Predicting rheological properties of HAMA/GelMA hybrid hydrogels via machine learning

Abstract

- Rheological properties are pivotal in determining the printability of biomaterials, directly impacting the success of 3D bioprinted constructs. Understanding the intricate relationship between biomaterial formulations, rheological behavior and printability can facilitate the advancement and rapid development of biomaterials. Herein, we critically measured the rheological properties of hyaluronic acid methacrylate (HAMA)/gelatin methacrylate (GelMA) hybrid hydrogels with varied formulations and generated a dataset to train a machine learning (ML) model. By utilizing four well-known algorithms, we developed the ML model for the viscosity and shear stress of HAMA/GelMA hydrogel mixtures. To improve model interpretability, we further created a multilayer perceptron framed model, known as HydroThermoMLP, by incorporating the Redlich-Kister polynomial as the thermodynamic representation of viscosity of mixtures. To accomplish the MLP learning on limited data, the shared loss function was formulated on the basis of the R-K presentation to guide the joint training process. The established HydroThermoMLP model, while maintaining the same accuracy as Random Forest, produces outputs that adhere to thermodynamic constraints and instill confidence in generalization applications with a simple algorithm informed by the R-K polynomial. It presents a robust predictive ML tool to forecast the viscosity of hybrid hydrogels and direct the design of biomaterials while appropriately abiding by thermodynamic constraints as essential guidelines.