Predicting Spray Dried Dispersion Particle Size Via Machine Learning Regression Methods.

Area of Science:

• Pharmaceutical Sciences
• Chemical Engineering
• Data Science

Background:

• Spray-dried dispersion particle size is crucial for drug bioavailability and manufacturability.
• Current methods require extensive experimentation, often limited to a single active pharmaceutical ingredient (API).

Purpose of the Study:

• To develop an API-independent predictive model for spray-dried dispersion particle size.
• To establish a strategy for defining initial process parameters based on formulation and target particle size.
• To reduce experimental burden in early-stage process development and scale-up.

Main Methods:

• An ensemble machine learning model was developed to predict particle size across pilot and commercial scales.
• Shapley additive explanations (SHAP) were used to understand parameter influence on predictions.
• An optimization strategy was implemented using the predictive model to estimate process parameters.

Main Results:

• The model achieved prediction errors between -7.7% and 18.6% (25th/75th percentiles) on a hold-out set.
• SHAP analysis confirmed mechanistic understanding of particle formation.
• The optimization strategy successfully estimated process parameters for scale-up.

Conclusions:

• This API-independent predictive modeling approach significantly reduces experimental effort for spray-dried dispersion development.
• The developed in-silico design space facilitates efficient process development and scale-up for new molecules.