Precise Control of Drug Release in Machine Learning-Designed Antibody-Eluting Implants for Postoperative Scarring Inhibition in Glaucoma.

Abstract

We have developed a scalable, smart subconjunctival micro-cylindrical implant system composed of polycaprolactone (PCL) and polyethylene glycol (PEG) for sustained delivery of basic fibroblast growth factor monoclonal antibody (FGFb mAb), a promising anti-fibrotic agent. This delivery platform allows precise prediction of drug release profiles through machine learning analysis based on key fabrication parameters, such as polymer composition, implant dimension, and drug content. Among the tested machine learning algorithms, LightGBM outperforms others in predicting drug release kinetics (R² = 0.9000 ± 0.0058). Besides, this model effectively elucidates the combined roles of various implant parameters in controlling release behavior and provides insights to guide formulation selection to maximize drug release. The optimized PCL-PEG implant, combined with 0.1% w/v poly (lactic-co-glycolic acid) (PLGA) bio-coating, exhibits sustained antibody release through synergistic diffusion-degradation kinetics. In vitro evaluation demonstrates the PCL-PEG/FGFb mAb@PLGA implant's ability to effectively inhibit 3D-fibroblast-mediated collagen contraction and fibrotic genes. In vivo studies in a rat GFS model, along with histological analysis, further validate the implant's efficacy to reduce fibrosis markers, highlighting the implant's potential to modulate wound healing and to prevent postoperative fibrosis. Finally, the PCL-PEG/FGFb mAb@PLGA implant demonstrates no significant toxicity and good biocompatibility both in vitro and in vivo.