Stripes x
Enter password to continue

Loading...

Home›Introduction›Cover
Stripes ×

Reimagining
Drug Discovery
with AI

Stripes is an NYC-based growth equity firm that invests in and actively supports best-in-class, category-defining companies. We believe Isomorphic Labs will revolutionize drug discovery.

$0BAUM
Fund VIICurrent Fund
NYCHeadquartered
Etched
Etched
Cognition
Cognition
Crusoe
Crusoe
Flatiron Health
Flatiron Health
On
On
Ramp
Ramp
Home›Introduction›Recent Investments
Category-Defining Investments

Our recent investments

Stripes partners with founders building n-of-1, category-defining companies with enormous opportunity. We look for amazing products with amazing market opportunities.

CompanyOverviewRound
Etched
Etched
Transformer-specialized AI inference chips$500M at $5B, Dec 2025
Cognition
Cognition
Autonomous AI software engineer (Devin)$400M at $10.2B, Sep 2025
Crusoe
Crusoe
AI infrastructure and data centers$1.375B at $10B+, Oct 2025
OpenAI
OpenAI
Frontier AI research and products$6.6B secondary at $500B, Oct 2025
AI
Applied Intuition
Vehicle intelligence and autonomy software$600M at $15B, Jun 2025
DB
Databricks
Leading data and AI platform$10B at $62B, Dec 2024
We believe Isomorphic Labs is a category defining company with the potential to have an enormous impact on human health, and create immense value in the long term.
Home›Industry Landscape›R&D Economics
Industry Landscape · 01

The economics of drug R&D
have to change

Returns on late-stage pipelines have improved for three years in a row, but the improvement depends almost entirely on one drug class. Outside GLP-1s, the economics of discovery have not changed.

0%
Projected IRR on late-stage pipelines, top 20 pharma, 2025
Up from 5.9% in 2024. Excluding GLP-1 drugs, the figure is 2.9%. Deloitte describes the recovery as fragile.
R&D IRR, with and without GLP-1s (2025)
All late-stage assets
7.0%
2024 baseline
5.9%
Excluding GLP-1 drugs
2.9%
Bar scale: 10% = full width
$0B
Cost to bring one drug from discovery to launch
Up from $2.23B in 2024. Includes the cost of programs that fail along the way.
0%
Of projected sales come from 9% of assets
54 blockbuster assets account for most of the projected value, so the outcome of any one program has a large effect on overall returns.
$0B+
Branded sales exposed to generics and biosimilars, 2026 to 2032
A larger loss-of-exclusivity wave than in either of the previous two periods.
Higher R&D spend alone will not close this gap. The industry needs discovery methods that produce better candidates at a lower cost per program.

Sources: Deloitte, Measuring the return from pharmaceutical innovation, 16th edition, May 2026 (IRR, cost per asset, blockbuster concentration). Evaluate Pharma loss-of-exclusivity analysis, 2026, as reported by DCAT Value Chain Insights (sales at risk).

Home›Industry Landscape›Where Drugs Die
Industry Landscape · 02

Roughly one in twenty candidates
makes it to patients

Clinical attrition has not improved meaningfully in two decades. Most failures happen in Phase 2, and many trace back to decisions made at the design stage: which target to pursue and how well the molecule binds it.

Clinical phase transition success, all modalities
Phase 1 → Phase 2
~50%
Phase 2 → Phase 3
~28%
Phase 3 → Filing
~50%
Phase 1 → Approval
~5%

Early evidence that AI-designed molecules perform differently in the clinic

Phase 1 success, historical industry
0%
Published estimates range from 40% to 65%
→
Phase 1 success, AI-discovered molecules
0%
80% to 90% in the first peer-reviewed analysis (2024). IQVIA's 2026 review of AI-enabled emerging biopharma found 75%.

Phase 2 success rates for AI-discovered molecules are so far in line with historical norms, at roughly 40%. The Phase 1 improvement suggests AI is already producing safer, more drug-like molecules. Whether it can also improve efficacy is still an open question, and it is the one Isomorphic's work on target selection and binding affinity is designed to answer.

Longest
in a decade
End-to-end development timelines, 2025
The gap between trials grew by three months in 2025, reversing recent gains.
0
Novel active substances launched globally, 2025
Output has been broadly flat while the cost per approval has risen.
Small improvements in candidate quality at the design stage carry through every later phase of development. This is where Isomorphic's platform is focused.

Sources: Nature Communications, "Dynamic clinical trial success rates for drugs in the 21st century", 2025 (phase transitions, ~5% overall). BIO, Informa and QLS Advisors, Clinical Development Success Rates 2011–2020 (Phase 2 → 3: 28.9%; likelihood of approval 7.9%). Jayatunga et al., Drug Discovery Today, 2024 (AI-discovered Phase 1 80–90%, Phase 2 ~40%). IQVIA Institute, Global R&D Trends 2026, March 2026 (AI-enabled Phase 1 75%, timelines, 79 launches).

