Data-Driven Identification of GelMA Hydrogel Formulations with Bioink-Relevant Properties Using Design of Experiments (DoE) and Machine Learning (ML)
4 affiliations
- 1Ho Chi Minh City University of Technology (HCMUT) , 268 Ly Thuong Kiet , Dien Hong Ward, ,
- 2Vietnam National University Ho Chi Minh City , Linh Xuan Ward, ,
- 3Vietnam National University Ho Chi Minh City University of Science (HCMUS) , 227 Nguyen Van Cu , Cho Quan Ward, ,
- 4University of Health Sciences, Vietnam National University HCMC , Dong Hoa Ward, ,
Abstract
Gelatin methacryloyl (GelMA) hydrogels are promising bioink candidates for extrusion-based 3D bioprinting, but their synthesis and formulation conditions must be optimized to achieve suitable gelation, printability, and mechanical stability. In our study, a design-of-experiments (DoE) and machine-learning-guided workflow was developed to optimize GelMA synthesis and identify bioink-relevant formulations. Two synthesis variables, the methacrylic anhydride-to-gelatin ratio (0.05–0.30 mL/g) and reaction time (1–4 h), were investigated in phosphate-buffered saline at pH 7.4 to determine their effects on the degree of substitution (DoS). The DoS increased from 39.72 ± 0.15% to 62.35 ± 0.01% with increasing methacrylic anhydride content and reaction time, confirming the positive influence of both variables. The DoE-derived synthesis outputs were then expanded into a 336-condition formulation data set by incorporating GelMA concentration, UV exposure time, and binary gelation outcome. Ensemble classifiers were used to predict gelation behavior, with random forest and AdaBoost achieving an accuracy of 95.59%, precision of 95.59%, recall of 95.59%, and F1 score of 95.57%; the random forest model further showed the highest ROC-AUC of 0.97. Gaussian process regression coupled with Bayesian optimization showed strong validation performance, with R2 = 0.9676 and RMSE = 0.0863, and was used to prioritize candidate formulations by balancing gelation probability, DoS, and UV exposure time. The top-ranked formulation included an MAA/gelatin ratio of 0.219 mL/g, a reaction time of 3.73 h, a predicted DoS of 62.3%, a GelMA concentration of 16.9% w/v, and a UV cross-linking time of 15.4 s, with a predicted gelation probability of 0.8422 and a composite desirability score of 0.842. Experimental validation showed that the optimized GelMA formulation exhibited pronounced shear-thinning behavior, with viscosity decreasing from approximately 2.5 Pa·s (2500 mPa·s) at very low shear rate to approximately 0.17–0.20 Pa·s (170–200 mPa·s) over the high-shear region up to 500 s–1. The photo-cross-linked hydrogel sustained compressive deformation up to approximately 60–62% strain, reaching a compressive stress of about 0.50 MPa. SEM analysis further revealed an open macroporous architecture with an average pore diameter of 21.5 ± 8.9 μm based on 150 pores. These results demonstrate that an integrated data–model–optimization–experiment strategy can facilitate GelMA formulation development and provide a practical framework for optimizing hydrogel formulations with properties relevant to extrusion-based bioink development. Further biological and printing validation is required to establish their suitability for cell-laden 3D bioprinting and regenerative medicine.
Related
- Effect of Cyclodextrin Type on the Photopolymerization and Network Properties of Oligo(L-lactide)-Extended Poly(ethylene glycol) Formulations
- Spectroscopic identification of gauche and trans conformers in L-phenylalanine Fiber and hydrogel via CH/CD stretching vibrations
- Optimizing mechanical properties and ionic conductivity of hybrid hydrogel electrolytes for flexible zinc-ion batteries
- Graphene Oxide-Regulated Calcium Alginate Molecularly Imprinted Hydrogel Films with Enhanced Physico-Mechanical Properties and Selective Protein Recognition.
