Machine learning-assisted optimization of injectable silk fibroin hydrogels functionalized with decellularized...

Silk fibroin (SF) hydrogels are promising injectable scaffolds for wound repair, but optimizing their multi-parametric gelation remains challenging due to complex formulation-processing interactions. We developed a bioinspired gelation strategy combining potassium-driven β-sheet nucleation with sonication-mediated conformational activation, guided by factorial design and machine learning. Artificial neural network (NN) and Random Forest (RF) algorithms mapped non-linear relationships between SF concentration, potassium (K⁺) concentration, sonication time, and gelation kinetics. Machine-learning-assisted analysis delineated an experimentally informed operational region associated with complete or near-complete gelation and generally favorable gelation kinetics, which was subsequently used to select formulations for technological characterization. All formulations showed physiological pH, acceptable injectability (extrusion forces <5 N), and sonication-tunable microstructure ranging from compact lamellar to porous sponge-like architectures. To enhance biological functionality, gels were combined with adipose-derived mesenchymal stromal cell-lyosecretome and human skin-derived decellularized extracellular matrix (dECM), with component-level testing supporting their cytocompatibility and identifying blood-interaction profiles that warrant further evaluation in the complete composite formulation. Functional analyses revealed distinct in vitro biological profiles: dECM produced the strongest pro-angiogenic response in the tube formation assay, accompanied by VEGFA upregulation and downregulation of NFKB1, HMGB1, and BAX in fibroblasts. Lyosecretome showed complementary effects, including selective SOD2 upregulation, FGF7 induction, and enhanced fibroblast migration when tested as a soluble component. SF-containing formulations attenuated fibroblast scratch closure relative to controls, indicating a matrix-dependent modulation of cell migration. These in vitro findings define distinct component-associated biological response profiles and provide a basis for subsequent mechanistic and preclinical evaluation.