What is AI Drug Discovery?
TL;DR
AI that integrates target identification, lead-compound screening, optimization, preclinical work and clinical trial design. Insilico/Recursion/Isomorphic Labs/AlphaFold 3/Schrödinger lead — discovery 12 yr → 6 yr, cost $2.6B → $800M.
AI Drug Discovery: Definition & Explanation
AI Drug Discovery integrates target identification, lead-compound discovery, molecule optimization, ADMET prediction, preclinical research, clinical-trial design and real-world evidence (RWE) analysis — a $5B+ global market in 2026 (+50% YoY), now ~25% of pharma R&D spend. Leading platforms: (1) Insilico Medicine (INS018_055 for IPF in Phase II, industry-first AI-designed candidate, HK + US listed, Sanofi $1.2B / Exelixis $80M deals), (2) Recursion Pharmaceuticals ($3B valuation, NYSE-listed, Roche $150M, Phenomics + AI from cell imaging, Exscientia merger 2024), (3) Isomorphic Labs (DeepMind spin-out, AlphaFold 3 commercialization, $3B Eli Lilly / Novartis deals, $3B funding), (4) Exscientia (first AI-designed Phase I 2020, BMS $1.2B), (5) BenevolentAI (AstraZeneca / Novartis partnerships, knowledge graph + LLM, ALS Phase II), (6) Atomwise (AI docking, 5M compounds/wk, Bayer / BridgeBio, $100K/yr+), (7) Schrödinger (physics + AI, $2B market cap, SLB-2 in Phase III, industry standard, $50K/yr+), (8) Iktos (France, generative chemistry, Servier / Sanofi), (9) Tempus Labs (oncology + AI, $8B valuation, Pfizer / Merck), (10) Saama Technologies (clinical AI, Pfizer COVID -50% time). Tech stack: AlphaFold 3 (joint protein + nucleic acid + ligand structure prediction) + generative chemistry (VAE / diffusion models for novel molecules) + Graph Neural Networks (molecular graph learning) + Knowledge Graph (literature + genes + diseases) + Phenomics (cell imaging AI) + federated learning (Owkin: train without exporting hospital data) + foundation models for biology (Meta ESMFold / Google ProtTrans / Salesforce ProGen). Use cases: (I) academia (~$10K/yr: AlphaFold 3 + ColabFold + Schrödinger Academic, -50% paper-writing), (II) seed biotech ($200K/yr: Schrödinger + Atomwise partnership + AlphaFold, lead in 6 mo, accelerates Series A), (III) mid-cap pharma ($2.2M/yr, 3x pipeline, -50% time-to-Phase-I), (IV) Big Pharma ($1B+/yr ~20-25% of R&D, 12 yr → 6 yr), (V) CRO ($1.1M/yr, -90% trial-data processing). Outcomes: discovery 12 yr → 6 yr, cost $2.6B → $800M (-70%), lead time 2 yr → 3 mo (Atomwise), PCC in 46 days (Insilico INS018_055 vs ~2-yr industry avg), Phase III control arms -25% (Unlearn.AI Digital Twin), -90% data cleaning (Pfizer COVID via Saama). Cautions: (1) clinical attrition stays high (Exscientia DSP-1181 OCD halted in Phase I — AI ≠ clinical success; triangulate with wet lab + experts), (2) FDA AI/ML guidance (2024) requires model transparency / validation / continual-learning controls + audit trails, (3) out-of-distribution (AlphaFold 3 weak on IDPs / novel families — multi-model + wet-lab confirmation), (4) IP / patent risk (Thaler v. Vidal: AI alone can't be inventor — document human chemist contributions), (5) HIPAA / GDPR risk (genomic + EHR can't train external AI — de-identification, federated learning, HIPAA / GDPR-compliant tools, IRB approvals). 2026 trends: AlphaFold 3 commercial rollout (Isomorphic Labs disrupts industry, $3B Eli Lilly / Novartis deals, lead time -90%), production-grade generative chemistry (Insilico Chemistry42 / Iktos Makya / Schrödinger LiveDesign — ADMET-optimized molecules in seconds), Digital Twin clinical trials (Unlearn.AI / Owkin / Pfizer adopt AI control arms, -25% cost / -30% duration), federated learning for medical data (Owkin + Roche / AstraZeneca consortia, full GDPR / HIPAA), foundation models for biology (Meta ESMFold / Google ProtTrans, the GPT of biology).