After a traumatic spinal cord injury, a cellular structure known as a glial scar forms around the site of damage – and while it prevents damage from spreading to healthy tissue, the glial scar also prevents neuronal regrowth. There is an enzyme that munches away at glial scars, making it a possible therapeutic spinal cord treatment. As luck would have it, though, the enzyme degrades when exposed to the temperature of the human body. 

“In the presence of a copolymer, the enzyme might be able to survive at physiological conditions for a greater period of time,” said Michael Webb, assistant professor of chemical and biological engineering at Princeton University. Exploring the use of a copolymer to tailor the functionality of an enzyme is one of the many problems in Webb’s research portfolio.

Webb works at the intersection of machine learning and polymer materials. A polymer is a substance composed of many repeating chemical units. Examples include natural and synthetic materials such as proteins, polyester, DNA, silicone, starches, and Teflon. “Polymers are big macromolecules that feature many different units with different chemical constitutions,” said Webb. “The interesting thing is you can take exactly the same constitutional units and arrange them in different ways – in different sequences, varying compositions, different architectures – and all of that can have a substantial impact on their materials properties.”

Using a type of molecular modeling on polymer materials, Webb and his colleagues gather data on complex polymeric structures. Complex polymers are made up of many different chemical units which repeat, instead of just a single repeating chemical unit. The data collected by the researchers is used to build machine learning models, which, in turn, are used to help understand how the material properties might be optimized – even for materials that don’t yet exist. “We are building tools and approaches that are able to address the materials of this complexity that may not yet have a physical manifestation,” said Webb.

Navigating a vast design space

On Oct. 1, Webb gave a seminar at the Center for Statistics and Machine Learning as a part of the center’s ongoing Lunchtime Faculty Seminar series hosted by the center. In his talk, Webb said that his group is particularly interested in creating “smart materials”, which work by responding to external stimuli. “We take a lot of inspiration from biological phenomena and create materials that can interact with systems from biology or otherwise mimic what can be achieved in biological context,” said Webb.

Webb’s group is predominantly centered around simulation rather than the actual creation of these types of materials. As a group concerned with simulation, Webb said he and his colleagues have found “increasing utility and opportunity for the use of machine learning” in their research. To alter the property of a material, its chemistries and architecture have to be arranged in different manners. “What we think of a lot as a group is how we might be able to utilize machine learning to help navigate this vast design space.”

In the future, Webb says, he and his colleagues are looking forward to hopefully learning how to manipulate material systems over time, effectuating changes in their responses to external stimuli. “I think that will require a new set of methods and strategies,” said Webb.