(665d) Using Machine Learning Tools in Bioprocess Scale up When the Number of Batches Is Small

Viktor Konakovsky¹, Graham McCreath¹ and Jarka Glassey²

¹ Fujifilm Diosynth Biotechnologies, Process Design, R&D, Billingham, United Kingdom

² School of Chemical Engineering and Advanced Materials, University of Newcastle, Newcastle upon Tyne, United Kingdom

Presenter and contact person: Viktor Konakovsky

Email: viktor.konakovsky@fujifilm.com, Phone: +44 (0)1642 363511

Modelling efforts in biotechnology historically focused around building and validating data using mechanistic, first principle models. During scale up from laboratory (2L), intermediate (10L), demonstration (200L) to manufacturing scale (2000L) there is limited time to build such models with the required diligence and care. For this reason, in the absence of an established platform process, historical small scale screening runs were analysed with machine learning tools to support scale up activities with exploratory and predictive modelling. However, there is a significant caveat by using such datasets; the data generated in scale down models may not necessarily be predictive enough as scale is changed, and a final validation of the model is not possible before the actual manufacturing scale run. We are aware of these challenges, and will present our approaches as well as recommendations, using a mAb-producing CHO cell process as case study.

Acknowledgements: This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 643056.

[1] Jason Brownlee, Machine Learning With R (2016)