Mining manufacturing data for discovery of high productivity process characteristics

Area of Science:

• Biotechnology
• Process Engineering
• Data Science

Background:

• Modern biopharmaceutical manufacturing relies on automated systems generating vast amounts of process data.
• Understanding complex bioprocesses and ensuring production robustness requires effective data analysis.

Purpose of the Study:

• To develop predictive models for bioprocess performance using historical manufacturing data.
• To identify critical process parameters influencing bioproduct yield and quality.

Main Methods:

• Utilized a kernel-based approach with support vector regression on cell culture data from 108 production runs.
• Integrated over one hundred on-line and off-line temporal parameters for model development.
• Ranked process parameters based on their predictive relevance for key performance indicators.

Main Results:

• Developed predictive models capable of forecasting production performance days before harvest.
• Identified strong associations between specific temporal parameters and final process outcomes.
• Uncovered key parameters crucial for enhancing bioprocess robustness.

Conclusions:

• Model-based data mining offers a data-driven approach for knowledge discovery in bioprocesses.
• Implementation can facilitate real-time decision-making to improve large-scale biomanufacturing.
• This methodology enhances the reliability and efficiency of biopharmaceutical production.