The identification of suitable cell lines for monoclonal antibody (mAb) manufacturing is a complex and resource-intensive process. We have developed an AI-driven approach using machine learning to accelerate clone selection. The model uses early-stage multi-omics data to predict late-stage cell line productivity, enabling early detection of high-producer cell lines. This shift to a model-centric process significantly reduces project timelines, accelerates mAb delivery, and fosters a data-driven culture in bioprocessing.

Digital twins have become a cornerstone in modern bioprocess development, enabling accelerated timelines, deeper process understanding, and potentially robust real-time control. This presentation will showcase industrial examples across monoclonal antibodies, viral vectors, and advanced therapies, highlighting applications from batch to continuous manufacturing. Case studies will demonstrate how tailored modelling, smart experimental design, and real-time deployment can drive end-to-end integration, seamless scale-up, and efficient technology transfer. By addressing both established and emerging modalities, the talk will illustrate how digital twins are moving beyond proof-of-concept to deliver industrial impact, fostering flexibility, resilience, and cost efficiency in next-generation biomanufacturing.

Efficient experimentation is critical in bioprocess development, where time, cost, and complexity constrain traditional Design of Experiments. We introduce a model-based algorithm that leverages Pareto fronts and Bayesian optimisation to balance exploitation of high-performing regions with exploration of uncertain areas. This approach accelerates learning, reduces experimental burden, and scales effectively to high-dimensional design spaces, providing a robust and resource-efficient strategy for modern bioprocess optimisation, while effectively exploring trade-offs among performance and quality objectives.

Broadening access by driving down the cost of manufacture using predictive and adaptive digital systems including the use of AI and AI agentic management systems.

Cell therapies are at a critical juncture with patient demand outstripping current manufacturing capacity. To address this challenge, the Cell and Gene Therapy Catapult has developed cell therapy manufacturing platforms that offer step-changes in throughput for autologous and allogeneic products by integrating end-to-end automation and digitising data flow and decision-making.

This presentation will show how digital shadows can be used to predict CAR T cell concentration based on online nutrient and metabolite data, including dissolved oxygen, glucose, and lactate concentrations. Such predictive modelling is designed to utilise the hard sensors that are already embedded in the bioreactor for process monitoring and control. As no additional process analytical technology is required, digital shadows provide a cost-effective soft sensor of cell concentration.

Post-translational modifications (PTMs) are critical for regulating protein function but challenging to analyse due to their complexity. This project aims to create an automated pipeline for high-throughput PTM screening, leveraging LC-MSE (a data-independent acquisition method), alongside customised data processing tools. The workflow automates the integration of spectral analysis, validation, and filtering of PTM fragments based on peptide attributes—ensuring accurate, reproducible results while reducing false-positive errors and minimising manual interpretation.

N-glycosylation strongly affects monoclonal antibody (mAb) quality and efficacy. We applied machine learning (ML) to predict N-glycan abundances in CHO cell fed-batch cultures under 12 media conditions. From mass spectrometry data, ML models reduced 167 peaks to 18 predictive features. Integrating simulated annealing enabled media design that improved titer while reducing mannosylation, illustrating how ML-guided monitoring and simulation can accelerate process optimisation in mAb biomanufacturing.

Hybrid modelling and machine learning techniques are rapidly advancing in process development to optimise performance and accelerate timelines. These techniques primarily focus on offline, iterative modelling, unlike manufacturing applications that require real-time decision support. The presentation will showcase the application of hybrid modelling and transfer learning from the manufacturing perspective and how advanced analytical techniques like Raman and mass spectrometry can enhance knowledge and improve online process monitoring and control.