Home›Industry Landscape›Promise to Proof
Industry Landscape · 03

AI drug design has crossed
from promise to proof

It has taken six years to go from AlphaFold 2 to the first Phase 3 trial of a drug designed with generative AI. Regulators have adjusted alongside the science. The milestones we have followed:

Nov 2020

AlphaFold 2 solves protein structure prediction

DeepMind's model reaches experimental-level accuracy at CASP14, addressing a problem that had been open for 50 years.

May 2024

AlphaFold 3 extends to ligands, nucleic acids and antibodies

Published in Nature by Google DeepMind and Isomorphic Labs. Structure prediction becomes directly useful for drug design.

Oct 2024

Nobel Prize in Chemistry

Demis Hassabis and John Jumper are recognised for AlphaFold.

Apr 2025

FDA moves to phase out animal-testing requirements

The agency publishes a roadmap encouraging AI-based computational models and other new approach methodologies in IND submissions, effective immediately. It reported meeting its first-year goals in April 2026.

Jun 2025

First AI-designed drug shows efficacy in a randomised trial

Rentosertib, for which both the target and the molecule were generated with AI, improves lung function in idiopathic pulmonary fibrosis. Published in Nature Medicine.

Feb 2026

Isomorphic Labs unveils its Drug Design Engine

IsoDDE more than doubles AlphaFold 3's accuracy on the hardest protein-ligand cases and improves on it 2.3x for antibody-antigen prediction. It predicts binding affinity at the accuracy of physics-based methods in a fraction of the time.

May 2026

Isomorphic raises $2.1B Series B

Led by Thrive Capital with Alphabet, GV, CapitalG, MGX, Temasek and the UK Sovereign AI Fund. The capital is intended to scale the engine and advance proprietary programs toward the clinic.

Sep 2026

First Phase 3 trial of a generative-AI-designed drug doses patients

Rentosertib enters a 320-patient Phase 3 trial, the first late-stage test of a drug designed this way.

The evidence now supports AI-designed molecules reaching the clinic. The open question is which platforms will produce the best ones.

Sources: Google DeepMind and Isomorphic Labs, Nature, May 2024 (AlphaFold 3). Nobel Foundation, October 2024. FDA press announcements, 10 April 2025 and April 2026. Nature Medicine, June 2025 (rentosertib Phase 2a). Isomorphic Labs, technical report and announcement, 10 February 2026 (IsoDDE); Series B announcement, 12 May 2026. Insilico Medicine press release, 10 September 2026 (GENESIS-IPF-3).

Home›Industry Landscape›Platform Deals
Industry Landscape · 04

Pharma is buying platforms,
not molecules

2026 opened with a series of large AI platform agreements. Major pharma companies are increasingly paying for design engines, compute and multi-target access rather than licensing individual assets.

~$3B
Isomorphic Labs × Eli Lilly and Novartis
Combined potential value of the two small-molecule collaborations announced in January 2024, excluding royalties. Upfront payments of $45M and $37.5M. Novartis expanded its program in 2025.
Jan 2024 · expanded 2025
Multi-target
Isomorphic Labs × Johnson & Johnson
Cross-modality collaboration covering small molecules and biologics. Isomorphic handles in silico design; J&J runs experimental assays and development.
Jan 2026
Up to $2.75B
Eli Lilly × Insilico Medicine
$115M upfront plus milestones and royalties for AI-designed oral therapeutics across several therapeutic areas.
Mar 2026
Up to $1B
Eli Lilly × NVIDIA
Five-year co-innovation lab in South San Francisco focused on closed-loop discovery and AI models for clinical development.
Jan 2026
Up to $1B
Merck × Google Cloud
Multi-year deployment of agentic AI across R&D, manufacturing and commercial functions.
Apr 2026
Model licences
Lilly × Chai · Pfizer × Boltz · GSK × Noetik
Agreements covering biologics design, structure and affinity models, and outcome-prediction models, all signed in the first weeks of 2026.
Jan 2026

Competitive pressure is also rising

All-time high
China-linked licensing deals, 2025
US and European companies are licensing more molecules from China to fill their pipelines.
0
First-in-class drugs launched only in China, 2025
China brought more new active substances to market than the US for the first time.
Most large pharma companies now want an AI design partner, and only a small number of platforms have the track record to fill that role. We believe Isomorphic is one of them.

Sources: Isomorphic Labs announcements, 7 January 2024 and 20 January 2026. Insilico Medicine press release, 29 March 2026. Eli Lilly investor release, 12 January 2026. Merck press release, 22 April 2026. Genetic Engineering & Biotechnology News, 22 January 2026 (Chai, Boltz, Noetik). IQVIA Institute, Global R&D Trends 2026, March 2026. Citeline, Pharma R&D Annual Review 2026.

