According to researchers at MIT and Duke University, new strategies for strengthening polymeric materials could lead to more durable plastics and reduce plastic waste.
Using machine learning, researchers were able to identify crosslinker molecules that could be added to polymeric materials and withstand more forces before tearing. These crosslinkers belong to a class of molecules known as mechanophores that change shape or other properties depending on mechanical force.
“These molecules help to make strong polymers depending on the force. You put some stress on them and see them that have a higher resilience rather than cracking or breaking them,” says Heather Kulik, professor of chemical engineering at MIT, who is also a senior author of chemistry.
The crosslinker identified by the researchers in this study is an iron-containing compound known as ferrocene, which has not been widely explored for its potential as a mechanophore up until now. Although it can take several weeks to experimentally evaluate a single menophore, researchers have shown that machine learning models can be used to dramatically speed up the process.
MIT Postdoc Ilia Kevlishvili is the lead author of the open access paper that appeared on Friday in ACS Central Science. Other authors include Jafer Vakil, a graduate student at Duke. David Kastner and Xiao Huang, graduate students of MIT. Stephen Craig, Duke's professor of chemistry.
The weakest link
Mechanophores are molecules that respond to forces in a unique way, usually by altering color, structure, or other properties. In a new study, the MIT and Duke team wanted to investigate whether the polymer could help make it more resilient to damage.
This new work is based on 2023 research from Craig and Jeremiah Johnson, A. Thomas Goutin, professor of chemistry at MIT, and his colleagues. In that study, the researchers surprisingly found that incorporating weak crosslinkers into the polymer network could lead to stronger overall materials. When materials with these weak crosslinkers extend to the breaking point, cracks propagating through the material avoid stronger bonds and instead try to pass through weak bonds. This means that if all bonds were of the same strength, the cracks must break more bonds.
To find a new way to exploit that phenomenon, Craig and Crik joined forces to identify mechanophores that could be used as weak crosslinkers.
“We had this new mechanical insight and opportunity, but it poses a huge challenge. Of all possible compositions of matter, how do we enter zero into something with the greatest potential?” Craig says. “A complete credit to Heather and Ilia for identifying and devising an approach to satisfy this challenge.”
Discovering and characterizing mechanophores is a challenging task that requires either time-consuming experiments or computationally intense simulations of molecular interactions. Most known mechanophores are organic compounds such as cyclobutane, which were used as crosslinkers in a 2023 study.
In the new study, researchers wanted to focus on a molecule known as ferrocene. Ferrocene is an organometallic compound with an iron atom sandwiched between two carbon-containing rings. These rings add different chemical groups to them, changing their chemical and mechanical properties.
Many ferrocenes are used as medicines or catalysts, and a handful are known to be good mechanophores, but most have not been evaluated for their use. Experimental testing of a single potential mechanophore can take weeks, while computational simulations are faster and take days. Evaluating thousands of candidates using these strategies can be a challenging task.
Recognizing that machine learning approaches can dramatically speed up the characterization of these molecules, MIT and the Duke team decided to use neural networks to identify ferrocenes that could become promising mechanophores.
They began with information from a database known as the Cambridge Structure Database, which contains the structures of 5,000 ferrocenes that have already been synthesized.
“I knew I didn't have to worry about the issue of synthesisability, at least from the perspective of the mechanophore itself, so I chose a very large space to explore with a lot of chemical diversity.
First, the researchers were able to perform approximately 400 computational simulations of these compounds, calculating the forces needed to separate atoms within each molecule. In this application, they were looking for molecules that would quickly fall apart, as these weak links could make polymeric materials resistant to tearing.
This data was then used to train a machine learning model using information about the structure of each compound. This model was able to predict the force required to activate the mechanophores affecting tear resistance for the remaining 4,500 compounds in the database, as well as the additional 7,000 compounds in the database lined up with atoms.
Researchers have discovered two major features that are likely to increase tear resistance. One was interactions between chemical groups attached to ferrocene rings. Furthermore, the presence of large bulky molecules attached to both rings of ferrocene increased the likelihood that the molecules would fall apart depending on their application force.
The first of these features wasn't surprising, but the second property was not predicted in advance by the chemists and could not be detected without AI, researchers say. “This was really amazing,” says Kulik.
More strict plastic
Once researchers identified around 100 promising candidates, Duke's Craig lab synthesized a polymeric material that incorporated one of them, known as M-TMS-FC. Within the material, M-TMS-FC functions as a crosslinker and connects the polymer chains that make up polyacrylate, a type of plastic.
By applying force to each polymer until it is tear, the researchers discovered that weak M-TMS-FC linkers produce strong tear-resistant polymers. This polymer was found to be about four times stronger than polymers made with standard ferrocene as a crosslinker.
“That really means a lot, because given the accumulation of all the plastics and all the plastic waste we use, if we make the materials tougher, that means they will have a longer lifespan.
Researchers now want to use machine learning approaches to identify mechanophores with other desirable properties, such as the ability to change colour and become catalytically active in response to forces. Such materials can be used as stress sensors or switchable catalysts. It is also useful for biomedical applications such as drug delivery.
In these studies, researchers plan to focus on other metal-containing mechanophores that have already been synthesized but whose properties are not fully understood.
“The transition metal mechanophores are relatively inconspicuous and perhaps a little more challenging to make,” Kulik says. “This computational workflow can be widely used to expand the space of menophores people have studied.”
This study was funded by the National Science Foundation Center for Molecularly Optimulized Networks (Monet).
