• driving efficiency and productivity in healthcare systems
• integrating and activating unstructured and disparate data
• enhancing insights and patient outcomes
• empowering patients, carers and healthcare professionals

The manufacturing of large molecule drug substances is inherently complex, resulting in significant variability in the Active Pharmaceutical Ingredients (API) produced. Often, the causes of this variability are unknown, posing challenges to reproducibility, supply chain management, raw material consumption, equipment usage, and planning overheads. To address these issues, AI and machine learning (ML) are employed to gain insights into variability and mitigate it, aiming to produce optimal “golden batches.” Increased API production (measured in kilograms) enhances business value by lowering the Cost of Goods Sold (COGS), conserving raw materials, streamlining the supply chain, reducing dependence on external manufacturing, and enabling quicker root cause analysis during manufacturing deviations. Additionally, AI/ML applications provide shop floor personnel with guided recommendations to ensure optimal decision-making for achieving golden batches.

Moderator: Enkelejda Miho, Switzerland

Moderator: Leila T. Alexander, Switzerland

The recent Nobel Prize in Chemistry highlighted the impact of artificial intelligence (AI) in protein design. Yet, antibody design, a subset of the protein design, remains a large challenge. The success rate of AI namely AlphaFold-Multimer version 2 (AF2) for antibody-antigen complexes has been benchmarked at 20%-30%, AlphaFold version 3 (AF3) improved the success rate to 40%. In contrast, the success rates for protein-protein complexes peak at 75% for AF2 and upward 80% for AF3. We discuss herein opportunities and challenges in advancing the frontier of protein design for the design of antibody therapeutics. We highlight the need for data in terms of scale and protein-protein interaction class (disordered vs ordered). The utility of deep learning in capturing the latent embedding space to describe long-range interaction, biophysical properties and downstream prediction task. Finally, the value in combining deep and shallow learning to operationalize AI for antibody design.

Carine Poussin, AI for Life Sciences, CSEM, Switzerland

Kyaw Zin Thann, Myanmar Country Program, PATH, Myanmar

Michele Svanera, Centre for Cognitive Neuroimaging, University of Glasgow, United Kingdom

Dalit Porat Ben Amy, Tzafon Medical Center, Bar Ilan University, CatAI – Oralyze, Israel