A new approach developed by researchers at MIT and Duke University could enhance the strength of polymer materials, potentially resulting in more durable plastics and a reduction in plastic waste.
By applying machine learning, the team discovered crosslinker molecules that, when integrated into polymers, improve their ability to resist tearing. These crosslinkers are mechanophores—molecules that change under mechanical stress.
“These molecules are promising for designing polymers that become tougher under stress. “They respond by becoming more resilient instead of breaking,” says Heather Kulik, senior author and MIT professor.
The specific crosslinkers identified are iron-based compounds known as ferrocenes, which had not been widely studied as mechanophores before. Testing one mechanophore can take weeks, but the team sped up the process using machine learning.
The study, published in ACS Central Science, lists MIT postdoctoral researcher Ilia Kevlishvili as lead author. Co-authors include Jafer Vakil, David Kastner, Xiao Huang, and Stephen Craig.
The Most Vulnerable Point
Mechanophores are molecules that uniquely react to mechanical force—often by altering their color, structure, or other characteristics. In a recent study, researchers from MIT and Duke explored whether these molecules could enhance the durability of polymer materials.
This research expands on a 2023 study led by Craig and MIT’s Jeremiah Johnson, the A. Thomas Guertin Professor of Chemistry. That earlier work revealed a surprising result: adding weak crosslinkers to a polymer network can actually strengthen the overall material. As the material nears breaking, cracks follow weaker bonds, forcing more breaks and boosting tear resistance.
To build on this insight, Craig teamed up with Kulik to identify mechanophores that could serve as weak crosslinkers.
“We gained a key insight,” Craig says, “but faced a big challenge: how to find the most promising molecules among countless options?” Credit goes to Heather and Ilia for recognizing this hurdle and developing a strategy to tackle it.”
Identifying mechanophores is difficult, requiring time-consuming experiments or resource-heavy simulations. Most studied mechanophores are organic, like cyclobutane used in a 2023 study.
Tuning Organometallic Compounds for Mechanophore Applications
In their latest work, scientists focused on ferrocenes, promising organometallic mechanophores. Ferrocenes feature an iron atom between two carbon rings and can be modified to alter their properties.
While many ferrocenes are already used in pharmaceuticals and catalysis, only a few have been investigated as mechanophores. Testing each candidate experimentally can take weeks, and even the faster computational methods still require days per molecule—making large-scale evaluation extremely difficult.
To overcome this bottleneck, the team from MIT and Duke turned to machine learning. They trained a neural network to predict which ferrocenes might function effectively as mechanophores.
Their starting point was the Cambridge Structural Database, which houses structural data for 5,000 ferrocenes that have already been synthesized.
“We knew we didn’t have to worry about whether these compounds could be synthesized—at least not the mechanophore portion—which gave us the freedom to explore a broad and chemically diverse set of molecules that were all synthetically accessible,” explains Kevlishvili.
Identifying Weak Points to Strengthen Polymers
To start, the team ran computational simulations on roughly 400 of the compounds to determine the amount of force needed to break atomic bonds within each molecule. For this study, they specifically sought compounds that would break apart more easily, as these weak points could enhance a polymer’s resistance to tearing.
Using the simulation results and structural data for each compound, the researchers trained a machine-learning model. This model could then estimate the activation force—how much force is needed to trigger the mechanophore—for the remaining 4,500 ferrocenes in the database, as well as for an additional 7,000 structurally similar compounds with slight atomic variations.
Through this process, they identified two key molecular features that seemed to boost tear resistance: interactions between the chemical groups attached to the ferrocene rings, and the presence of large, bulky side groups on both rings, which made the molecules more prone to breaking under stress.
According to the researchers, the first feature aligned with existing chemical knowledge, but the second was unexpected and likely wouldn’t have been predicted by a chemist alone—it only became apparent through the use of AI. “This was a genuinely surprising finding,” says Kulik.
Reinforced Plastics
After narrowing down around 100 promising candidates, Craig’s lab at Duke University created a polymer using one of them—m-TMS-Fc—as a crosslinker. In this setup, m-TMS-Fc connects the strands of polyacrylate, a common plastic.
When force was applied to these polymers until they broke, the team discovered that the seemingly weak m-TMS-Fc linker actually yielded a highly tear-resistant material. The resulting polymer was roughly four times tougher than those using traditional ferrocene crosslinkers.
“This finding is significant,” says Kevlishvili. “Stronger plastics could last longer in use, potentially cutting down the need for frequent replacements and helping to curb plastic waste over time.”
The team now aims to apply their machine-learning method to identify mechanophores with additional useful traits, such as color change or catalytic activity triggered by mechanical stress. These could serve as stress indicators, tunable catalysts, or even support biomedical functions like controlled drug release.
Future research will center on ferrocenes and other metal-based mechanophores that have been synthesized but remain poorly characterized.
“Transition metal-based mechanophores haven’t been studied extensively and are likely more difficult to synthesize,” Kulik explains. “However, this computational approach offers a powerful way to expand the range of mechanophores researchers can investigate.”
The study received support from the National Science Foundation’s Center for the Chemistry of Molecularly Optimized Networks (MONET).
Read the original article on: MIT
Read more: Microplastics Could Be Allies in Monitoring Sewage in Rivers