The DOI matches its Crossref record; 49 of the 49 references it cites were found in a registry or an index; 18 of the 36 sources its claims cite were checked for support.
Data-driven de novo design of super-adhesive hydrogels
Open accessPublished version of a preprint
Abstract
Data-driven methodologies have transformed the discovery and prediction of hard materials with well-defined atomic structures by leveraging standardized datasets, enabling accurate property predictions and facilitating efficient exploration of design spaces1–3. However, their application to soft materials remains challenging because of complex, multiscale structure–property relationships4–6. Here we present a data-driven approach that integrates data mining, experimentation and machine learning to design high-performance adhesive hydrogels from scratch, tailored for demanding underwater environments. By leveraging protein databases, we developed a descriptor strategy to statistically replicate protein sequence patterns in polymer strands by ideal random copolymerization, enabling targeted hydrogel design and dataset construction. Using machine learning, we optimized hydrogel formulations from an initial dataset of 180 bioinspired hydrogels, achieving remarkable improvements in adhesive strength, with a maximum value exceeding 1 MPa. These super-adhesive hydrogels hold immense potential across diverse applications, from biomedical engineering to deep-sea exploration, marking a notable advancement in data-driven innovation for soft materials. A data-driven approach integrates data mining, experimentation and machine learning to design high-performance adhesive hydrogels from scratch, tailored for demanding underwater environments.
Overview
CoChat Summary
TL;DR
This paper integrates adhesive-protein data mining, hydrogel synthesis, and machine learning to design underwater adhesive hydrogels from scratch. Starting from 180 bioinspired formulations, ML optimization produced hydrogels with maximum underwater adhesive strength exceeding 1 MPa on glass in normal saline.
Key findings
- Mining 24,707 adhesive proteins from 3,822 organisms and selecting the top 200 species yielded 180 unique monomer compositions after removing 20 duplicates; these were used to synthesize DM-driven hydrogels.
- Among the 180 DM-driven hydrogels, 16 showed adhesive strength \(F_a > 100\) kPa and 83 showed \(F_a > 46\) kPa; the best, G-042/G-max, reached 147 kPa.
- ML optimization expanded the dataset to 341 hydrogels; R1-max achieved maximum \(F_a > 1\) MPa on glass in normal saline, maintained robust adhesion over 200 attachment–detachment cycles, and sustained a 1-kg shear load for over 1 year.
- SHAP analysis showed high \(\phi_{BA}\) and \(\phi_{PEA}\) enhance \(F_a\), high \(\phi_{HEA}\), \(\phi_{CBEA}\), and \(\phi_{AAm}\) reduce it, and \(\phi_{ATAC}\) has a dual effect; the top ML hydrogels were composed exclusively of BA, PEA, and ATAC.
Methods
- Adhesive protein data were compiled from NCBI using the keyword “adhesive protein”; consensus sequences for the top 200 species were generated with Clustal Omega, and the 20 canonical amino acids were grouped into six physicochemical classes. Relative compositions were derived from the top five functional-class pairs, accounting for approximately 50% of occurrences.
- Hydrogels were synthesized by one-pot free-radical copolymerization of six functional monomers in DMSO with 0.1 mol% crosslinker and 2.4 M total monomer, then solvent-exchanged to normal saline. Underwater adhesion was screened by tack tests on glass in normal saline using a 10-N loading force and 10-s contact time; lap shear and 180° peeling tests were also used.
- ML used a six-dimensional monomer-proportion feature vector to predict \(F_a\). Nine models were benchmarked; Gaussian process and random forest regression were selected as base models, and batched sequential model-based optimization with expected improvement proposed new formulations. Validation rounds expanded the dataset from 180 to 341 hydrogels.
Limitations
- The paper states that challenges remain because of limitations in monomer diversity, polymer synthesis technologies for controlling monomer sequences at a scale suitable for materials development, and dataset scalability. It also states that the performance plateau after the first ML round was suspected to arise from the functionalities of the adopted monomer species, and that further optimization rounds were not pursued.
- The paper notes that hydrogel performance varies with environmental conditions, underscoring the importance of data consistency in ML optimization.
