Beyond Patterns: The Case for Physics-Grounded AI in Drug Discovery
summary
Claims that physics‑based simulation plus AI predicts binding, potency and safety before synthesis, backed by concrete results: partnership work with Sanofi reduces ~9,000 candidates to 14 for tissue validation; a diagnostic project trims ~20 million designs to ~100 clones and finds 3 one‑shot tetravalent binders; prion work with UCSF uses prion doc to screen 2,000 compounds in a year or 8 million in a week; approach uses Large Quantitative Models (LQMs) integrated with LLMs and deployed by SandboxAQ, funded with close to $1B.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:02LLM limits in drug discovery
- 2:30Physics-grounded AI solution
- 4:13Physics models + LLMs
- 6:00Sandbox AQ team
- 7:44Physics-grounded AI results
- 9:07Tetravalent antibody design
- 10:36Prion disease drug discovery
- 13:05Precision drug discovery vision
- 14:50Democratizing AI access
- 15:11Closing thoughts
related talks
30mAI in Drug Discovery
08mEngineering the Future of Pharma: AI-Driven Drug Discovery
26mQuantum, AI & Health
34mThe True Revolution Lies Ahead of Us
30mMachine Learning: A New Approach to Drug Discovery
36mAlgorithmic Alchemy: Transforming Drug Discovery with Open AI
11mAI Based Patent Search For Chemical Compound & Drug Discovery
10mGenesis Final Round: Nanyang Biologics