Home›Our Perspective›Our Thesis
Our Perspective

How we think about
AI drug discovery

Six observations about where AI drug discovery is heading. Taken together, they point to the kind of company the industry's structure will favour. Each is expanded on the pages that follow.

1
AI has moved from predicting structures, to generating leads, to compressing the workflow
The first wave predicted protein structures. The second generates actual drug candidates against a target. The third turns sequential design-and-test rounds into parallel computation.
2
Advantage is shifting from algorithms to data, feedback loops and stack integration
Algorithmic improvements drove early progress in model capability. As those capabilities advance and commoditise, differentiation moves to proprietary data, fast experimental feedback and integration across discovery and development.
3
Leads for easier targets are largely solved. The frontier is hard targets and clinical success
Two fronts: generating leads against hard target classes, and improving the odds a candidate succeeds in humans. Discovery AI and preclinical AI are becoming distinct layers.
4
Proprietary biological and human outcome data become the deepest moat
Organoids, in-vitro systems and animal models help, but only go so far. Human outcomes are the ground truth, and only a full-stack platform can close the loop from design decision to clinical result.
5
Owning drug economics beats selling discovery software
Software budgets are bounded by workflow ROI. A successful therapeutic is worth billions. The best platforms will move toward milestones, royalties and their own assets.
6
Winners will look like AI-native drug companies, not software companies
Frontier models, proprietary data, wet-lab validation, clinical development and asset ownership in one closed loop. The industry's structure favours a scaled, full-stack player.
The thesis in one sentence

AI drug discovery has progressed from structure prediction to lead creation. As the frontier shifts toward harder biology and, ultimately, predicting clinical success, competitive advantage should increasingly accrue to scaled, full-stack platforms that can compound proprietary experimental and human outcome data while capturing the economics of the drugs they create. We believe this positions Isomorphic particularly well.

Home›Our Perspective›1 · Structure to Creation
Our Perspective · 01

AI drug discovery has moved from predicting structures,
to generating leads, to compressing the workflow

The first wave of AI breakthroughs was about understanding biology, most notably predicting protein structures. The field has since progressed to generating actual drug candidates against a target, and is now compressing meaningful parts of the discovery workflow into computation.

Stage one
Predict structures
What does the protein look like, alone and in complex with ligands, nucleic acids and antibodies. AlphaFold 2 and 3 made this routine.
→
Stage two
Generate leads
Design molecules that bind a chosen target: small molecules, antibodies, peptides. Generative models now produce real drug candidates, not just predictions.
→
Stage three
Compress the workflow
Optimise binding, developability, cross-species reactivity and bispecific behaviour at the same time, so sequential design-and-test rounds become parallel computation.

Properties that can now be optimised together

Binding affinityDevelopabilityCross-species reactivityBispecific behaviourSelectivity

Bispecifics are a good example. In the wet lab they are painful. On the computer you design both arms at once.

AI Researcher, Frontier Drug Discovery Lab
Historically: sequential
Design → Make → Test → Learn
Design → Make → Test → Learn
Design → Make → Test → Learn
Each property improved in its own round of experiments
Increasingly: parallel
Binding
Developability
Cross-species
Bispecific
Properties optimised jointly in computation before the first molecule is made

What we see in the market

AlphaFold 3 modelled complexes, not just proteins
The May 2024 model from Google DeepMind and Isomorphic Labs predicts joint structures of proteins with small molecules, nucleic acids and antibodies. That is the step from describing biology to designing against it.
De novo antibody design now works at useful hit rates
Chai Discovery's Chai-2 (July 2025) designed binders for 26 of 52 targets with a roughly 16% hit rate, going from design to wet-lab confirmation in under two weeks. Earlier computational methods were well below 1%.
Affinity prediction is catching physics-based methods
Isomorphic's IsoDDE (February 2026) reports binding-affinity accuracy that can exceed free-energy perturbation, at a fraction of the time and cost, and identifies new binding pockets from sequence alone.
A fully AI-generated drug has reached Phase 3
For rentosertib, both the target and the molecule were produced by AI. It showed efficacy in a randomised Phase 2a (June 2025) and began Phase 3 in September 2026.

Sources: Abramson et al., Nature, May 2024. Chai Discovery, "Zero-shot antibody design in a 24-well plate", bioRxiv, July 2025. Isomorphic Labs, IsoDDE technical report, 10 February 2026. Nature Medicine, June 2025; Insilico Medicine, 10 September 2026.

What used to be a chain of separate wet-lab campaigns, find a binder, then fix developability, then fix cross-species reactivity, is becoming a single design step. On a hard program that takes one to two years out of the front end.

AI Researcher, Frontier Drug Discovery Lab
The workflow is changing from a series of experimental cycles into a largely computational process, with experiments used to confirm rather than to explore.
Home›Our Perspective›2 · Algorithms to Data
Our Perspective · 02

Competitive advantage is shifting from algorithms
toward data, feedback loops and stack integration

Early advances in AI drug discovery were driven mainly by algorithmic innovation: better architectures, training techniques and ways of representing biological structure. As those capabilities advance and become commoditised, differentiation is moving to proprietary data, the experimental loops that generate it, and integration across the stack.

