Unraveling the molecular magic: AI explains the formation of the most stretchable hydrogel
Creators
- 1. Department of Biomedical Engineering, Amirkabir University of Technology, USA
- 2. Department of Applied Science and Technology, Polytechnic University of Turin, USA
-
3.
Marquette University
-
4.
California Institute of Technology
-
5.
University of California, Irvine
Abstract
We synthesize a novel hydrogel with exceptional stretchability, capable of extending up to 260 times its original length. Its synthesis is guided by systematic optimization of key components: ammonium persulfate (APS), methylenebisacrylamide, dimethylacrylamide, and polyethylene oxide (PEO). We hypothesize that this extreme stretchability arises from a unique architecture—termed span networks—in which the primary dimethylacrylamide-based polymer network is cross-linked by methylenebisacrylamide and connected by linear PEO chains capable of undergoing random chain scission. Given the intractability of exhaustively analyzing all possible reaction pathways, we employ an AI-based reaction prediction system to investigate the underlying chemistry. This approach reveals a novel network formation mechanism involving PEO chains that link polymer networks through scission–prone interactions. These predicted mechanisms, including chain scission events between PEO and carboxyl groups, are experimentally validated using Fourier-transform infrared (FTIR) spectroscopy.
Copyright and License
© The Royal Society of Chemistry 2026.
Data Availability
The main dataset used to train the radical reaction predictor is the RMechDB dataset which is accessible at https://deeprxn.ics.uci.edu/rmechdb. The radical predictor models, both single step and pathway search, are available online under the DeepRXN platform at: https://deeprxn.igb.uci.edu/rmechrp. The supplementary information (SI) provides details on the computational and experimental data, including hydrogel classification, radical reaction predictions, decomposition, polymerization, and cross-linking pathways, as well as additional mechanical testing results. See DOI: https://doi.org/10.1039/d5re00389j.
Finally, all experimental data supporting the findings of this study, including mechanical testing (stress--strain, relaxation, and cyclic tests), rheological measurements (axial and shear storage and loss moduli), and FTIR spectroscopy results, are publicly accessible viahttps://doi.org/10.7910/DVN/KSRAZW.
Additional details
Dates
- Accepted
-
2025-10-22Accepted
- Available
-
0225-11-05First published