Developing and Deploying Digital & AI Solutions in Biopharma

In the rapidly evolving field of pharmaceutical bioprocess development, the application of hybrid models represents a transformative opportunity to enhance process understanding, while highly improving R&D effectiveness through a significant reduction of experiments and resources. These capabilities are further enhanced by (i) transfer learning, which enables the effective reuse of historical process data—across different products, scales, and equipment, to inform the development of models for new products; and (ii) through Pareto-based Bayesian optimization, which effectively leverages model prediction confidence to enable optimal, risk-informed decisions, such as designing new experiments.

These technologies represent the core of the “DataHow Digital Development Framework”, which closely resembles the well-known PDCA cycle used in lean development and manufacturing. In this context, the novelty is represented by the introduction of machine learning (ML) technologies and methods within an optimal development workflow that systematically builds process knowledge. This is reshaping the scope of models from a tool to understand and closely reproduce mature production processes to a development aid that can be used to learn the features of new processes while providing continuous support for decision-making, from early screening to tech transfer to manufacturing.

The overall outcome of embracing this ML-adapted approach is accelerated learning with less development effort, time, and cost. It also offers enhanced flexibility in all phases of the product life cycle, with more robust decisions and shorter time-to-decision.

Bringing these approaches to life, the talk will cover multiple industrial study cases from different modalities, from mAbs to CGTs, across many different unit operations.