Inside Isomorphic Labs: Demis Hassabis's Moonshot in Digital Biology

Introduction

When Demis Hassabis announced Isomorphic Labs in November 2021, the pharmaceutical world took notice. Fresh off AlphaFold's revolutionary success in solving protein folding, Hassabis wasn't content to simply publish papers. Isomorphic Labs represents an ambitious bet: that the same AI-first methodology that conquered protein structure can transform drug discovery itself.

Two years later, with major partnerships with Eli Lilly and Novartis worth up to $3 billion, Isomorphic Labs is proving this bet may pay off.

The Origin Story

From DeepMind to Isomorphic

The Genesis:

• 2020: AlphaFold 2 achieves unprecedented accuracy in CASP14
• 2021: AlphaFold database released with 350,000+ structures
• November 2021: Hassabis announces Isomorphic Labs as Alphabet subsidiary
• Goal: Apply AI directly to drug discovery, not just research tools

The Name: "Isomorphic" refers to mathematical structures that are equivalent in form—a metaphor for finding mappings between biological problems and computational solutions.

The Founding Hypothesis

Hassabis's core insight:

Drug discovery is fundamentally a computational problem—understanding molecular interactions, predicting binding, optimizing properties—all are information processing challenges that AI can solve better than traditional methods.

The Isomorphic Approach

AI-First, Not AI-Assisted

Unlike traditional pharma companies adding AI tools, Isomorphic was built from the ground up around AI:

Traditional Drug Discovery:

1. Choose biological target
2. High-throughput screening of compound libraries
3. Lead optimization through medicinal chemistry
4. Preclinical and clinical testing

Isomorphic's Vision:

1. AI predicts protein structures and interactions
2. Generative models design optimal molecules computationally
3. Computational validation reduces experimental needs
4. Only top candidates synthesized and tested
5. Results feed back to improve models

The Technology Stack

While specifics remain proprietary, Isomorphic likely integrates:

Structure Prediction:

• AlphaFold for protein structures
• Extended AlphaFold 3 for protein-ligand complexes
• Protein dynamics prediction

Molecule Generation:

• Generative models for de novo drug design
• Optimization for multiple objectives simultaneously
• Synthetic accessibility scoring

Property Prediction:

• ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity)
• Binding affinity prediction
• Selectivity and off-target effects

Active Learning:

• Iterative experimental feedback
• Model refinement from real-world results
• Closed-loop discovery cycle

Major Partnerships and Validation

Eli Lilly Collaboration (2023)

Deal structure:

• Worth up to $1.7 billion
• Multiple drug targets across various diseases
• Combines Lilly's biological expertise with Isomorphic's AI

Strategic significance:

• Validation from Big Pharma
• Access to vast experimental data
• Testing platform at scale

Novartis Partnership (2024)

Deal structure:

• Up to $1.2 billion
• Focus on multiple therapeutic areas
• Long-term collaboration framework

Implications:

• Second major pharma endorsement
• Expanding therapeutic coverage
• Proving technology generalizes

What These Deals Mean

$3 billion in partnerships before announcing a single drug candidate signals:

• Pharma recognizes transformative potential
• Isomorphic's technology is mature enough for commercial application
• Traditional drug discovery inefficiencies create massive willingness to pay

The Team and Culture

Leadership

Demis Hassabis (CEO):

• Co-founder of DeepMind
• Neuroscientist and AI researcher
• Chess master and game design background
• Track record: AlphaGo, AlphaFold, AlphaZero

Key Hires:

• Recruited top talent from DeepMind, pharma, and academia
• Blend of AI researchers and drug discovery experts
• Emphasis on interdisciplinary collaboration

The Culture

Drawing from DeepMind's playbook:

Long-term thinking:

• Not chasing quick wins
• Building foundational AI capabilities
• Patient capital from Alphabet

Research excellence:

• Publishing in top journals
• Contributing to open science (where appropriate)
• Maintaining academic rigor

Ambitious goals:

• Not incremental improvements
• Seeking step-change innovations
• Moonshot mentality

The Competitive Landscape

Isomorphic Labs operates in an increasingly crowded space:

Other AI-First Drug Discovery Companies

Insilico Medicine:

• Established player (founded 2014)
• Multiple drugs in clinical trials
• Strong generative chemistry platform

Recursion Pharmaceuticals:

• High-content imaging + AI
• Massive wet lab automation
• Public company (NASDAQ: RXRX)

