An interactive Bayesian optimization framework for intelligent design of HAMA/GelMA hybrid hydrogels

- Hyaluronic acid methacrylate (HAMA)/gelatin methacrylate (GelMA) hybrid hydrogels are extensively utilized in biomanufacturing and tissue engineering, where their rheological properties are determinants of bioprintability and functional performance. However, optimizing these behaviors remains challenging due to the complex nonlinearity and high-dimensional design space defined by hydrogel concentration and temperature. Compared with previous machine-learning studies on hydrogel systems that primarily perform forward prediction of rheological or mechanical properties, here we introduce an interactive Bayesian optimization (IBO) framework that integrates Bayesian optimization with both an environment model and a discriminative model to optimize concentration–temperature values to achieve a target viscosity. The multilayer perceptron–based environment model here proposed exhibits high predictive performance (R 2 ≥ 0.994, RMSE = 4.68), while the support vector machine–based discriminator achieved F1 > 0.95 and AUC > 0.998 in distinguishing thermosensitive regions. Through feedback-driven iterations, IBO improved efficiency and robustness in targeting viscosity, with its mean value converging from 66.01 ± 8.76 Pa·s to 51.81 ± 4.38 Pa·s across three rounds, reaching a qualified rate of 80%. Even under a constrained HAMA content of 0.40% (w/v), IBO generated near-target viscosities (47.64–49.64 Pa·s). These results collectively demonstrate that IBO can efficiently navigate complex, nonlinear rheological landscapes and reliably converge toward user-defined performance targets with low experimental data cost, while maintaining robustness under practical formulation constraints, thereby enabling efficient and directed formulation design. Overall, IBO provides an efficient, reliable, and scalable paradigm for viscosity-guided formulation design of HAMA/GelMA hybrid hydrogels, with potential applicability to soft matter and polymer systems. These findings can further assist in developing hydrogel formulations with improved printability and performance in biomanufacturing and related biomedical applications. • An interactive Bayesian optimization (IBO) framework was developed to optimize the viscosity of HAMA/GelMA hybrid hydrogels. • A physics-informed MLP and an SVM discriminator enforced non-negative viscosity and flagged thermosensitive regions to stabilize convergence. • Feedback-driven iterations achieved 80% experimental success by the third round and maintained performance under HAMA concentration limits.