Automated active learning to optimize hydrogel drug release profiles.

Area of Science:

• Biomaterials Science
• Drug Delivery Systems
• Computational Chemistry

Background:

• Hydrogels offer excellent biocompatibility and tunable release kinetics for therapeutics.
• Optimizing hydrogel formulations for specific drug release profiles is traditionally time-consuming and labor-intensive.

Purpose of the Study:

• To develop an automated, high-throughput, and machine learning (ML)-guided framework for efficient alginate hydrogel formulation optimization.
• To accelerate the design of controlled release systems for sensitive therapeutics.

Main Methods:

• Utilized a liquid handling robot to create a diverse library of 120 alginate hydrogel formulations with bovine serum albumin (BSA).
• Employed Gaussian process regression (GPR) ML models to predict cumulative release profiles over time.
• Applied Shapley additive explanations (SHAP) for feature importance analysis to identify key release kinetic factors.
• Implemented Bayesian optimization and active learning for iterative formulation selection and refinement.

Main Results:

• Identified alginate molecular weight, concentration, and time as critical factors influencing drug release kinetics.
• Achieved near zero-order release profiles through iterative ML-guided optimization.
• Successfully translated optimized formulations to achieve sustained release of chondroitinase ABC single-enzyme nanoparticles (chABC-SENs).

Conclusions:

• Demonstrated a scalable, data-driven strategy for rapid hydrogel formulation optimization.
• Highlighted the significant potential of ML in accelerating the development of advanced controlled release technologies.
• Validated the framework's efficacy for optimizing the release of complex therapeutic payloads like enzyme nanoparticles.