Interpretable machine learning-driven optimization of physicochemical properties in hydrogel scaffolds to promote macrophage polarization.

The physicochemical properties of hydrogel scaffolds, including storage modulus, loss modulus, swelling ratio, porosity, pore size, and roughness, are capable of promoting macrophage polarization into anti-inflammatory phenotype (M2) to accelerate tissue repair. However, most current studies focus on the effects of individual properties on M2 polarization, examining each in isolation. Enhancing the synergistic effects of multiple physicochemical properties is a challenge. In this work, a novel strategy called interpretable machine learning-driven optimization of physicochemical properties (IML-OPP) is proposed to address this challenge. In the IML-OPP strategy, the optimal value of each physicochemical property was sequentially determined based on its ranked importance. First, an initial value was identified for each property by maximizing the individual effect to promote M2 polarization. Then, these initial values were optimized based on their interactive effects. Once all the optimal values were determined, an optimized combination of physicochemical properties was designed to construct a hydrogel scaffold optimized for promoting M2 polarization. To assess the robustness and universality of the IML-OPP strategy, three optimized combinations of physicochemical properties were generated and evaluated. These results offer theoretical guidance for designing hydrogel scaffolds aimed at promoting M2 polarization.