- Evident from the text: the initial dataset was small (180 hydrogels) and the final dataset remained limited (341 hydrogels); only six functional monomers were used, and the clearest design principle is for adhesion to glass surfaces using BA, PEA, and ATAC.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures















Verification
Claims the paper makes
8 Valid · 2 Not valid · 9 Not checked · 81 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Despite their diversity, these proteins share common sequence patterns that offer valuable insights into designing underwater adhesives22.”
§ Data-driven de novo design of super-adhesive hydrogels > Main
What the sources say
- Disputedcites [22] Lee, B. P., Messersmith, P. B., Israelachvili, J. N. & Waite, J. H. Mussel-inspired adhesives and coatings. Annu. Rev. Mater. Res. 41, 99–132 (2011). sourceExperimental — accuracy not yet measured
Section 3.3.2 (page not specified) says the plaque proteins all contain Dopa and have high isoelectric points, but 'otherwise have widely different sequences'—contrary to the claim that they share common sequence patterns. Section 4.1 supports using these proteins to inspire wet-adhesive design, but not the asserted shared sequence patterns. (LLM, conf=0.88)
Checked by a reasoning model — may be wrong.
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“This finding underscores the important role of ideal random copolymerization of functional monomers (with near-unity reactivity ratios) in achieving the statistical sequence features essential for mimicking protein functions19,27.”
§ Data-driven de novo design of super-adhesive hydrogels > Synthesis of DM-driven hydrogels
What the sources say
- Partly supportedcites [19] Ruan, Z. et al. Population-based heteropolymer design to mimic protein mixtures. Nature 615, 251–258 (2023). sourceExperimental — accuracy not yet measured
The ‘Design and synthesis of RHP ensembles’ section describes statistically sequence-defined heteropolymers designed to match protein segmental features, and the concluding design discussion says segmental sequence information can guide statistical sequence control to replicate protein functions (page numbers unavailable). The paper uses experimentally determined monomer reactivity ratios to simulate sequences, but our copy does not say the ratios are near unity or that ideal random copolymerization is important to achieving those features. (LLM, conf=0.85)
- Partly supportedcites [27] Fan, H. L. et al. Adjacent cationic-aromatic sequences yield strong electrostatic adhesion of hydrogels in seawater. Nat. Commun. 10, 5127 (2019). sourceExperimental — accuracy not yet measured
The Introduction links adjacent cationic–aromatic sequences to a protein-inspired adhesion mechanism, and the Results report equal incorporation of certain monomer pairs. But the Results attribute that outcome to cation–π interactions and matched vinyl heads, not ideal random copolymerization or near-unity reactivity ratios; the paper does not establish that statistical sequences are essential for mimicking protein functions. Our copy provides section labels but no page numbers. (LLM, conf=0.82)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, their application to soft materials remains challenging because of complex, multiscale structure–property relationships4–6.”
§ Data-driven de novo design of super-adhesive hydrogels > Abstract
What the sources say
- Not checked — no full text of the sourcecites [4] Li, F. et al. Design of self-assembly dipeptide hydrogels and machine learning via their chemical features. Proc. Natl Acad. Sci. USA 116, 11259–11264 (2019). source
- Partly supportedcites [5] Xu, T. et al. Accelerating the prediction and discovery of peptide hydrogels with human-in-the-loop. Nat. Commun. 14, 3880 (2023). sourceExperimental — accuracy not yet measured
The Introduction (page unavailable) identifies hydrogels as soft materials and says predicting and designing peptide hydrogels remains challenging. It attributes the difficulty to inefficient design methods and the vast number of peptide sequences, not to complex, multiscale structure–property relationships. It also does not extend that explanation to soft materials generally. (LLM, conf=0.82)
- Partly supportedcites [6] Tamasi, M. J. et al. Machine learning on a robotic platform for the design of polymer-protein hybrids. Adv. Mater. 34, 2201809 (2022). sourceExperimental — accuracy not yet measured
The Introduction (page not provided) says ML-guided copolymer design is limited, particularly by insufficient quality data and laborious synthesis and characterization. That supports the general point that applying data-driven methods to these materials is challenging, but it does not identify complex, multiscale structure–property relationships as the reason or establish the claim for soft materials generally. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Their pairwise reactivity ratios, determined by 1H NMR analysis, were close to unity when copolymerized in the cosolvent dimethyl sulfoxide (DMSO) using free-radical polymerization (Supplementary Fig. 5 and Supplementary Table 3).”