Where the marginal advantage comes from
Early phase
Algorithms
Data
Next phase
Algorithms
Proprietary data · feedback loops · integration
Illustrative, not measured

Most of the progress so far came from better algorithms. That phase is ending, because everyone now trains on the same public data and shares the same blind spots. Differentiation has to come from data nobody else has.

AI Researcher, Frontier Drug Discovery Lab
Why algorithms diffuse
  • Architectures and training methods are published and reproduced quickly.
  • Open-source structure and affinity models are closing the gap on proprietary ones.
  • Compute is available to any well-funded team.
Why data does not
  • Public datasets cover the biology that is already well understood.
  • Harder problems need new experimental data that does not yet exist.
  • The ability to generate that data quickly, in a form models can learn from, is difficult to replicate.
Why integration compounds
  • When the same organisation owns the models, the lab and the development programs, each experiment can be designed to teach the model.
  • Results flow back without contractual friction or data leaving the building.
  • A point-solution vendor depends on partners for every step of that loop.

The lab is what made the algorithms improve. Every recipe change was tested against real binders, and the speed of that loop decided who moved fastest.

AI Researcher, Frontier Drug Discovery Lab

What we see in the market

Open models are closing the gap fast
Boltz-2, released under an MIT licence by MIT and Recursion in June 2025, approaches the accuracy of physics-based affinity methods at roughly 1,000x the speed. Capabilities that were proprietary a year earlier are now free to use.
Pharma is treating its experimental data as the scarce asset
Lilly TuneLab (September 2025) offers biotechs models trained on Lilly's disposition, safety and preclinical data on hundreds of thousands of molecules, gathered at a cost of more than $1B. Access is federated and partners contribute data back.
Recent deals are about data, not model weights
Chai Discovery's January 2026 agreement with Lilly includes an exclusive model trained on Lilly's proprietary data. The differentiation is in what the model learns from, not the architecture.
Isomorphic kept its engine closed
IsoDDE was published as a technical report, not as open weights. That is consistent with a view that the model, and the data behind it, are the business.

Sources: Recursion and MIT Jameel Clinic, 6 June 2025 (Boltz-2). Eli Lilly investor release, 9 September 2025 (TuneLab). Genetic Engineering & Biotechnology News, 22 January 2026. Isomorphic Labs, 10 February 2026.

A thousand molecules designed by your own model and put through your own assays teach the model more than a hundred thousand random molecules bought from anyone else.

AI Researcher, Frontier Drug Discovery Lab
As models move beyond problems that are well represented in public data, the platforms that can run their own experiments, feed the results back into the model, and do so across the whole discovery-to-development stack will pull ahead.
Home›Our Perspective›3 · The Next Frontier
Our Perspective · 03

Leads for easier targets are largely solved.
The frontier is hard targets and clinical success

AI has made substantial progress on the easier lead-generation problems: targets with well-characterised structures and plenty of data. The frontier is now moving on two fronts at the same time.

Direction one
Generating leads against hard targets

Target classes where structural states are more complex and training data are scarce. Models are beginning to produce leads here, and the first successes are appearing.

GPCRsIon channelsPeptide-MHC complexesIntrinsically disordered proteins
Direction two
Improving the odds of clinical success

Beyond a molecule that binds its target, toward a candidate more likely to work in patients. This is where a large share of value sits: most of the cost of drug development is spent on candidates that fail in the clinic.

ADMETToxicityImmunogenicityEfficacyPatient heterogeneity

Look closely at where the published hits are and it is mostly targets pharma already had antibodies against, or that sit in the Protein Data Bank. The easy problem is solved. The hard targets are not.

AI Researcher, Frontier Drug Discovery Lab

The AI stack is separating into layers

Orchestration and AI science models
Broad reasoning systems that read the literature, form hypotheses and coordinate specialist tools. Target identification still often sits here rather than in a dedicated model.
Claude for Life Sciences · Phylo (Biomni)
Discovery AI
Identifies, generates and designs promising leads against a target. Specialist structure, affinity and generative models.
Isomorphic Labs · Chai Discovery · Xaira · Boltz
Preclinical AI
Works out, and improves, the odds that a candidate translates into a successful human drug: toxicity, ADMET, efficacy in human tissue.
Axiom Bio · Vivodyne

Discovery AI and preclinical AI are different problems with different data. The first asks whether a molecule can be made to bind. The second asks whether it will be safe and effective in a person. Because clinical attrition compounds, improvements in the second are worth at least as much as improvements in the first.

Since the middle of 2025 the frontier has moved to GPCRs, ion channels and peptide-MHC. Binders against those are no longer impossible, but they are not routine, and they need much more wet-lab throughput than a one-shot model can give you.

