Greedy design space construction based on regression and latent space extraction for pharmaceutical development

Area of Science:

• Pharmaceutical Manufacturing
• Chemical Engineering
• Regulatory Science

Background:

• The design space (DS) is crucial for ensuring drug product quality and regulatory approval.
• High-dimensional statistical models are used to define the DS, but they lack visualization capabilities for feasible input ranges.
• Existing methods struggle to balance comprehensive process understanding with practical DS visualization.

Purpose of the Study:

• To propose a greedy approach for constructing an extensive and flexible low-dimensional DS.
• To integrate high-dimensional statistical models with observed internal representations for improved DS visualization.
• To satisfy both comprehensive process understanding and DS visualization capabilities for pharmaceutical development.

Main Methods:

• Developed a greedy approach to construct a low-dimensional DS from a high-dimensional statistical model.
• Utilized observed correlation structures to reduce the dimensionality of the DS.
• Fixed non-critical controllable parameters and considered variations in non-critical non-controllable parameters for visualization.

Main Results:

• Successfully constructed a low-dimensional DS that is both extensive and flexible.
• Demonstrated improved visualization of feasible input parameter ranges compared to high-dimensional models.
• The approach effectively balances process understanding with practical DS visualization.

Conclusions:

• The proposed greedy approach is useful for developing pharmaceutical manufacturing processes.
• This method enhances the ability to visualize and understand the design space for drug product quality.
• The study provides a valuable tool for regulatory filings and process optimization in the pharmaceutical industry.