Self-Driving Development of Perfusion Processes for Monoclonal Antibody Production.

Area of Science:

• Biotechnology
• Bioprocess Engineering
• Computational Biology

Background:

• Biopharmaceutical development requires significant time and resources, particularly in optimizing upstream conditions for monoclonal antibody production.
• Current high-throughput mini-bioreactor systems lack advanced computational tools for integrated experimental planning and autonomous execution.
• Machine learning integration in later-stage bioprocess development remains underdeveloped.

Purpose of the Study:

• To develop an autonomous experimental machine for bioprocess development.
• To integrate Bayesian experimental design (BED) with a cognitive digital twin for bioprocess optimization.
• To accelerate the development and transfer of processes for clinical material generation.

Main Methods:

• Developed an integrated software framework combining a Bayesian experimental design (BED) algorithm and a cognitive digital twin.
• Digitally linked the software framework to a 24-parallel mini-bioreactor perfusion platform.
• Created an autonomous experimental machine capable of knowledge embedding, real-time learning, prediction, and autonomous operation.

Main Results:

• Demonstrated an autonomous experimental machine for bioprocess optimization.
• Successfully achieved challenging cultivation goals, including increased viable cell volume (VCV) and maximized viability.
• Operated a 27-day cultivation, with the autonomous agent managing operations for 20 days.

Conclusions:

• The developed autonomous system significantly enhances bioprocess development efficiency.
• Autonomous agents integrated with digital twins and BED algorithms can overcome limitations in current bioprocess optimization.
• This approach promises to decrease the time and resources needed for biopharmaceutical process development and clinical material generation.