Deep learning is a promising approach for building sequence-to-expression models for strain optimisation. But these need large and costly data that create steep entry barriers for many laboratories. In this talk I will discuss data requirements and how they impact predictive accuracy, alongside training strategies for improved prediction of protein expression in new regions of the sequence space. These results provide guidelines for balancing data cost/quality in predictive strain design.

This presentation will delve into strategies for developing accurate predictive models from limited and yet complex data, the necessary data inputs, and the potential challenges faced during implementation. Based on many industrial use cases attendees will gain insights into how these models can forecast performance, reduce variability, and streamline manufacturing, improving both efficiency and product quality in biologic drug development.

Complex self-driving intelligent experiments for bioprocess development are only possible by integrating all aspects of cell cultivation, analytics and modeling into a comprehensive framework and steered by an effective Workflow Management System. This is realised in the KIWI-biolab and its opportunities for process optimisation and collaboration are demonstrated by a number of developmental projects. The strict implementation of such full automated approaches promotes the application of Findable, Accessible, Interoperable and Reusable (FAIR) principles, allowing researchers to readily share protocols, models, methods and data.

The application of modelling tools in bioprocess development and manufacturing has garnered considerable interest. But what does it really take to develop digital bioprocess twins? This talk dives into key topics such as the business impact of process modelling, experimental design strategies, tailored modelling approaches, accelerated process development, seamless scale-up, and the real-time use of models for monitoring and control. These principles will be demonstrated through several diverse industrial case studies, showcasing the Toolbox capabilities, from accelerated upstream development to gene and cell therapy (G&GT).

While AI's transformative power is well recognized across industries, its potential in pharmaceutical bioprocessing remains underexploited due to limited data. In 2019, Raissi et al. introduced Physics-Informed Neural Networks (PINNs), creating a new paradigm by integrating deep learning with first-principles laws. This method enables the use of AI even with scarce data, presenting a groundbreaking chance to revolutionise biopharmaceutical processes by cutting costs and accelerating the time-to-market for new therapies.

The pharmaceutical industry is focusing on patient convenience and decentralised care, driving the need for high-concentration liquid formulations with stable protein behavior. However, predicting protein behavior at high concentrations—including risks like viscosity or aggregation—remains challenging, especially with complex molecular designs. We present an early screening process using high-throughput assays to assess critical solution parameters and predict developability risks across various molecule formats early in drug discovery.

Molecular properties that impact developability attributes and outcomes comprises of conformational, chemical, colloidal, and other interactions. These attributes are measured using relevant analytical methods to assess the developability/ manufacturability of the molecule in different formulation. Developability assessment of mAbs has been studied and applying this assessment using the right tools to new modalities such AAV will help streamline capsid selection and candidate selection from discovery to development for new modalities

Ensuring the success of novel biotherapeutics requires rigorous developability assessments and formulation strategies. This presentation delves into methods for evaluating the physicochemical properties and stability of new biologic candidates. Learn how early-stage assessments can inform formulation development to improve manufacturability, enhance stability, and address potential challenges before clinical stages, ultimately accelerating the path to market for innovative therapeutics.

The development and formulation of therapeutic antibodies is a highly complex optimisation task requiring significant time and resources. This can be particularly problematic for emerging engineered antibody formats, such as fragments and bispecifics, which can suffer from developability issues. Here we show a Bayesian optimisation method to improve multiple developability properties simultaneously through formulation design.

We will provide an overview of how various PAT and Data-Driven approaches are used in different process units in our Drug Substance Process Development. we will showcase several PAT applications and how we use them to improve productivity and reduce development timelines. Finally, we will address challenges and pitfalls encountered during implementation of these approaches and share our future vision of using these approaches to transform pharmaceutical manufacturing.

Though AI, ML, and other modern algorithmic data analysis techniques are poised to provide a range of benefits across the biopharmaceutical industry, these advances will be particularly impactful to analytical development. This talk will highlight key considerations from a CMC perspective when implementing AI/ML in analytical development. Additional focus will be placed on recent regulatory guidance pertaining to the development and validation of analytical methodologies utilising algorithmic data processing techniques.

Data obtained and processed near real-time can provide information to confer consistent product quality, less product and resource wastage, and increased productivity, irrespective of variations in process, materials, and operating environment. The practical implementation of Process Analytical Technologies (PAT) requires, however, analytical methods offering sufficient speed, selectivity, and sensitivity. An online Liquid Chromatography (LC) setup enabling real-time monitoring of product quality will be presented.

Clinical therapies based on cells have the potential to cure many diseases. However, populations of cells cannot be purified or analyzed as stringently as drugs. By analyzing changes in macromolecular cell composition, Raman microscopy offers a label-free approach to validate both biomanufacturing processes and final cell products. Using Raman microscopy, we have detected early apoptosis, distinguished stem cells from their differentiated progeny, as well as T-cell subtypes and activation states.

In the rapidly growing field of mRNA medicines, real-time process monitoring is essential to ensure efficiency, product integrity, and overall manufacturing robustness. Conventional analytical techniques often require sample preparation, are time-consuming, and may not provide the level of process understanding needed for optimization. This presentation explores the innovative use of Raman spectroscopy as a non-destructive, in-line analytical tool for mRNA manufacturing. A data-driven case study will highlight how Raman spectroscopy enables real-time monitoring of critical process parameters, providing insights into nucleoside dynamics and RNA formation, ultimately enhancing process control and product consistency.