Temporal Profile Predictions for Cell Culture Using FDE and PLS (2026-EU-30MP-2778)

Managing cell culture operations during bioreactor experiments is demanding, requiring constant adjustment. Coordination with downstream purification and analytics adds complexity, where misalignment drives inefficiency and delays. These challenges are magnified during technology transfer, where scale, equipment, and operational differences introduce additional uncertainty.

To mitigate these challenges, we have developed a “cell culture digital twin” that predicts key results from ongoing bioreactor data, enabling proactive decisions and real-time cross-functional alignment.

Our predictive models capture the dynamics of the bioreactor process and was developed through three main steps: (i) establishing a data foundation using ~2000 structured and quality-checked bioreactor runs from Symphogen's in-house built scientific data management system Mimer; (ii) constructing predictive models for key results in JMP Pro, involving imputation of missing data using non-linear fit, and model building based on Functional Data Explorer (FDE) and partial least squares (PLS); and (iii) applying the resultant predictive models to ongoing experimental data within a Python-based Databricks application, thereby making cell culture predictions immediately available to all users.

Digital twins deliver predictive insights that align teams, streamline operations, and de-risk technology transfer, making a strong case for broader rollout across departments