What if plastics could heal, sense stress, or last longer? A machine learning method finds hidden molecules that may change how we design materials.
A strategy developed by researchers at MIT and Duke University uses machine learning to identify crosslinker molecules that make polymers more resistant to tearing. This may help reduce plastic waste. The method focuses on mechanophores, which are molecules that change structure or behavior when exposed to force, and uses a neural network to screen candidates.
The researchers applied this method to ferrocenes, iron-containing compounds not widely studied as mechanophores. Tests showed that the material was about four times tougher than those made with standard ferrocene-based crosslinkers.
The first step involved simulating about 400 ferrocenes to calculate the force needed to break bonds. This data trained a machine-learning model to predict tear resistance in other compounds. The model found two features that improved performance — interactions between chemical groups on the rings and the presence of large groups on both sides of the ferrocene. While the first feature was expected, the second was a result found by the AI and not predicted by a human chemist.
Finding and testing mechanophores is usually a slow process. Most known mechanophores are organic, like cyclobutane, which has been used as a crosslinker. Testing one mechanophore in the lab can take weeks. Even simulations take days. This makes traditional screening hard when there are thousands of options.
Ferrocenes are common in catalysis and drug research but have not been tested much as mechanophores. These organometallic compounds have an iron atom between two carbon-based rings. The rings carry chemical groups that affect how the molecule reacts to force.
By working with ferrocenes and using machine learning, the team showed a more scalable way to find useful mechanophores. This approach could also help find compounds that respond to force by changing color, triggering reactions, or other behaviors. This could be useful in sensing stress, switching catalysts, or delivering drugs.
Future work will likely look at ferrocenes and other metal-based compounds to find more mechanophores and develop new materials.