- Solution Versus Suspension: How Does the Pharmaceutical Form of Insulin Shape the Mechanical Properties and Release Kinetics of a Carbopol® Ultrez™ 10/Sodium Alginate Hybrid Hydrogel with a Molecular-Docking Rationale for Diffusion-Controlled Delivery
BibTeX
@article{pham2026datadriven,
title = {{Data-Driven Identification of GelMA Hydrogel Formulations with Bioink-Relevant Properties Using Design of Experiments (DoE) and Machine Learning (ML)}},
author = {Tan Thi Pham and Le Thao Vy Huynh and Gia Huy Duong and Nguyen Thien Han Le and Minh Hien Nguyen},
journal = {ACS Omega},
year = {2026},
month = oct,
doi = {10.1021/acsomega.6c06504},
url = {https://pubs.acs.org/acsodf/article/11/42/64973/5444969/Data-Driven-Identification-of-GelMA-Hydrogel},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Data-Driven Identification of GelMA Hydrogel Formulations with Bioink-Relevant Properties Using Design of Experiments (DoE) and Machine Learning (ML)
AU - Tan Thi Pham
AU - Le Thao Vy Huynh
AU - Gia Huy Duong
AU - Nguyen Thien Han Le
AU - Minh Hien Nguyen
PY - 2026
DA - 2026/10/06
JO - ACS Omega
DO - 10.1021/acsomega.6c06504
UR - https://pubs.acs.org/acsodf/article/11/42/64973/5444969/Data-Driven-Identification-of-GelMA-Hydrogel
AB - Gelatin methacryloyl (GelMA) hydrogels are promising bioink candidates for extrusion-based 3D bioprinting, but their synthesis and formulation conditions must be optimized to achieve suitable gelation, printability, and mechanical stability. In our study, a design-of-experiments (DoE) and machine-learning-guided workflow was developed to optimize GelMA synthesis and identify bioink-relevant formulations. Two synthesis variables, the methacrylic anhydride-to-gelatin ratio (0.05–0.30 mL/g) and reaction time (1–4 h), were investigated in phosphate-buffered saline at pH 7.4 to determine their effects on the degree of substitution (DoS). The DoS increased from 39.72 ± 0.15% to 62.35 ± 0.01% with increasing methacrylic anhydride content and reaction time, confirming the positive influence of both variables. The DoE-derived synthesis outputs were then expanded into a 336-condition formulation data set by incorporating GelMA concentration, UV exposure time, and binary gelation outcome. Ensemble classifiers were used to predict gelation behavior, with random forest and AdaBoost achieving an accuracy of 95.59%, precision of 95.59%, recall of 95.59%, and F1 score of 95.57%; the random forest model further showed the highest ROC-AUC of 0.97. Gaussian process regression coupled with Bayesian optimization showed strong validation performance, with R2 = 0.9676 and RMSE = 0.0863, and was used to prioritize candidate formulations by balancing gelation probability, DoS, and UV exposure time. The top-ranked formulation included an MAA/gelatin ratio of 0.219 mL/g, a reaction time of 3.73 h, a predicted DoS of 62.3%, a GelMA concentration of 16.9% w/v, and a UV cross-linking time of 15.4 s, with a predicted gelation probability of 0.8422 and a composite desirability score of 0.842. Experimental validation showed that the optimized GelMA formulation exhibited pronounced shear-thinning behavior, with viscosity decreasing from approximately 2.5 Pa·s (2500 mPa·s) at very low shear rate to approximately 0.17–0.20 Pa·s (170–200 mPa·s) over the high-shear region up to 500 s–1. The photo-cross-linked hydrogel sustained compressive deformation up to approximately 60–62% strain, reaching a compressive stress of about 0.50 MPa. SEM analysis further revealed an open macroporous architecture with an average pore diameter of 21.5 ± 8.9 μm based on 150 pores. These results demonstrate that an integrated data–model–optimization–experiment strategy can facilitate GelMA formulation development and provide a practical framework for optimizing hydrogel formulations with properties relevant to extrusion-based bioink development. Further biological and printing validation is required to establish their suitability for cell-laden 3D bioprinting and regenerative medicine.
SN - 2470-1343
ER -