Day One (13th May 2025)Tuesday, 13 May 2025 - CET/CEST (Cent Europe Summer, GMT+2)

In modern biopharmaceutical manufacturing, data analytics and machine learning (DA/ML) play a vital role in improving process efficiency and ensuring high product quality. This presentation will explore methodologies for applying DA/ML tools to analyze comprehensive biomanufacturing datasets, with a focus on monoclonal antibody production. By strategically selecting predictive models tailored to specific data, these methods improve process optimization and accuracy. The integration of batch and time-series data further enhances model precision by capturing batch-to-batch correlations, demonstrating the broad applicability of these approaches across biomanufacturing processes

While AI's transformative power is well recognized across industries, its potential in pharmaceutical bioprocessing remains underexploited, largely due to limited data availability. In 2019, Raissi et al. introduced Physics-Informed Neural Networks (PINNs), creating a new paradigm by integrating deep learning with first-principles physical laws. This innovative approach enables the effective use of AI even in data-scarce environments, presenting a groundbreaking opportunity to revolutionize biopharmaceutical processes by reducing costs and accelerating the time-to-market for new therapies. This presentation will showcase one of the first applications of physics-informed AI in bioprocessing, featuring practical case studies developed in collaboration with industrial partners.

• Biopharma industry needs a workforce proficient in both traditional and advanced technologies.

• Significant skills gap exists in areas like data science, automation, cybersecurity, and sensor technologies.

• Targeted training and upskilling programs are needed to address the gap.

• Collaboration between education and industry is essential for future workforce preparedness.

To date, tedious off-line analytics are performed to monitor the IVT step, which do not allow for immediate action if reaction conditions and/or productivity targets are not met. This issue could be resolved with Raman spectroscopy, which allows for much faster measurements of CPPs & CQAs, enabling operators to make faster decisions to ensure optimal process conditions.

A bioreactor is a complex system goverened by physical, chemical, and biochemical phenomena, fluid dynamics, and process control. A comprehensive mathematical model should include mass balances for the headspace, bubbles, and culture medium, gassing and mixing cascades, feeding, sampling, as well as control loops for pH and dissolved oxygen control. Correlations for mass transfer coefficients and specific power inputs are needed for each bioreactor considered, and a fitting procedure can be used for obtaining the kinetic expressions for the basic cell metabolites.

The developed digital twins can then be used to simulate operation under various conditions, and can support scale-up, transfer, experiment planning, small-scale model development, process characterization, and for troubleshooting.