteries. A research group from the Department of Materials Science and Engineering at MIT in the USA used ML-aided coarse-grained molecular dynamics simulations to design polymer electrolytes with high ionic conductivity and enhanced electrochemical stability. Their research was published in the journal Chemistry of Materials on April 22nd, 2020.
Conducting Polymers for Electronic Applications
Although most polymers are insulators, a class of intrinsically conducting polymers called conjugated polymers is extensively used in light-emitting diode and organic solar cells applications. Molecular doping can enhance the electrical properties of polymers, but discovering optimal polymer-dopant pairs providing optimal conductivity can be a time-consuming process given the complex nature of the electron transfer processes in polymers. Several high-performing donor/acceptor pairs for organic solar cells were recently discovered by employing regression-based property prediction models for organic molecules.
Polymer Membranes for Gas and Liquid Separation
Polymers with high surface area show great promise as membrane materials in fluid separation applications. Finding promising polymer candidates with high intrinsic microporosity and high permeability for the desired gases or liquids is a trivial task, but a potential solution may have been found. Researchers from the Georgia Institute of Technology, Atlanta, USA, accelerated their polymer membrane research utilizing Polymer Genome to build gas permeability prediction models for different polymer compounds.
Discovery of Novel Depolymerizable and Biodegradable Polymers
Creating polymers capable of reversible polymerization-depolymerization when exposed to particular stimuli is one of the biggest challenges to polymer science. Such recyclable polymers would permit feeding resources back into the polymer production with minimum value loss.
Recently, a report was published by researchers from the University of Akron in the USA. Here, data-driven computational tools were employed to investigate the energy landscapes of different ring-opening polymerization-depolymerization reactions of cyclic monomers, thus screening potential candidates for chemically recyclable industrial polymers.
A team from Los Alamos National Laboratory in New Mexico, USA, took a similar approach, proposing an ML-based system for rapid structure-property mapping of bio-based polymers from existing polymer databases to discover new biodegradable polymer candidates.
Challenges and Future Prospects for Polymer Informatics
The availability of large open databases suitable for ML applications forms the foundation of polymer informatics. Establishing these databases, however, is challenging for several reasons.
Inconsistencies throughout the history of polymer science and lack of data sharing due to privately (industrially) owned research hinder creating a comprehensive polymer database. Besides, some of the existing data on the hierarchical structure of polymers prove difficult to encode for ML purposes.
Nonetheless, academic and industrial research groups around the globe are taking steps towards establishing fully data-driven research and development processes for polymers. The real benefit from this would be the release of polymer scientists from the inefficient trial-and-error design process, allowing scientists to focus on broader design concepts and theoretical advancements.
References and Further Reading
Chen, L., et al. (2021) Polymer informatics: Current status and critical next steps. Mater. Sci. Eng. R Rep. 144, 100595. Available at: https://doi.org/10.1016/j.mser.2020.100595
Kuenneth, K., et al. (2021) Polymer informatics with multi-task learning. Patterns 2, 4, 100238. Available at: https://doi.org/10.1016/j.patter.2021.100238
Chen, G., at al. (2021) Predicting Polymers’ Glass Transition Temperature by a Chemical Language Processing Model. Polymers 13, 1898. Available at: https://doi.org/10.3390/polym13111898
Chen, G., et al. (2020) Machine-Learning-Assisted De Novo Design of Organic Molecules and Polymers: Opportunities and Challenges. Polymers 12, 163. Available at: https://doi.org/10.3390/polym12010163
Wang, Y., et al. (2020) Toward Designing Highly Conductive Polymer Electrolytes by Machine Learning Assisted Coarse-Grained Molecular Dynamics. Chemistry of Materials 32 (10), 4144-4151. Available at: https://doi.org/10.1021/acs.chemmater.9b04830
Tran, H. D., et al. (2020) Machine-learning predictions of polymer properties with Polymer Genome. Journal of Applied Physics 128, 171104. Available at: https://doi.org/10.1063/5.0023759
Pilania, G., et al. (2019) Machine-Learning-Based Predictive Modeling of Glass Transition Temperatures: A Case of Polyhydroxyalkanoate Homopolymers and Copolymers. Journal of Chemical Information and Modeling 59 (12), 5013-5025. Available at: https://doi.org/10.1021/acs.jcim.9b00807
Kim, C., et al. (2018) Polymer Genome: A Data-Powered Polymer Informatics Platform for Property Predictions. J. Phys. Chem. C 122, 17575−17585. Available at: https://doi.org/10.1021/acs.jpcc.8b02913