Self-Driving Development of Perfusion Processes for Monoclonal Antibody Production

Abstract

It is essential to increase the number of autonomous agents bioprocess development for biopharma innovation to shorten time and resource utilization in the path from product to process. While robotics and machine learning have significantly accelerated drug discovery and initial screening, the later stages of development have seen improvement only in the experimental automation but lack advanced computational tools for experimental planning and execution. For instance, during development of new monoclonal antibodies, the search for optimal upstream conditions (feeding strategy, pH, temperature, media composition, etc.) is often performed in highly advanced high-throughput (HT) mini-bioreactor systems. However, the integration of machine learning tools for experiment design and operation in these systems remains underdeveloped. In this study, we introduce an integrated framework composed by a Bayesian experimental design algorithm, a cognitive digital twin of the cultivation system, and an advanced 24 parallel mini-bioreactor perfusion experimental setup. The result is an autonomous experimental machine capable of 1. embedding existing process knowledge, 2. learning during experimentation, 3. Using information from similar processes, 4. Notifying events in the near future, and 5. Autonomously operating the parallel cultivation setup to reach challenging objectives. As a proof of concept, we present experimental results of 27 days long cultivations operated by an autonomous software agent reaching challenging goals as are increasing the VCV and maximizing the viability of the cultivation up to its end.