A Data-Driven Workflow for Nanomedicine Optimization Using Active Learning and Automated Experimentation

Abstract

Nanomedicines are an advanced class of drug formulations that hold significant promise, particularly in enhancing the solubility of hydrophobic drugs. However, current state-of-the-art methodologies for developing nanomedicines are often inefficient, limiting both the systematic screening of dosage forms and the fine-tuning of individual formulations. To overcome these challenges, this study introduces a data-driven workflow that integrates active learning with experimental automation to rapidly identify optimal nanoformulations, using aceclofenac as a model drug with poor solubility. The initial formulation design space comprised combinations of the drug with 12 different excipients, resulting in approximately 17 billion possible formulations. To strategically identify the optimal candidates, the active learning-robotic system was first employed to narrow the vast space into a manageable subset. Next, this refined subset was further explored using a design of experiments approach, with selected formulations manually prepared and then subjected to purification processes and characterization techniques that are challenging to automate. Through this workflow, a panel of high-performing lead nanoformulations was identified within a few weeks, demonstrating improved solubility, small and uniform particle size, and stability during storage. These findings highlight the power of combining AI-driven design with automation to accelerate nanomedicine development and lay the groundwork for more efficient formulation development.