AI Researcher, Frontier Drug Discovery Lab
The question was
"Can we design a molecule that binds this target?"
→
The question becomes
"Can we predict what happens when we put that molecule into a human?"

Better molecules do not solve Phase 2. Not one AI-designed antibody has been in a human yet, and the in-vitro assays for immunogenicity tell you little about what will happen in a patient.

AI Researcher, Frontier Drug Discovery Lab

Why this frontier matters

Large target classes remain hard
About a third of FDA-approved drugs act on GPCRs, and that class alone covers more than 100 distinct targets. Ion channels and peptide-MHC complexes are similarly important and similarly under-represented in structural data.
Efficacy, not safety, is what kills drugs in Phase 2
In recent analyses, 59% of Phase 2 failures were for lack of efficacy and 22% for safety. Better molecules alone do not fix this. Better target choice and better prediction of human response do.
AI has moved Phase 1, not yet Phase 2
IQVIA's 2026 review found AI-enabled programs from emerging biopharma passing Phase 1 at 75%, but Phase 2 rates on par with peers. The field's next proof point is efficacy.
Preclinical AI is emerging as its own layer
Axiom Bio raised $40M in 2025 for AI liver-toxicity models trained on more than 115,000 molecules, and reports flagging the risk in two Pfizer programs later discontinued. Vivodyne raised $40M in May 2025 to test candidates on thousands of lab-grown human tissues. Both aim squarely at the clinical failure rate.
Early signs of reaching new biology
IsoDDE identified both the known and a recently discovered cryptic pocket on cereblon from sequence alone. Cryptic pockets and molecular-glue biology are exactly where public structural data runs out.

Sources: Hauser et al., Nature Reviews Drug Discovery, 2017 (GPCRs). Dowden and Munro, Nature Reviews Drug Discovery, 2019 (Phase 2 failure causes). IQVIA Institute, Global R&D Trends 2026, March 2026. Endpoints News and BioPharmaTrend on Axiom Bio, 2025; Vivodyne Series A release, 28 May 2025. Isomorphic Labs, 10 February 2026.

Predicting whether a molecule will work in the clinic is the real prize. It will take years, but even moving the success rate by ten or twenty points would change the economics of the industry.

AI Researcher, Frontier Drug Discovery Lab
Both directions require data that is scarce in the public domain. Progress will depend less on model design and more on who can generate and learn from that data.
Home›Our Perspective›4 · The Data Moat
Our Perspective · 04

Proprietary biological and human outcome data
become the industry's deepest moat

Organoids, organs-on-chip, virtual cells and animal models improve preclinical prediction, but they can only take a model so far. If the problem is predicting clinical success, large-scale human outcome data is the ground truth. Platforms with scale, their own assets and large pharma partners can accumulate that data, and it compounds back into better models.

The compounding loop
Design
→
Experiment
→
Preclinical
→
Human trial
→
Human outcome
→
Better models
→
Better next drug
↻ each cycle improves the next

Who can observe the most of this loop

Point-solution AI vendor
Sees design and, at best, early assay data. Rarely learns what happened in the clinic.
Standalone biotech
Sees the full loop, but only for a handful of its own programs.
Scaled full-stack platform
Own assets plus programs across several large pharma pipelines. A much broader set of experiments and outcomes to learn from.

What we see in the market

Human outcome data already commands a premium
Roche paid $1.9B for Flatiron Health, a Stripes portfolio company, in 2018. Flatiron sold oncology EHR software, but Roche's stated rationale was driven in large part by its real-world evidence: regulatory-grade patient outcome data from more than 265 cancer clinics. That was before AI models could learn from such data at scale.
Pharma values its preclinical and safety data in the billions
Lilly describes the experimental data behind TuneLab as costing more than $1B to generate. This is the kind of data a design platform needs and cannot download.
Regulators are opening the door to human-relevant models
The FDA's April 2025 roadmap encourages computational and human-relevant methods in place of animal studies. Platforms that can link predictions to human data will benefit most.
Isomorphic's partnerships widen what it can observe
Programs with Eli Lilly, Novartis and Johnson & Johnson, across small molecules and biologics, put Isomorphic-designed candidates into three large development organisations alongside its own pipeline.

Sources: Roche media release, 15 February 2018. Eli Lilly investor release, 9 September 2025. FDA press announcement, 10 April 2025. Isomorphic Labs announcements, 7 January 2024 and 20 January 2026.

A full-stack company can close this loop. A point-solution software or model provider cannot: it sees the design step and rarely learns what happened in the clinic. That is why scale and full-stack ownership, not model quality alone, will decide who accumulates the deepest dataset.
Home›Our Perspective›5 · Owning Economics
Our Perspective · 05

Owning economics in successful drugs
is a superior value-capture model to selling software

Pharma's willingness to pay for software and research services is constrained by workflow ROI and R&D budgets. The value created by a successful therapeutic can reach billions of dollars. The asymmetry is large.

