AI in drug discovery has attracted billions in investment and generated endless hype. Claims of revolutionizing pharmaceutical R&D are common.

But what’s actually happening? Which AI approaches are creating real value? And where are we in the long journey from molecules to medicines?

The Promise

Why AI matters for drug discovery:

Speed: Drug development takes 10-15 years. AI could accelerate early stages significantly.

Cost: Average drug costs $2+ billion to develop. AI could reduce failed candidates.

Success rate: 90%+ of candidates fail clinical trials. Better predictions could improve odds.

Novel targets: AI might identify drug targets humans would miss.

Personalization: AI could enable more targeted, personalized treatments.

The potential is genuine—if AI delivers on its promises.

What’s Actually Working

Current AI applications with evidence:

Target identification: AI analyzing biological data to identify promising drug targets. Multiple companies showing results.

Molecule design: AI generating and optimizing molecular structures. Novel molecules now in clinical trials.

Protein structure prediction: AlphaFold and successors transforming understanding of protein biology.

Clinical trial optimization: Patient selection, endpoint prediction, and trial design improvement.

Drug repurposing: Identifying new uses for existing drugs faster than traditional methods.

Literature mining: Extracting insights from vast scientific literature.

Companies Showing Progress

Who’s demonstrating real capabilities:

Insilico Medicine: AI-discovered drug in Phase 2 clinical trials. End-to-end AI-first approach.

Recursion: Large-scale biological data and AI for drug discovery. Multiple programs advancing.

Isomorphic Labs: DeepMind’s drug discovery spinout. Early but well-resourced.

Exscientia: AI-designed molecules in clinical trials. Acquisition by Recursion validates approach.

AbCellera: AI for antibody discovery. Commercial products reaching market.

Traditional pharma: Roche, Novartis, AstraZeneca, and others building AI capabilities.

What’s Not Working Yet

Where AI struggles:

Clinical success prediction: AI hasn’t yet reliably predicted which candidates will succeed in human trials.

Complex biology: Multi-target, multi-pathway drugs remain challenging to optimize.

Rare diseases: Limited data hampers AI approaches.

Novel mechanisms: AI excels at optimizing known approaches; breakthrough innovation is harder.

Integration with biology: Computational predictions often fail when tested experimentally.

The gap between molecular design and clinical success remains wide.

The Pipeline Reality

AI drug candidates by development stage (approximate 2025 data):

• Preclinical: Hundreds of AI-designed candidates
• Phase 1: Dozens of programs
• Phase 2: Approximately 10-20 programs
• Phase 3: A handful
• Approved: None yet purely AI-designed

The first AI-designed drugs reaching market is likely 2026-2028, with significant uncertainty.

What’s Needed

For AI to transform drug discovery:

Better biology integration: Closer connection between computational and experimental work.

Clinical outcome data: More data on what actually works in humans, not just in silico.

Multi-modal approaches: Combining molecular, clinical, and real-world data.

Regulatory adaptation: Frameworks for evaluating AI-designed drugs appropriately.

Talent integration: Teams that combine AI expertise with biological and clinical knowledge.

Realistic expectations: Acceptance that AI accelerates rather than replaces traditional processes.

Investment Considerations

For investors and strategists:

Long timelines: Drug development takes years regardless of AI involvement.

Validation requirements: Regulatory approval is the only validation that matters.

Platform vs. pipeline: Some value AI platforms; others specific drug candidates.

Pharma partnerships: External validation through licensing deals with major pharma.

Technical due diligence: Evaluating AI claims requires scientific sophistication.

Portfolio approach: Most individual candidates fail; diversification essential.

What’s Coming

AI drug discovery evolution:

First approvals: AI-designed drugs likely reaching market within 3-5 years.

Capability expansion: AI tackling more complex drug types (biologics, gene therapies).

Integration maturation: AI becoming embedded in standard pharmaceutical R&D.

Cost validation: Evidence on whether AI actually reduces development costs.

Personalization progress: AI enabling more targeted, individualized treatments.

The Bottom Line

AI is making real progress in drug discovery—but slower and more incremental than hype suggests.

The technology works for specific applications: target identification, molecule optimization, clinical trial design. Whether it will fundamentally transform drug development remains unproven.

For now, AI is a valuable tool within pharmaceutical R&D, not a replacement for the complex, expensive process of developing safe and effective medicines.

---

Tracking the actual progress of AI in pharmaceutical research and development.