Machine Learning-Guided Development of Anti-Tuberculosis Dry Powder for Inhalation Prepared by Co-Spray Drying.

Area of Science:

• Pharmaceutical Sciences
• Drug Delivery Systems
• Computational Chemistry

Background:

• Tuberculosis (TB) remains a significant global health challenge.
• Current anti-TB drug administration methods have limitations, including systemic side effects and poor lung distribution.
• Dry powder inhalers (DPIs) offer a promising alternative for targeted lung delivery.

Purpose of the Study:

• To investigate the aerodynamic performance of co-spray-dried DPIs containing rifampin or pyrazinamide and amino acids.
• To develop and utilize machine learning (ML) models for predicting DPI performance.
• To accelerate the development of inhalable anti-TB drug formulations.

Main Methods:

• Preparation of 72 formulations with varying drug-amino acid combinations, molar ratios, and spray-drying parameters.
• Evaluation of aerodynamic performance using a Next Generation Impactor.
• Development and application of four ML models (including XGBoost) to predict fine particle dose (FPD), fine particle fraction (FPF), mass median aerodynamic diameter (MMAD), and geometric standard deviation (GSD).

Main Results:

• Optimal formulations included rifampin-L-lysine acetate and pyrazinamide-L-leucine, achieving high FPF (73.37%, 87.74%) and optimal MMAD (2.59 µm, 1.88 µm).
• XGBoost model demonstrated excellent predictive performance with R² values up to 0.991.
• Molecular weight and LogP were identified as key features influencing DPI performance predictions.

Conclusions:

• Machine learning is feasible for accelerating the development of inhalable spray-dried anti-TB drugs.
• ML models can accurately predict the aerodynamic performance of DPI formulations.
• This approach facilitates the optimization of drug-amino acid combinations for effective lung delivery.