Company profile · 3 min read
CellType: agentic drug discovery around human-biology models
CellType combines biological foundation models and AI agents to simulate human biology and prioritize drug-discovery decisions for pharma partners.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
CellType is building an agentic drug company around biological foundation models that simulate human biology. Its intended partner is a pharma or biotech team that wants earlier evidence about targets, toxicity and translation before committing to a long experimental or clinical program.
What it does
The YC profile says CellType combines AI agents with models of human biology to run the drug-discovery pipeline and is working with Top 10 pharma. The YC launch describes Cell2Sentence, a model that represents cellular biology, and a “virtual human” used for target discovery, toxicity prediction, translational prediction, patient stratification and virtual trials.
CellType says it screened more than 4,000 drugs, predicted a new cancer-treatment signal and validated it in cell lines and intact human tumor microenvironments. Google’s public write-up is linked from the company’s launch and describes the broader research context. These are company and partner-reported discovery claims, not a clinical result, approved treatment or replacement for human trials.
Why I’d look closer
The advantage is choosing a human-centered biological model before running every physical experiment. The founder context is unusually strong: David van Dijk is described as a Yale professor who built biological foundation models with Google, while Ivan Vrkic co-developed Cell2Sentence, published at ICML and worked on large-scale model training and CERN software.
The tradeoff is translation. A model can represent more of biology than a simple cell line while still missing dose, delivery, immune response, patient heterogeneity and clinical operations. Drug companies also need to understand which findings are licensed, reproducible and ready for a partner’s validation pipeline.
What I’d ask
Which predictions have independent wet-lab replication? How does CellType distinguish training data from validation data? What is the model’s performance by indication, population and modality? Can a partner inspect the evidence behind a target or toxicity prediction, and what exactly enters an IND-enabling program?
My editorial take
Shortlist CellType for a discovery partnership if translational risk is the bottleneck and your team can test the claims against owned assays. Do not treat a virtual human as a clinical substitute. The product earns its value when it changes which experiments get run and those choices hold up outside the model.
Quick facts
| Field | Sourced detail |
|---|---|
| Product | Biological foundation models and AI agents for drug discovery |
| Buyer | Pharma and biotech R&D teams |
| Public evidence | Company reports 4,000-plus drug screening and validated cancer-signal work |
| Pricing | Not published in the checked pages |
| Main question | Does the model improve translational decisions beyond existing preclinical assays? |
Sources checked
| Source | Checked |
|---|---|
| YC company profile | 2026-09-19 |
| CellType YC launch | 2026-09-19 |
| Google research context | 2026-09-19 |
Cohort context
CellType is listed in Winter 2026. In our 2026-09-18 directory snapshot, 16 of 199 listed companies in that cohort have YC’s primary industry label Healthcare (8.0%). This is a current-directory comparison, not an original intake count or a performance ranking. Nine-cohort dataset.
Public website snapshot
Observed 2026-09-19T16:20:06.374Z in raw homepage HTML. This records visible metadata and advertised links, not agent execution or product quality.
| Signal | Homepage observation |
|---|---|
| Product description metadata | Not observed in this response |
| Canonical link | Not observed in this response |
| H1 or H2 heading | Not observed in this response |
| Typed structured data | Not observed in this response |
| Docs/developer link | Not observed in this response |
| Pricing link | Not observed in this response |
| llms.txt link | Not observed in this response |
| Markdown alternate | Not observed in this response |
Public observations · Collection method. Missing links here do not establish that a capability or file is absent elsewhere.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