Predicting Powder Blend Flowability from Individual Constituent Properties Using Machine Learning.

Area of Science:

• Pharmaceutical Sciences
• Materials Science
• Computational Chemistry

Background:

• Predicting powder blend flowability is crucial for efficient pharmaceutical manufacturing.
• Current methods are often challenging, resource-intensive, and lack predictive power.
• Need for advanced computational approaches to optimize blend properties.

Purpose of the Study:

• Develop machine learning (ML) models to predict powder blend flowability across multiple categories.
• Identify critical features influencing blend flowability.
• Guide the design of pharmaceutical formulations with enhanced flow properties.

Main Methods:

• Analyzed a dataset of 410 blends using various active pharmaceutical ingredients and excipients.
• Employed supervised ML models (Random Forest, XGBoost) to predict flowability categories.
• Utilized particle size, morphology, surface properties, and dry coating parameters as predictive features.

Main Results:

• Achieved high prediction accuracy for flowability regimes (up to 87% for blends).
• Dry coating parameters emerged as the most influential features, followed by particle size and morphology.
• ML models successfully identified transitions between flow regimes, aiding blend optimization.

Conclusions:

• Integrated ML models effectively predict powder blend flowability and elucidate feature-property relationships.
• This approach facilitates rational design of blends with improved flow properties.
• Reduces experimental effort in pharmaceutical process and product development.