Can machine learning predict drug nanocrystals?

Area of Science:

• Pharmaceutical Nanotechnology
• Computational Chemistry
• Materials Science

Background:

• Nanocrystals enhance dissolution rates of poorly soluble drugs by reducing particle size to the nanoscale.
• Current nanocrystal formulation development relies heavily on experienced personnel and time-consuming trial-and-error methods.
• Developing predictive models for nanocrystal properties is crucial for efficient formulation.

Purpose of the Study:

• To develop and validate machine learning models for predicting nanocrystal particle size and polydispersity index (PDI).
• To assess the performance of machine learning models across different nanocrystal preparation methods.
• To identify key factors influencing nanocrystal properties during preparation.

Main Methods:

• Collected 910 nanocrystal size and 341 PDI data points from ball wet milling (BWM), high-pressure homogenization (HPH), and antisolvent precipitation (ASP) methods.
• Utilized the Light Gradient Boosting Machine (LightGBM) algorithm to construct predictive models.
• Experimentally validated the model's predictive accuracy with newly prepared nanocrystals.

Main Results:

• LightGBM demonstrated strong predictive performance for nanocrystal size and PDI using BWM and HPH methods.
• Predictive performance was poorer for the ASP method, likely due to data quality issues related to reproducibility and stability.
• Identified critical influencing factors: milling time for BWM, cycle index for HPH, and stabilizer concentration for ASP.

Conclusions:

• Machine learning, specifically LightGBM, is effective for predicting nanocrystal properties prepared via BWM and HPH methods.
• The study highlights the limitations of the ASP method for reliable nanocrystal production and prediction.
• This research offers a novel, data-driven approach for optimizing nanotechnology-based drug manufacturing.