Recently, Associate Professor Hailong Fan, a newly appointed member of the Hundred Talents Program at the College of Chemistry and Environmental Engineering, together with Professor Jianping Gong of Hokkaido University and collaborators, achieved a significant breakthrough in the field of hydrogels. Drawing on a comprehensive protein database, the team innovatively proposed a “three-in-one” design strategy that integrates data mining, biomimetic experimental design, and machine learning. Using this approach, they successfully predicted and developed super-adhesive hydrogels with underwater adhesion strengths reaching the megapascal (MPa) level. This achievement demonstrates a complete materials design pathway transitioning from biomimetic experience to data-driven approaches. The results were published in Nature on October 6, 2025, and featured on the journal’s cover. Associate Professor Hailong Fan is one of the corresponding authors.

Designing soft materials such as hydrogels and elastomers is a complex challenge. It requires selecting the right structural units (e.g., monomers) and determining how they are arranged within the material. The resulting design space contains nearly infinite possible combinations. Further complicating matters, the interplay of weak molecular interactions and thermal fluctuations produces intricate multiscale behaviors, where mesoscale structures play a critical role in linking structure to performance. While data-driven methods have transformed the discovery of hard materials with well-defined atomic structures—enabling precise performance prediction and efficient exploration—applying similar approaches to soft materials has remained challenging due to their inherently complex structure–property relationships.

In this study, the researchers proposed a fully data-driven strategy that combines data mining, experimental validation, and machine learning to design high-performance underwater adhesive hydrogels from scratch. By mining a protein sequence database, they developed a descriptor-based method that statistically reproduces protein sequence motifs in synthetic polymer chains via ideal copolymerization. This enabled the targeted design of hydrogels and the creation of a standardized dataset. Based on an initial dataset of 180 bio-inspired hydrogel formulations, machine learning optimization led to a significant improvement in adhesion strength, with the best-performing sample surpassing 1 MPa.

These ultra-strong adhesive hydrogels hold great potential in fields ranging from biomedical engineering to deep-sea exploration, marking a significant milestone in data-driven innovation for soft materials.

Link to paper: https://www.nature.com/articles/s41586-025-09269-4