AI is the Future of Drug Discovery
Download as PPTX, PDF2 likes520 views
Artificial intelligence (AI) has the potential to transform drug discovery through the use of semantic networks to represent biomedical knowledge as facts and infer new insights, probabilistic rules to evaluate beliefs about candidate drugs, and optimization and simulation techniques to iteratively improve outcomes. By continuously learning from diverse sources of data, AI systems could automate and enhance the processes of target identification, compound screening and testing in ways that accelerate research and development compared to traditional analytical tools.
AI is the Future of Drug Discovery
- 1. Artificial Intelligence is the Future of Drug Discovery David E Leahy Discovery Bus Ltd
- 2. • What is AI? • How do we build AI systems? • Why is this a good thing?
- 3. What is AI? • Software that makes decisions • Ambient • Independent • Declarative
- 4. Elements of AI Facts Rules Beliefs Inference Learning Optimise Simulate Sensors Decisions
- 5. Facts and Inference • mol1 binds protein1 • target involved disease1 • Protein subClass Target • protein1 similar protein2 • mol1 (binds (protein1 similar (protein2(Target(disease2))
- 6. The Semantic Web Bio2RDF (9 Billion Facts): OpenPhacts (3-4 million Facts)
- 7. Belief • mol binds protein • mol active assay • But • multiple measures • multiple predictions • Beliefs are probabilities, Facts are truths (see: Dempster-Shafer Theory for reasoning with uncertainty)
- 9. Belief Networks Structure and dynamics of molecular networks. Pharmacology & Therapeutics 138 (2013) 333–408. Peter Csermely et al Knowledge Vault. A web-scale approach to probabilistic knowledge fusion. K. Murphy et al,
- 10. Rules • Facts: When FactA & FactB & FactC then Decision • Beliefs: Costs, Benefits and Risks of Decision • Decisions • Molecule Facts -> Active • Research Target Profile (RTP) -> Candidate • Project Facts -> Invest
- 11. Search and Optimisation • Search World of Known Facts • Pass Rules? • Search World of Possible Facts • Move through World • Create new beliefs • Pass Rules?
- 12. Simulation • Dynamic simulations of protein structures (eg FEP) • Pharmacokinetic simulation from in vitro ADME • Efficacy and safety simulations
- 13. Elements of AI Facts Rules Beliefs Inference Learning Optimise Simulate Sensors Decisions
- 14. From Tools to AI • Always on, always learning • Generic processes & objective criteria • Multiple lines of evidence • Common Language
- 15. Why AI? • What would AI be like? • What would be different? • What would be better?