§ Data-driven de novo design of super-adhesive hydrogels > Synthesis of DM-driven hydrogels
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The resulting distributions of monomer block lengths and pairwise frequencies (Fig. 3b,c) closely matched those observed in adhesive proteins (Fig. 2b,c)”
§ Data-driven de novo design of super-adhesive hydrogels > Synthesis of DM-driven hydrogels
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
61 more claims
Create a free account to see all 100 claims and to verify the 2 references still open.
- Probably not a substantial claim. 17
- It may state two separate claims as one. 23
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Data-driven de novo design of super-adhesive hydrogels > Methods > ML methods
“GP_KB, GP_CLmax and GP_CLmin simplified the joint q-EI probability calculation36 by using the GP prediction value as a hypothetical value for selecting the next data points with EI maximums.”
§ Data-driven de novo design of super-adhesive hydrogels > Methods > ML methods
“GP_CLmax: used the maximum Fa (y_max) from the training set as a hypothetical value for selecting the next data points with EI maximums.GP_CLmin: used the minimum Fa (y_min) for selecting the next data points with EI maximums.”
Citations
Cites
Greedy function approximation: A gradient boosting machine
Friedman, J. H. (2001) · The Annals of Statistics · doi:10.1214/aos/1013203451 · Cited by 30,731 · cited 5×
“• RFR-GBM: RFR as the hypothetical value provider and GBM as the EI maximizer.”
InfluentialUMAP: Uniform Manifold Approximation and Projection
McInnes, L., Healy, J., Saul, N., & Großberger, L. (2018) · Journal of Open Source Software · doi:10.21105/joss.00861 · Cited by 9,826 · cited 1×
“4b, using uniform manifold approximation and projection (UMAP) 38 for dimensional reduction (from six to two dimensions).”
Geurts, P., Ernst, D., & Wehenkel, L. (2006) · Machine Learning · doi:10.1007/s10994-006-6226-1 · Cited by 9,300 · cited 6×
“Non-linear models comprised k -nearest neighbours (KNN), kernel ridge regression (KRR), support vector regression (SVR), random forest regression (RFR), gradient boosting regression with XGBoost (XGB), extra trees regression (ETR) and Gaussian process (GP) with a Matérn kernel 32,34 .”
InfluentialLarian Studios Dev Team (2024) · doi:10.1145/3292500.3330701 · Cited by 8,618
Efficient Global Optimization of Expensive Black-Box Functions
Jones, D. R., Schonlau, M., & Welch, W. J. (1998) · Journal of Global Optimization · doi:10.1023/a:1008306431147 · Cited by 8,070 · cited 1×
“These non-linear models encompass non-parametric (KNN), kernel-based (KRR, SVR and GP) and tree-ensemble (RFR, XGB and ETR) approaches, enabling a comprehensive comparison 34,35,47 .”
Biopython: freely available python tools for computational molecular biology and bioinformatics
Cock, P. J. A., Antao, T., Chang, J. T., Chapman, B. A., Cox, C. J., Dalke, A., Friedberg, I., Hamelryck, T., Kauff, F., Wilczynski, B., et al. (2009) · Bioinformatics · doi:10.1093/bioinformatics/btp163 · Cited by 6,080 · cited 1×
“SeqIO interface in BioPython 46 .”
Machine learning for molecular and materials science
Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O., & Walsh, A. (2018) · Nature · doi:10.1038/s41586-018-0337-2 · Cited by 4,968 · cited 1×
“To reduce experimental demands, data-driven strategies are becoming increasingly essential 8,9 .”
Algorithm 778: L-BFGS-B: Fortran subroutines for large-scale bound-constrained optimization
Zhu, C., Byrd, R. H., Lu, P., & Nocedal, J. (1997) · ACM Transactions on Mathematical Software · doi:10.1145/279232.279236 · Cited by 3,462
Sequential Model-Based Optimization for General Algorithm Configuration
Hutter, F., Hoos, H. H., & Leyton-Brown, K. (2011) · Lecture Notes in Computer Science · doi:10.1007/978-3-642-25566-3_40 · Cited by 2,469 · cited 3×
“Among nine ML models benchmarked (Supplementary Tables 5 and 6), Gaussian process (GP) 32 and random forest regression (RFR) 33 emerged as the most effective base models for predicting F a from ϕ i , achieving low test error while minimizing overfitting (Extended Data Fig.”