Software or services fee
Bounded by R&D budgets
Value of a successful therapeutic
Billions
Illustrative scale

How value capture moves downstream

1
Software licences
2
Research services
3
Milestones
4
Royalties
5
Co-development
6
Proprietary assets

What the numbers say

The software ceiling is visible
Schrödinger, the leading physics-based discovery software vendor, reported $200M of software revenue in 2025 after more than three decades. Its drug-discovery revenue doubled to $56M in the same year.
The pattern is industry-wide
Lilly's March 2026 agreement with Insilico pays $115M upfront against up to $2.75B in milestones and royalties, and includes an exclusive licence to Insilico-designed assets.
One successful asset is worth more than a software business
Deloitte puts average forecast peak sales for a late-stage asset at $598M a year in 2025. A single approved drug can exceed the lifetime revenue of most discovery-software companies.

Sources: Schrödinger, fourth-quarter and full-year 2025 results, February 2026. Insilico Medicine, 29 March 2026. Deloitte, Measuring the return from pharmaceutical innovation, May 2026.

Platforms that genuinely improve the probability of creating successful drugs should want to participate directly in the value of those drugs, and the most capable ones will be able to negotiate for it.
Home›Our Perspective›6 · AI-Native Drug Companies
Our Perspective · 06

The long-term winners should look less like software companies
and more like AI-native drug companies

Taken together, we believe AI drug discovery is evolving from a model-quality race into a race to build the best closed-loop drug development system.

What the strongest platforms combine
Frontier models
Structure, affinity and property prediction at the state of the art
Proprietary experimental data
Generated in-house, on the problems public data does not cover
Wet-lab validation
Experiments that confirm predictions and feed the model
Clinical development
Programs advanced into and through human trials
Asset ownership
Direct participation in the value of the drugs created
Compounding advantages in both model capability and economic value capture

Why this positions Isomorphic well

Accumulating the data for harder problems
  • Scale and technical depth from the DeepMind lineage.
  • Multi-target programs with Eli Lilly, Novartis and Johnson & Johnson across small molecules and biologics.
  • Proprietary programs moving toward the clinic.
Capturing value as AI moves downstream
  • Partnership structures with milestones and royalties rather than fees alone.
  • Capital to develop its own assets and participate deeply in drug development economics.
  • A position to benefit as prediction extends from binding toward clinical success.

The field is converging on this model

Xaira Therapeutics
Launched in April 2024 with more than $1B, built around de novo protein design from the Baker lab, with in-house wet labs and its own pipeline.
Insilico Medicine
Own pipeline through to Phase 3, plus a $2.75B licensing and research agreement with Lilly. Discovery software has become a smaller part of the story.
Recursion
Public company combining automated wet labs and large proprietary biological datasets with open models such as Boltz-2 and its own clinical programs.
Isomorphic Labs
$2.7B raised across two rounds, a proprietary engine that leads AlphaFold 3, three large-pharma partners across modalities, and a first-in-human oncology trial targeted for 2026.

Sources: Xaira Therapeutics launch release, 23 April 2024. Insilico Medicine, 29 March and 10 September 2026. Recursion investor releases, 2025. Isomorphic Labs, 12 May 2026; CEO public statements, 2026.

We believe Isomorphic is well positioned both to accumulate the data required for the next generation of harder problems and to capture the value created as AI moves further toward clinical success.
Home›Our Team›Scale Team
Scale Team

Dedicated operating resources
to help Isomorphic scale faster

Stripes' in-house Scale Team provides hands-on support across every critical function — not just capital, but execution firepower embedded alongside your team.

Talent

Executive search, org design, compensation benchmarking, employer branding

Marketing & Brand

Brand positioning, content strategy, demand generation, event marketing

Finance & Legal

FP&A setup, audit readiness, legal counsel, entity structuring, tax strategy

Data & Analytics

Data infrastructure, KPI dashboards, customer analytics, pricing optimization

Growth & Sales

GTM strategy, enterprise sales playbooks, pipeline management, partnership sourcing

Operations

Procurement, vendor management, international expansion, systems & process design

Immediate Value-Add for Isomorphic Labs

Enterprise GTM Buildout

Help design and staff the enterprise sales motion for pharma partnerships — including territory planning, pricing strategy, and customer success frameworks

Executive Talent Pipeline

Source and recruit senior commercial leaders with pharma/biotech experience — CRO, VP Sales, Head of Partnerships — through our proprietary network

Pharma Customer Intros

Leverage advisory council relationships to open doors at top 20 pharma — warm intros to R&D decision-makers and Chief Digital Officers

Brand & Thought Leadership

Position Isomorphic as the category leader in AI drug discovery — conference strategy, media relations, case study development, and analyst engagement

Home›Our Team›Healthcare & AI Expertise
Healthcare & AI Expertise

Deep experience in
healthcare & life sciences

Healthcare is a core vertical for Stripes. We've built category leaders across pharma services, health tech, clinical analytics, and digital health — giving us pattern recognition directly relevant to Isomorphic.

