A structured oral formulation database for machine learning: Uncovering data-informed design strategies to facilitate...

Area of Science:

• Pharmaceutical Sciences
• Computational Pharmaceutics
• Data Science in Drug Development

Background:

• Oral dosage form development traditionally relies on empirical methods.
• The increasing number of poorly soluble drugs necessitates structured data for formulation.
• Existing data is often unstructured, limiting advanced computational approaches.

Purpose of the Study:

• To develop the Computational Pharmaceutics Intelligent Manufacturing Database (CPIMD) for machine learning applications.
• To integrate physicochemical properties, excipient compositions, release categories, and dissolution data.
• To create a structured dataset for data-informed oral formulation development.

Main Methods:

• Constructed CPIMD integrating diverse data from 683 Japanese PMDA-approved oral dosage forms.
• Implemented a standardized workflow for data cleaning, feature encoding, and dissolution profile digitalization.
• Applied unsupervised clustering to identify formulation patterns and a random forest model for release type prediction.

Main Results:

• Identified four major formulation pattern clusters based on drug properties, release categories, and excipient combinations.
• Developed a formulation pattern matrix linking API properties, release objectives, and functional excipients.
• Achieved 97.1% test accuracy in predicting release type using binary excipient features with a random forest model.

Conclusions:

• CPIMD provides a valuable, structured dataset for oral dosage form development.
• The database supports machine learning applications, including predictive modeling and excipient screening.
• CPIMD facilitates a shift towards more data-informed and efficient formulation strategies.