Machine learning unlocks fluorescent molecular tools for information encryption

Researchers in Switzerland and Australia have used machine learning to crack the code governing charge transfer and colour emission in chains of molecules.

Chains of molecules, known as polymers, can be put together in patterns to create different visual effects, such as emitting a certain colour when exposed to ultraviolet light or other light sources.

Polymers are used in data storage, security inks, organic light-emitting diodes (OLEDs), and even the solar energy industry.

Until now, getting the molecules in the right order to create the desired effect has been a slow process of trial and error, limiting its practical application and usefulness.

To solve this problem, Exciton Science Research Fellow Dr Nastaran Meftahi of RMIT University, under the supervision of Professor Salvy Russo, trained a machine learning model to better understand the behaviour occurring inside and between the molecules.

Despite only having a limited amount of data to study, the model Nastaran created proved to be a fast learner.

It has revealed significant new information about how to assemble polymers more efficiently and effectively for a wide variety of experimental and commercial purposes.

Guided by the model, PhD student Suiying Ye and her supervisor, senior scientist Dr Yinyin Bao at ETH Zurich, together with their collaborators, were able to confirm that the transfer of charge within the chain or network is dependent on the pattern the molecules are arranged in, and their distance from one another.

With this information, the researchers are now able to create molecule chains that will display designated colours in response to different stimuli, including light, chemicals and energy.

The results have been published in the journal Chem and are available here.

“Once we had the machine learning model that told us how to get all the other colors, we could achieve a full spectrum,” Nastaran said.

“And that's where the value is. The model makes the experimental research go better and faster.

“And it's not just about making things better and better, but it's finding things that you may not have been able to find by trial and error.

“You just might never have hit that spot that you wanted to hit on the colour spectrum without machine learning.”

One of the key advantages of the new polymer synthesis process is that it is simple and easy to replicate, and uses widely available materials.

Yinyin said: “This work could be inspiring for the polymer chemistry field, and for people working on the stimuli responsive materials.”

By using the data collected from the latest experiment, an even more comprehensive machine learning model could be trained to provide new insights in the future.

“People who develop new materials might also think of using machine learning models trained on existing data to explore more possibilities,” Suiying said.