Machine learning-guided design of cooperative multi-size hydrogel microspheres for osteoarthritis therapy

Osteoarthritis (OA) is a multifactorial joint disease propelled by interrelated pathological processes including synovial inflammation, failure of joint lubrication, and impaired cartilage regeneration. Hydrogel microspheres have demonstrated potential in the treatment of OA. However, previous studies primarily concentrated on the functionalization of hydrogel microspheres, while overlooking the influence of size effects on OA repair. Considering the intricate multicellular characteristic of the joint microenvironment, the regulatory function of particle size on biological responses differs, thus making a quantitative design strategy necessary. Here, we present a size-engineered and machine learning (ML)-guided strategy for OA therapy based on cooperatively engineered multi-size HA hydrogel microspheres. We further elucidate the relationships between microsphere size and their corresponding biological functions. In vitro and in vivo results revealed clear functional specialization: small microspheres enhanced interfacial lubrication, medium-sized microspheres exhibited optimal immunomodulatory activity, and large microspheres preferentially supported chondrocyte proliferation and extracellular matrix (ECM) production. To quantitatively optimize their cooperative effects, a deep kernel learning Gaussian process (DKL-GP) model was employed for Bayesian optimization of microsphere composition. The ML-optimized formulation significantly outperformed uniform-ratio mixtures and single-size controls in vivo, promoting cartilage protection and favorable immune microenvironment remodeling. Collectively, this work establishes a size-engineered hydrogel microsphere platform for OA therapy and introduces a generalizable paradigm for designing injectable biomaterials tailored to multiscale degenerative diseases.