Flatiron Health
Flatiron Health
Oncology SaaS & Real-World Evidence • Acquired by Roche for $1.9B
Placed CFO, Chief People Officer, IPO Controller; intros into RCM and RWE businesses. $29M → $90M ARR (57% CAGR)
Stripes placed three C-suite executives and provided strategic introductions that helped Flatiron scale its real-world evidence platform, ultimately leading to Roche's $1.9B acquisition. Directly relevant to Isomorphic as a pharma platform company that scaled into enterprise deals with large pharma.
Pomelo Care
Pomelo Care
Virtual Maternal Care Platform
Advised on performance marketing, strategy to expand to Medicare/Commercial. $11M → $66M revenue run rate (230% CAGR)
Our Scale Team connected Pomelo with payer network contacts and helped design their expansion strategy into Medicare and commercial insurance markets.
Equip
Equip
Virtual Eating Disorder Treatment
Evidence-based, family-centered eating disorder treatment delivered via telehealth. Rapid scale through payer partnerships and clinical outcomes data.
Equip demonstrates Stripes' ability to scale digital health platforms that require clinical rigor, payer contracting, and evidence-based validation — capabilities directly transferable to Isomorphic's pharma engagement model.
Boulder Care
Boulder Care
Virtual Addiction Medicine
Tech-enabled substance use disorder treatment platform. Scaled through Medicaid and commercial payer partnerships.
Boulder's model of using technology to improve clinical outcomes in a highly regulated environment mirrors the challenges Isomorphic faces in gaining pharma trust and regulatory acceptance.
Hello Heart
Hello Heart
Digital Cardiovascular Health
AI-powered heart health platform with clinically validated outcomes. Deployed through employer and health plan channels.
Hello Heart's combination of AI/ML-driven insights with clinical validation is a strong parallel to Isomorphic's approach — both require rigorous evidence to win enterprise healthcare buyers.
Chapter
Chapter
Medicare Advisory Platform
Tech-enabled Medicare advisory helping seniors navigate plan selection. Scaled through direct-to-consumer and B2B2C channels.
Chapter shows Stripes' expertise in scaling healthcare platforms that require deep domain knowledge, regulatory navigation, and trust-building — all relevant to Isomorphic's pharma engagement.

Pattern recognition: Across these investments we've learned how to sell complex technology platforms into pharma R&D organizations — long sales cycles, multi-stakeholder buying committees, compliance requirements, and the importance of clinical validation.

AI Expertise

Deep conviction in AI infrastructure and applications — direct pattern recognition for Isomorphic's AI-native approach to drug discovery.

Cognition
Cognition
AI Software Engineering
Creator of Devin, the first AI software engineer. Building autonomous AI systems capable of complex multi-step reasoning and execution.
Cognition's approach to building AI agents that can reason, plan, and execute autonomously mirrors the frontier AI capabilities that underpin Isomorphic's drug discovery platform. Both companies are pushing the boundaries of what AI systems can achieve in complex, high-stakes domains.
Flock Safety
Flock Safety
AI-Powered Public Safety Platform
AI-driven public safety technology used by thousands of communities and law enforcement agencies across the U.S. to reduce crime.
Flock Safety demonstrates Stripes' ability to scale AI platforms in highly regulated, mission-critical environments — a direct parallel to deploying AI in pharmaceutical R&D where accuracy, compliance, and trust are paramount.
Applied Intuition
Applied Intuition
AI-Driven Autonomous Mobility
Full-stack simulation and infrastructure platform for autonomous vehicle development, trusted by the world's largest automakers.
Applied Intuition's approach of building comprehensive AI tooling for a complex, safety-critical domain parallels Isomorphic's full-stack strategy for drug discovery. Both companies provide the critical AI infrastructure layer that enables their industries to adopt autonomy at scale.
Etched
Etched
Purpose-Built AI Inference Chips
Building Sohu, a transformer-specific ASIC that delivers an order-of-magnitude improvement in AI inference performance over general-purpose GPUs.
Etched is building the hardware layer that will power next-generation AI workloads. As Isomorphic's models grow in complexity, specialized inference hardware becomes critical infrastructure for running large-scale molecular simulations at production speed.
Crusoe
Crusoe
AI Cloud Infrastructure & Sustainable Compute
Clean-energy AI cloud platform providing high-performance compute for AI training and inference workloads at scale.
Crusoe addresses one of the fundamental bottlenecks in AI — access to cost-effective, sustainable compute at scale. As AI-driven drug discovery models become increasingly compute-intensive, infrastructure like Crusoe's becomes essential to the ecosystem Isomorphic operates within.
Home›Our Team›Our Network
Our Network

Dedicated operating resources
to help Isomorphic scale faster

Six Operating Partners work hands-on inside portfolio companies. Around 120 senior advisors and council members across pharma, healthcare, technology and enterprise provide access to decision-makers. The names below are the ones most relevant to Isomorphic; the full network is larger.

