Implementing model-based real-time monitoring in biopharmaceutical production represents a key advancement toward Quality by Design and provides the foundation for model-predictive control. Data-driven models are capable of making predictions for monoclonal antibody (mAb), double-stranded DNA, host cell protein, and high molecular weight impurity concentrations during elution from a Protein A chromatography capture step. Ensuring explainability of machine learning models and applying efficient experimental design are key to enhancing modern process control.

We present a physics-grounded binding isotherm for ion-exchange chromatography derived from electrical double layer (EDL) theory. Incorporating nonlinear screening, ion valency, and crowding, the model achieves predictive elution from minimal calibration. Its parameters map directly to physical properties, enhancing interpretability over existing models. Calibrated on a single dataset, we validate predictions across gradients, step elutions, salt types, and load densities with model proteins, and highlight the effect of divalent ions.

Optimising sandwich ELISA performance necessitates the selection of compatible mAb pairs as capture and detection antibodies. In this study, we investigated the antibody-antibody interactions that contribute to diminished signal and elevated background during therapeutic antibody detection in sandwich ELISA. We identified signal loss due to overlapping binding sites, and elevated background from cross-reactivity between antibodies. Additionally, interactions at distinct epitopes were found to induce varying steric hindrance, affecting oligomerisation.

In silico computations play an increasing role in drug development, but platforms combining multiple models and comprehensively evaluating therapeutic proteins' developability are currently lacking. This presentation covers this gap, showing how different models, spanning structure prediction, bioinformatics, machine learning, and molecular dynamics, can be combined within an automated platform to speed-up and de-risk candidate selection, lead characterisation, and formulation development. Relevant case studies will be presented.

Developability Assessment (DAS) is vital in drug development, enabling early risk identification in lead candidates. With growing biotherapeutic diversity and advanced protein engineering, DAS must adapt case-by-case. Integrating automation, data management, and machine learning reduces this burden and opens new pathways for efficient assessment—key themes to be explored in the upcoming presentation.

Highly potent novel biologics require extensive characterisation and formulation approaches to optimise their stability. This presentation will highlight approaches taken in formulation development to optimise the stability of such molecules. Another unique challenge of highly potent molecules is their administration. We will present an approach that demonstrates effective dosing of these highly sensitive molecules.

Bioprocess control involves managing non-linear, time-evolving cell populations, unlike traditional chemical or pharmaceutical processes. This complexity requires adaptive strategies supported by multivariate monitoring and historical data to inform control actions. In this work, we propose a novel data-driven control scheme for fed-batch upstream operations, using graph theory to identify dynamic control parameters beyond conventional ones. Machine learning and optimisation enable real-time recalibration and reactive control based on historical insights. This closed-loop, multi-attribute model ensures cultures follow a desired trajectory, effectively addressing variability and enhancing control across diverse conditions, cell types, and products where static strategies may fall short.

During the course of a bioprocess strain, physiology is likely to change. By natural selection, the performance of the cultivation can be positively affected—or more often—during the production of recombinant proteins the culture is degenerated and loses its metabolic capabilities. Under usage of models this changes can be foreseen and preventative control actions can occur to stabilize the cultivation. It will be shown how to measure in real-time adaption and degeneration in E. coli and C. glutamicum cultivations and how to use this information to optimise and stabilise the process?

Hybrid modeling workflows are transforming continuous microbial bioprocess development. In this presentation, we extend our previous case study to multiple products, integrating a two-stage bioreactor with inline lysis, membrane filtration, and multi-column chromatography. By leveraging mechanistic knowledge and machine learning, we demonstrate how such workflows accelerate development, reduce experimental load, and enable rapid transition to real-time control.

Raman spectroscopy is a powerful tool in process analytical technology (PAT) for bioprocess monitoring. However, developing accurate models is slow and costly. We propose an innovative strategy, where pure spectral fingerprints of target analytes are in silico added to calibration datasets. This approach eliminates extensive physical experiments, accelerates workflow, and improves adaptability across diverse bioprocess conditions. The method has potential to significantly streamline PAT model development while maintaining predictive performance.

In previous work, we demonstrated the benefits of harvesting oncolytic viruses during production e.g. using tangential flow depth filtration (TFDF). This approach enables integrated clarification within the upstream process, eliminating a downstream unit operation and enhancing both viral titers and productivity. Building on this, we present our results on high-cell-density perfusion cultures (>20 × 10^6 cells/mL) of HEK293 and avian cell lines (EB66 and CCE.X10) for the production of three different oncolytic viruses: Newcastle disease virus (NDV), herpes simplex virus type 1 (HSV-1), and a recombinant vesicular stomatitis virus–Newcastle disease virus chimera (VSV-NDV). We discuss virus-specific challenges encountered during intensification and strategies to maximise productivity and product quality.

The increasingly higher demand of viral vectors for gene therapy, such as rAAV, makes current state-of-the-art of biomanufacturing (batch, fed-batch) incapable of satisfying such needs. Continuous processes have been shaping the trend of biomanufacturing evolution, but so far have been modestly applied for rAAV production. Here, a continuous multi-stage bioreactor process was designed to produce recombinant adeno-associated viruses (rAAV) using the HeLaS3 producer cell line (PCL) infected with wild-type adenovirus type 5 (wtAd5). At downstream, cell lysis for rAAV release, DNA digestion, and lysate clarification, were also implemented in continuous mode, towards implementation of an end-to-end continuous rAAV biomanufacturing pipeline.