Clustal Omega for making accurate alignments of many protein sequences
Sievers, F., & Higgins, D. G. (2018) · Protein Science · doi:10.1002/pro.3290 · Cited by 2,137 · cited 3×
“Clustal Omega was executed with the command: .”
Influential
Research works printed but not linked: 2
- Myers, R. H., Montgomery, D. C. & Anderson-Cook, C. M. Response Surface Methodology: Process and Product Optimization Using Designed Experiments (Wiley, 2016).
- González, J., Dai, Z., Hennig, P. & Lawrence, N. Batch Bayesian optimization via local penalization. In Proc. Artificial Intelligence and Statistics 648–657 (PMLR, 2016).
Cited by
Ionic Hydrogel Sensors toward Next‐Gen Personalized Healthcare
Sun, Y., Tian, G., Deng, W., & Yang, W. (2025) · Advanced Materials · doi:10.1002/adma.202509122 · Cited by 82
Malashin, I. P., Martysyuk, D., Тынченко, В. С., Gantimurov, A., Nelyub, V., & Бородулин, А. С. (2025) · Polymers · doi:10.3390/polym17182557 · Cited by 40
Xiang, G., Yin, B., Heidari, B. S., Youssef, G., Gosecka, M., Gosecki, M., Torres, F. G., Wong, S. H. D., & Dodda, J. M. (2025) · Advanced Science · doi:10.1002/advs.202512037 · Cited by 45
Artificial intelligence-enabled hydrogels: innovations and applications
Zhang, S., Wang, H., Liu, F., Su, Y., Han, K., Liu, Y., Guan, F., Liu, H., & Ma, S. (2025) · Journal of Materials Chemistry B · doi:10.1039/d5tb01944c · Cited by 36
Artificial Intelligence Informed Hydrogel Biomaterials in Additive Manufacturing
Zhang, Z., Tao, Z. Z., Du, R., Huo, R., & Zheng, X. (2025) · Gels · doi:10.3390/gels11120981 · Cited by 33
Metadata
- Authors
- Liao, H., Hu, S., Yang, H., Wang, L., Tanaka, S., Takigawa, I., Li, W., Fan, H., & Gong, J. P.
- Affiliations
- Hokkaido University (JP)
- Shenzhen University (CN)
- Corresponding author
- Takigawa, Ichigaku
- Li, Wei
- Fan, Hailong
- Gong, Jian Ping
- ORCID
- Venue
- Nature
- Published
- 2025-08-06
- Volume
- 644
- Issue
- 8075
- Pages
- 89-95
- ISSN
- 0028-0836 (ISSN-L)
- 1476-4687
- PMC
- PMC12328221
- OpenAlex
- W4413020503
- Semantic Scholar
- 83748147a6ed0eb041876b65a0aaf87ab3310470
- Cited by
- 169
- References
- 49
- Citations by year
- 2026: 138 · 2025: 31
- FWCI
- 35.49
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by-nc-nd
- creativecommons.org/licenses/by-nc-nd/4.0 (tdm)
- creativecommons.org/licenses/by-nc-nd/4.0 (vor)
- Locations
- Nature (publishedVersion, cc-by-nc-nd)
- Research Square (acceptedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by-nc-nd)
- PubMed (publishedVersion)
- Institutional Repositories DataBase (IRDB) (submittedVersion)
- Kyoto University Research Information Repository (Kyoto University) (submittedVersion, cc-by-nc-nd)
- Topics
- Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
- Physical Sciences › Materials Science › Biomaterials › Hydrogels: synthesis, properties, applications
- Physical Sciences › Materials Science › Surfaces, Coatings and Films › Polymer Surface Interaction Studies
- Keywords
- bioinspired polymers, protein sequence descriptors, random copolymerization, structure-property relationships, adhesion strength, wet adhesion
- MeSH terms
- Machine Learning, Adhesiveness, Adhesives, Proteins, Hydrogels, Databases, Protein, Data Mining, Polymerization
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Data-driven de novo design of super-adhesive hydrogels. https://app.cochat.ai/papers/10.1038/s41586-025-09269-4
- The paper
- Liao, H., Hu, S., Yang, H., Wang, L., Tanaka, S., Takigawa, I., Li, W., Fan, H., & Gong, J. P. (2025). Data-driven de novo design of super-adhesive hydrogels. Nature. https://doi.org/10.1038/s41586-025-09269-4
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.