Machine Learning-Driven Design of Multicomponent Bone Inorganic Matrix Mimicking Scaffolds for Osteogenesis Enhanced...

Bone defects require materials with osteogenic, neurogenic, and angiogenic activity, yet designing such materials within high-dimensional compositional spaces remains challenging. Here, we report a machine learning (ML) driven strategy to accelerate the design of multicomponent mesoporous bioactive glasses (MBG). A dataset of 169 formulations was used to train a Support Vector Machine (SVM) model to classify osteogenic potential into two classes, high (relative ALP activity ≥0.83) and low (relative ALP activity <0.83). The model screened 10 000 candidates and identified 126 as high potential. These 126 candidates were then subjected to experimental mineralization assays using BMSCs, and selected four high-performance candidates (Candidates 2-5) and one lower-performance control (Candidate 1). The corresponding MBG cores (MBG1-MBG5) were incorporated into NGF-loaded silk fibroin shells to fabricate core-shell nanofibrous scaffolds via coaxial electrospinning. Among them, MBG5@SF/NGF exhibited the highest osteogenic activity, promoting BMSCs viability, differentiation, matrix mineralization, and Erk1/2 phosphorylation, along with upregulation of osteogenic and neuro-inductive markers. Transcriptomic sequencing confirmed activation of osteogenic and neurogenic pathways. In mice calvarial defect model, MBG5@SF/NGF reconstructed a neuro-vascular-bone regenerative microenvironment. We conclude that the ML accelerated screening cascade, combined with experimental validation, identified MBG5@SF/NGF as a promising scaffold for functional bone regeneration.