Sanofi has been actively using AI to accelerate compound optimization across its discovery programs. Here’s a concise view of how that typically plays out and who they’ve partnered with:
What AI compound optimization means at Sanofi
- Potency and selectivity: Predicting target engagement and off-target effects early to steer designs toward more selective compounds.
- ADME/Tox and PK predictions: Using ML to forecast absorption, distribution, metabolism, excretion, and safety liabilities to de-risk leads sooner.
- Multi-parameter optimization (MPO): Balancing potency, safety, pharmacokinetics, and manufacturability in one design cycle.
- De novo design and generative chemistry: Using generative models to propose novel chemotypes and explore wider chemical space than traditional medicinal chemistry alone.
- Retrosynthesis and synthetic feasibility: Assessing how readily proposed molecules can be made and scaled.
- Virtual screening and docking: Prioritizing candidates with favorable predicted binding and profiles before synthesis.
- Data-driven decision making: Integrating internal Sanofi data with external AI tools to improve predictive power and reduce false positives.
Key partnerships and collaborations
- Exscientia: Publicly announced collaboration aimed at AI-driven drug discovery and lead optimization across multiple programs. This partnership is intended to accelerate design, screening, and progression of candidates.
- Schrödinger: Collaboration leveraging Schrödinger’s computational platform to enhance structure-based design, MPO, and predictive modeling for drug discovery efforts.
- In addition to these partnerships, Sanofi has been building and integrating internal AI capabilities (data science teams, platforms, and workflows) to scale AI methods across discovery and early development.
How it’s typically set up
- Center of Excellence and/ or dedicated AI platforms: An internal team and platform to curate data, deploy ML models, and integrate predictions with medicinal chemistry workflows.
- Data governance and quality: Emphasis on high-quality, well-annotated data to train robust models, with ongoing efforts to improve data curation and reproducibility.
- Cross-functional workflows: Close collaboration between data scientists, computational chemists, medicinal chemists, and project teams to translate predictions into actionable synthesis and screening plans.
- Validation loop: Experimental validation of AI predictions to refine models and reduce uncertainty over time.
What to consider or ask about
- Specific programs: Public press releases announce collaborations, but many program details are proprietary. If you want, I can summarize known public collaborations and their stated goals.
- Tools and methods: Typical tech includes graph neural networks for property prediction, generative models for design, MPO-focused scoring, and retrosynthesis AI. Do you want a deeper dive into any of these methods?
- Output and impact: AI accelerates exploration and decision-making, but experimental validation remains essential. I can outline how teams typically balance speed with scientific rigor.
If you want, tell me what aspect you’re most interested in (partnership specifics, the types of AI methods used, or how to implement a similar program in-house), and I can tailor the details.