Relay Therapeutics:

• Protein dynamics focus
• Public company (NASDAQ: RLAY)
• Drugs in clinical development

Exscientia:

• AI-designed molecules in trials
• Partnership with pharma companies
• Public company (NASDAQ: EXAI)

Isomorphic's Differentiators

AlphaFold advantage:

• Direct lineage from most successful AI-for-biology tool
• Proven track record of breakthrough innovation
• Access to DeepMind's AI infrastructure

Alphabet backing:

• Nearly unlimited computational resources
• Patient, long-term capital
• Not pressured by quarterly earnings

Talent density:

• Ability to recruit top AI researchers
• DeepMind pedigree
• Compelling mission

The Scientific Approach

Integration Over Point Solutions

Rather than just better screening or better design, Isomorphic aims for end-to-end optimization:

• Integrated pipeline from target to candidate
• Each component informs others
• Global optimization, not local improvements

Data Strategy

Leveraging multiple data sources:

• Public databases (PDB, ChEMBL, etc.)
• Literature mining
• Partner data from Lilly and Novartis
• Proprietary experimental results

Data flywheel:

• More experiments → Better models
• Better models → Better predictions
• Better predictions → More efficient experiments
• Virtuous cycle accelerates

Computational Scale

With Alphabet's resources:

• Massive parallel computation
• Ability to simulate millions of molecules
• Explore chemical space systematically
• Run experiments digitally before physically

Challenges and Risks

The Translation Gap

From computation to clinic:

• Computational predictions must hold in real biology
• Human pharmacology is complex
• Clinical trials still required

Mitigation:

• Conservative predictions
• Extensive validation
• Partnership with experienced pharma companies

The Timeline Question

Drug development is slow:

• 10-15 years from discovery to market
• High failure rates (90%+ fail in trials)
• Isomorphic's first drugs won't launch for years

Competition

The field is moving fast:

• Multiple well-funded competitors
• Traditional pharma building AI capabilities
• Risk of being leapfrogged by next breakthrough

The Bigger Picture

Changing How Science Is Done

Isomorphic represents a new model:

Not just a tool:

• Not selling software to pharma
• Using AI themselves to discover drugs
• Proving AI can replace traditional R&D

Vertical integration:

• Control entire pipeline
• Optimize globally, not locally
• Capture full value of innovation

The Hassabis Philosophy

Demis Hassabis has consistently articulated a vision:

AI's greatest impact will be accelerating scientific discovery itself—not just automating existing processes, but fundamentally changing how research is conducted.

Isomorphic Labs is this philosophy incarnate.

What Success Looks Like

Near-term (2-5 years)

• Multiple drug candidates in clinical trials
• Demonstrated faster, cheaper discovery than traditional methods
• Additional pharma partnerships

Medium-term (5-10 years)

• First Isomorphic-discovered drug approved
• Proven hit rate superior to traditional discovery
• Industry-wide adoption of AI-first approaches

Long-term (10+ years)

• Dozens of approved drugs
• AI-designed drugs are the norm, not exception
• Pharmaceutical development fundamentally transformed

Lessons for the Broader Field

Isomorphic's approach offers insights for AI in science:

Deep domain integration:

• Not just applying ML to existing data
• Fundamentally rethinking the problem
• AI + domain expertise

Patient capital:

• Long-term investments needed
• Don't expect immediate returns
• Breakthroughs take time

Close the loop:

• Computational and experimental tightly coupled
• Rapid iteration
• Active learning essential

Conclusion

Isomorphic Labs represents the most ambitious attempt yet to apply AI to drug discovery. With Demis Hassabis's track record, Alphabet's resources, and major pharma partnerships, it has the ingredients for success.

But the real significance extends beyond any single company. Isomorphic is testing a hypothesis about the future of science itself: that AI can not just assist but transform how we discover new knowledge. If successful, Isomorphic won't just create new drugs—it will create a new way of creating drugs.

As we watch Isomorphic's progress over the coming years, we're witnessing a live experiment in science at digital speed. The outcome will shape not just pharmaceutical development but the entire enterprise of scientific discovery in the age of AI.

References

1. Hassabis, D. (2021). Announcing Isomorphic Labs. Isomorphic Labs Blog.
2. Jumper, J. et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596, 583–589.
3. Abramson, J. et al. (2024). Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, 630, 493–500.
4. Various industry reports on Isomorphic Labs partnerships (2023-2024).