0
Operating Partners
Former C-suite operators embedded with portfolio companies
0
Senior Advisors
Including the former head of R&D at Pfizer
0+
Council members
Tech, CISO and Engineering councils of sitting CIOs, CTOs and CISOs
0
Executive Advisors
Functional experts in go-to-market, product and operations

Operating Partners

Full-time members of the Stripes team. They work inside portfolio companies on execution, not as outside advisers.

NameBackgroundRelevance to Isomorphic
Paul MelchiorreFormer CRO, AnaplanEnterprise sales: territory design, pricing and repeatable large-deal processes for pharma buyers
Scott AronsonFormer COO, ClouderaTurning bespoke technical engagements into a scalable platform offering
Sharon RothsteinFormer Global CMO, StarbucksCategory positioning as the company moves from research lab to commercial partner
Julie HerendeenFormer CMO, DropboxDemand generation and pipeline within pharma and biotech
Barb MessingFormer Chief Marketing & People Experience Officer, Roblox; former CMO, Walmart US and TripAdvisorEmployer brand and people operations through rapid headcount growth
Brigitte KleineFormer President, Tory Burch; former President, Michael KorsScaling operations from a founder-led organisation to a global one

Senior Advisors most relevant to Isomorphic

Senior Advisor
Mikael Dolsten

Former Chief Scientific Officer and President of Worldwide Research, Development and Medical at Pfizer, where he led one of the largest R&D organisations in the industry, including its external partnering.

Senior Advisor
Diana McKenzie

Former SVP and CIO of Amgen, following 17 years at Eli Lilly, and the first CIO of Workday. Board member at Vertex Pharmaceuticals and MetLife.

Eight further Senior Advisors include former CIOs and CTOs of PayPal, Zoom, AIG and Lockheed Martin.

Advisory Councils

Sitting executives who meet with portfolio companies as prospective customers, reference buyers and advisers. Listed here are the members at pharma, healthcare and AI-infrastructure organisations.

Tech Council · 40 members
Greg Meyers, Chief Digital & Technology Officer, Bristol Myers Squibb
Jim Swanson, Global CIO, Johnson & Johnson
Ron Kim, former CTO, Merck
Ryan Snyder, SVP & CIO, Thermo Fisher Scientific
Jacqui Nevils, Global CIO, Fresenius Medical Care
Muru Murugappan, Chief Information & Transformation Officer, Alcon
Jorge Populo, CIO, UnitedHealth Group
Michelle Greene, Global CIO, Cardinal Health
Daniel Barchi, CIO, CommonSpirit Health
BJ Moore, former CIO, Providence
Sandy Venugopal, CIO, CoreWeave
Plus CIOs and CTOs at Boeing, United Airlines, Sephora, Wayfair, Toast, Hasbro and others
CISO Council · 15 members
Aimee Cardwell, former EVP & CISO, UnitedHealth Group
Nick Vigier, CISO, Oscar Health
James Beeson, Global CISO, R1 RCM; former CISO, Cigna
Aanchal Gupta, Chief Security Officer, Adobe
Lakshmi Hanspal, Chief Trust Officer, DigiCert; former Global CISO, Amazon
Rinki Sethi, CISO, Upwind Security
Plus CISOs at Chevron, Coca-Cola, Electronic Arts, CBRE, IHG and others
Executive Advisors · 50 advisers
Tiffany Inglis, Chief Medical Officer, Sera Prognostics
David Weathington, VP Network & Value-Based Solutions, Elevance
Kevin McGavick, VP National Growth, ChenMed
Mike Bowersox, former Medicare President, Humana
Jason Parrott, SVP Enterprise Growth & Partnerships, Vida
Michael Hoff, former Global Head of Partners & Alliances, Mistral AI
Ricky Robinett, Senior Director, Developer Marketing, Google Cloud
Robert Welborn, former Head of Decision Science, Meta
Plus go-to-market, product and finance leaders from Stripe, HubSpot, Brex, monday.com, Klaviyo and others
Engineering & IT Council · 7 members
Senior engineering and infrastructure leaders at Sony, Nokia, KeyBank, Global Payments, DIRECTV and Daimler Truck. Useful as Isomorphic scales its own compute and data infrastructure.
The network is run by a dedicated in-house team, led by our Head of Network and Head of Talent, and is available to Isomorphic from the first day of a partnership.

Source: Stripes, Scale Resources and team pages, stripes.co, as of 29 September 2026. Titles as listed there.

Home›Our Team›Thank You
Stripes ×

Thank You

We believe Isomorphic Labs can revolutionize drug discovery. Let's build the future of medicine together.

Stripes × Isomorphic Labs • Confidential — For Discussion Purposes Only