Machine learning powers discovery of new cryoprotectants for cold storage

Scientists from the University of Warwick and the University of Manchester have developed a cutting-edge computational framework that enhances the safe freezing of medicines and vaccines.

Treatments such as vaccines, fertility materials, blood donations, and cancer therapies often require rapid freezing to maintain their effectiveness. The molecules used in this process, known as “cryoprotectants,” are crucial to enabling these treatments. In fact, without cryopreservation, such therapies must be deployed immediately, thus limiting their availability for future use.

The breakthrough, published in Nature Communications, enables hundreds of new molecules to be tested virtually using a machine learning-based, data-driven model.

Prof. Gabriele Sosso, who led the research at Warwick, explained, “It’s important to understand that machine learning isn’t a magic solution for every scientific problem. In this work, we used it as one tool among many, and its success came from its synergy with molecular simulations and, most importantly, integration with experimental work.”

This innovative approach represents a significant shift in how cryoprotectants are discovered, replacing the costly and time-consuming trial-and-error methods currently in use.

Importantly, through this work, the research team identified a new molecule capable of preventing ice crystals from growing during freezing. This is key, as ice crystal growth during both freezing and thawing presents a major challenge in cryopreservation. Existing cryoprotectants are effective at protecting cells, but they do not stop ice crystals from forming.

The team developed a computer model that was used to analyze large libraries of chemical compounds, identifying which ones would be most effective as cryoprotectants.

Dr. Matt Warren, the Ph.D. student who spearheaded the project, remarked, “After years of labor-intensive data collection in the lab, it’s incredibly exciting to now have a machine learning model that enables a data-driven approach to predicting cryoprotective activity.

“This is a prime example of how machine learning can accelerate scientific research, reducing the time researchers spend on routine experiments and allowing them to focus on more complex challenges that still require human ingenuity and expertise.”

The team also conducted experiments using blood, demonstrating that the amount of conventional cryoprotectant required for blood storage could be reduced by adding the newly discovered molecules. This development could speed up the post-freezing blood washing process, allowing blood to be transfused more quickly.

These findings have the potential to accelerate the discovery of novel, more efficient cryoprotectants—and may also allow for the repurposing of molecules already known to slow or stop ice growth.

Prof. Matthew Gibson, from The University of Manchester, added, “My team has spent more than a decade studying how ice-binding proteins, found in polar fish, can interact with ice crystals, and we’ve been developing new molecules and materials that mimic their activity. This has been a slow process, but collaborating with Prof. Sosso has revolutionized our approach.

“The results of the computer model were astonishing, identifying active molecules I never would have chosen, even with my years of expertise. This truly demonstrates the power of machine learning.”