The Application of Machine Learning in the Development of Co-Amorphous Dry Powder Inhalation.

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Drug Delivery Systems

Background:

• Co-amorphous systems (COAMS) enhance drug solubility, stability, and pulmonary delivery efficiency for dry powder inhalation (DPI).
• Traditional formulation development relies on laborious trial-and-error, with limited understanding of COAMS' impact on pulmonary deposition.
• A data-driven approach is needed to optimize COAMS development for DPI.

Purpose of the Study:

• To apply machine learning (ML) models for predicting the formation of co-amorphous systems (COAMS) for DPI.
• To evaluate the performance of various ML models in identifying critical molecular features for COAMS formation.
• To experimentally validate the ML-guided development of co-amorphous DPIs.

Main Methods:

• Literature mining was used to construct a database of COAMS.
• Multiple ML models (logistic regression, random forests, XGBoost, LightGBM, SVM) were developed and evaluated using molecular representation.
• SHapley Additive ex Planations (SHAP) analysis identified key molecular features influencing COAMS formation.

Main Results:

• All five ML models achieved satisfactory predictive accuracy (ACC ≈ 0.80) for COAMS formation.
• The LightGBM model demonstrated the highest predictive performance (ACC = 0.845 in the testing subset).
• Experimental validation using salbutamol sulfate and indomethacin showed that ML-guided co-amorphous DPIs achieved favorable aerodynamic performance (fine particle fractions of 41.87%-69.30%).

Conclusions:

• Machine learning provides a feasible and efficient approach to guide the development of co-amorphous systems for DPI.
• This ML-driven strategy can accelerate the optimization of drug formulation for improved pulmonary delivery.
• The study highlights the potential of computational methods to overcome limitations in conventional drug formulation development.