Guides to AI Drug Development

What is AI drug discovery (AIDD)?

AI-driven drug discovery (AIDD) applies machine learning, generative models, multi-omics data, physics-based computation, and automated experimentation to the drug R&D process. It spans target identification, protein structure prediction, molecule and antibody design, ADMET prediction, synthesis route planning, and increasingly clinical trial design and patient stratification.

Why it matters

Traditional drug development takes 10–15 years and averages over US$2B per drug, with a failure rate above 90%. AI attacks the earliest, most failure-prone stages — turning "broad screening and slow validation" into "data-guided, precision targeting."

The core loop

AIDD is not about a single lucky prediction. It is a closed loop: Design → Make → Test → Analyze (DMTA), repeated with ever-improving models. Insilico's rentosertib went from target discovery to preclinical candidate nomination in about 18 months.

Key AI technologies in drug development

• Target discovery: PandaOmics, knowledge graphs (BenevolentAI), LLM-based scoring to identify and prioritize drug targets.
• Structure prediction: AlphaFold 3 (Isomorphic Labs) predicts protein and molecule interactions.
• Generative chemistry: Chemistry42 (Insilico), Centaur (Exscientia), Nach01 — generate novel molecules optimized for potency, selectivity, ADMET.
• Phenotypic screening at scale: Recursion's automated microscopy + CRISPR + 65PB data.
• Physics + ML hybrids: Schrödinger's LiveDesign platform.
• AI compute: NVIDIA BioNeMo, BioHive-2, Gefion supercomputer.

What the data actually shows

• Preclinical compression is real: target-to-IND in ~12–18 months vs 4–6 years traditionally; AI-designed molecules clear Phase I at 80–90% vs ~52% historically.
• Phase II is the wall: ~40% success — statistically no better than conventional development.
• No full FDA approval yet: as of August 2026, ~175 AI-originated programs have entered human trials, none fully approved.
• Money is flowing: US$11B raised by AI/ML drug discovery companies in 2025 (348 deals), up from US$8.9B in 2024 (264 deals); 2024–2025 partnership and M&A value exceeded US$55B.

How to evaluate the sector

1. Follow late-stage assets: The real test is Phase II/III — track rentosertib (Phase III, IPF) and others.
2. Check platform economics: software revenue, pharma adoption (e.g., 13/20 top pharma using Pharma.AI), and repeat deals.
3. Separate discovery speed from approval: faster candidates ≠ approved drugs.
4. Watch compute partnerships: NVIDIA/Novo Nordisk, Recursion BioHive-2 signal where infrastructure advantage sits.
5. Cross-check market figures: market-size estimates vary widely by research firm (e.g., US$2.35B–4.6B for 2025 depending on scope).