Researchers frequently pursue innovative materials derived from polymers. Instead of initiating a polymer search from the ground up, they conserve resources and time by merging existing polymers to attain specific attributes.

However, pinpointing the most effective blend presents a tricky dilemma. There is an virtually infinite array of possible combinations, and polymers engage with one another in intricate ways, making the characteristics of a new blend difficult to forecast.

To hasten the discovery of novel materials, MIT scientists created a fully automated experimental platform that can proficiently pinpoint optimal polymer blends.

The closed-loop process employs a sophisticated algorithm to navigate an extensive range of potential polymer blends, delivering a selection of pairings to a robotic system that combines materials and examines each blend.

Depending on the findings, the algorithm determines the subsequent experiments to carry out, continuing the cycle until the new polymer aligns with the user’s objectives.

Throughout the experimentation, the system independently identified numerous blends that surpassed the performance of their individual polymers. Notably, the researchers discovered that the highest-performing blends did not necessarily incorporate the best individual components.

“I found this to be strong confirmation of the utility of an optimization algorithm that takes the entire design space into consideration concurrently,” states Connor Coley, the Class of 1957 Career Development Assistant Professor in the MIT departments of Chemical Engineering and Electrical Engineering and Computer Science, as well as senior author of a publication on this innovative method. “By acknowledging the complete formulation space, one can potentially discover new or superior properties. Adopting a different methodology might easily lead to overlooking the underperforming components that are crucial to the most effective blend.”

This workflow could eventually assist in the discovery of polymer blend materials that facilitate progress in areas such as enhanced battery electrolytes, more affordable solar panels, or customized nanoparticles for safer drug delivery.

Coley collaborated on the publication with lead author Guangqi Wu, a former MIT postdoctoral researcher currently a Marie Skłodowska-Curie Postdoctoral Fellow at Oxford University; Tianyi Jin, an MIT graduate student; and Alfredo Alexander-Katz, the Michael and Sonja Koerner Professor in the MIT Department of Materials Science and Engineering. The findings are published today in Matter.

Formulating superior blends

When researchers create new polymer blends, they encounter a virtually boundless selection of possible polymers to choose from. After selecting a few for mixing, they still need to determine the composition of each polymer and the concentration of polymers within the blend.

“Having such an expansive design space requires algorithmic solutions and high-throughput workflows because you simply cannot evaluate all combinations through brute force,” Coley elaborates.

While scholars have investigated autonomous workflows for single polymers, less emphasis has been placed on polymer blends due to the significantly larger design space.

In this investigation, the MIT scientists sought new random heteropolymer blends, formulated by combining two or more polymers with distinct structural characteristics. These adaptable polymers have demonstrated considerable potential in high-temperature enzymatic catalysis, a process that accelerates chemical reactions.

Their closed-loop workflow initiates with an algorithm that autonomously selects a few promising polymer blends based on the desired properties defined by the user.

The researchers initially employed a machine-learning model to project the efficacy of new blends, but achieving accurate predictions across the astronomically extensive range of possibilities proved challenging. Consequently, they adopted a genetic algorithm, which employs biologically inspired methods like selection and mutation to discover an optimal solution.

The system encodes the makeup of a polymer blend into what can be regarded as a digital chromosome, which the genetic algorithm incrementally refines to isolate the most promising combinations.

“This algorithm is not new, but we needed to adapt it to fit within our system. For example, we had to restrict the number of polymers in a single material to enhance discovery efficiency,” Wu notes.

Furthermore, given the vastness of the search space, they adjusted the algorithm to balance its exploratory choices (seeking random polymers) versus its exploitative actions (optimizing the best polymers from preceding experiments).

The algorithm dispatches 96 polymer blends simultaneously to the autonomous robotic platform, which mixes the materials and assesses the properties of each.

The experiments aimed at enhancing the thermal stability of enzymes by optimizing retained enzymatic activity (REA), a metric to gauge how stable an enzyme is after interacting with the polymer blends and exposure to high temperatures.

The outcomes are relayed back to the algorithm, which utilizes them to produce a new set of polymers until it identifies the most suitable blend.

Expediting discovery

Constructing the robotic system presented numerous obstacles, including developing methods for uniformly heating polymers and enhancing the speed at which the pipette tip oscillates.

“In autonomous discovery platforms, we prioritize algorithmic innovations, yet numerous intricate and nuanced elements of the process require validation before the information can be trusted,” Coley remarks.

Upon testing, the optimal blends recognized by their system frequently outperformed the polymers from which they originated. The best overall blend achieved performance 18 percent superior to any of its individual components, reaching an REA of 73 percent.

“This suggests that rather than creating new polymers, we can sometimes blend existing polymers to formulate new materials that outperform individual polymers,” Wu states.

Moreover, their autonomous platform can generate and evaluate 700 new polymer blends daily, only necessitating human involvement for refilling and replacing materials.

While this study concentrated on polymers for protein stabilization, their platform could be adapted for other applications, such as developing new plastics or battery electrolytes.

In addition to exploring further polymer attributes, the researchers aim to leverage experimental data to enhance their algorithm’s efficiency and create new algorithms to refine the operations of the autonomous liquid handler.

“Technologically, there is a pressing need to improve the thermal stability of proteins and enzymes. The results presented here are notably impressive. Being a platform technology and given the rapid progress in machine learning and AI for materials science, one can envision the potential for this team to further enhance random heteropolymer performances or optimize designs based on specific needs and applications,” comments Ting Xu, an associate professor at the University of California at Berkeley, who was not involved in this research.

This work is partly funded by the U.S. Department of Energy, the National Science Foundation, and the Class of 1947 Career Development Chair.