A practical framework for clone selection, media-feed screening, and upstream process parameter optimization for a mAb CHO cell culture process.

This study presents a data-driven workflow for upstream process development in biologics manufacturing, aimed at improving consistency, efficiency, and decision-making using established statistical and machine learning tools. Rather than relying on subjective interpretation, the framework integrates multiple analytical methods in a structured manner to support key process development decisions. For clone selection, K-means clustering was applied and benchmarked with hierarchical clustering to identify top-performing candidates from a large clone pool. By incorporating multiple product quality attributes into the evaluation, the approach improves selection consistency and reduces subjectivity. In the media and feed screening stage, principal component analysis (PCA) was used to explore how specific combinations influence glycosylation patterns, enabling rapid identification of promising conditions for further optimization while reducing experimental burden. For upstream process parameter refinement, a sequential learning strategy based on multi-objective Bayesian optimization (MOBO) was applied to adaptively explore trade-offs among competing quality attributes and titer. This approach enabled more efficient identification of improved operating conditions compared with static experimental designs while minimizing experimental runs. While the methods used are well-established, their structured integration into a cohesive workflow demonstrates practical utility for industrial applications. Aligned with Bioprocessing 4.0 principles, this approach supports improved process understanding and informed decision-making in upstream bioprocess development.