Stefania Evoli

As the conference closes, this panel will summarize technical and scientific insights to assess how far the computational and AI/ML space has progressed in biologics design and what must evolve by 2027 to confidently derisk early CMC and formulation decision-making using AI-driven approaches.

• As AI models are increasingly used to generate and prioritize biologics candidates, how much mechanistic or biophysical understanding is needed to advance molecules with confidence, particularly as programs approach early CMC?
• Based on the use cases shared across the last couple of days, where have predictive models clearly enabled better decisions, and where do computational x AI metrics still fall short of delivering scientific or manufacturing confidence?
• At which points do AI-driven insights most often break down between in silico design, wet-lab validation, and early CMC?
• Do we think that confidence scores and model rankings meaningfully reduce experimental burden, or simply reshuffle risk for liabilities like stability, aggregation, viscosity, and immunogenicity?
• How should discovery teams balance speed and exploration with the need for traceability and rigorous scientific justification when progressing biologics candidates?
• Looking to 2027, what evidence will be required for AI to be trusted to prioritize or terminate biologics programs ahead of major early CMC investment?
• Are current limitations in interpretability and confidence temporary, or enduring constraints that will continue to shape AI’s role in biologics R&D?