CellType Simulates Human Biology to Find Cancer Treatments That Actually Work

The Macro: Drug Discovery Is Expensive Because Biology Is Hard to Predict

It costs over $2 billion to bring a single drug to market. Most of that cost comes from failure. For every drug that succeeds, dozens fail in clinical trials. The attrition rate in oncology is particularly brutal, with 96% of cancer drug candidates failing to reach approval.

The fundamental challenge is that biology is messy. Human cells interact in ways that are difficult to model. A drug that kills cancer cells in a petri dish might do nothing in a living patient. A compound that shows promise in mice might be toxic in humans. The only way to know for sure is to run clinical trials, which take years and cost hundreds of millions of dollars.

The promise of computational biology is to simulate enough of this complexity in silico that you can filter out the losers before spending billions on clinical trials. Recursion Pharmaceuticals, Insitro, and others are pursuing versions of this approach. But the models have historically been limited by data availability and biological fidelity. You cannot simulate what you cannot model.

Foundation models changed the equation. Trained on massive biological datasets, these models can capture patterns and relationships that traditional computational methods miss. The question is whether they are good enough to make real predictions about drug candidates that translate to clinical outcomes.

The Micro: A Yale Professor’s Models Are Already in Top 10 Pharma

David van Dijk and Ivan Vrkic cofounded CellType. David is a Yale Professor with 11,000+ citations. Ivan is an ML researcher from Yale and EPFL who co-developed the core technology and previously built software to control CERN’s Large Hadron Collider. They are a two-person team from YC Winter 2026 with Ankit Gupta.

The core technology is Cell2Sentence, a method for representing biological data that was published at ICML 2024. The approach was developed in collaboration with DeepMind and was featured by Sundar Pichai in a presentation viewed by 7 million people. It has already discovered and validated a novel cancer treatment signal.

CellType builds AI agents that run the full drug discovery pipeline on top of biological foundation models that simulate human biology. The company positions itself as a “biological reasoning engine.” They are already working with Top 10 pharma companies, and 100% of their deals are inbound. When the world’s largest pharmaceutical companies come to you, that says something about the technology.

The platform lives at celltype.com, though the site itself is minimal. The research credentials and commercial traction tell the story better than any marketing page could.

The Verdict

CellType has the kind of scientific credibility that most biotech startups can only dream of. ICML publications, DeepMind collaboration, a Yale professor with 11,000 citations, and Top 10 pharma customers. The question is not whether the technology is real. It is whether it can translate from research validation to commercial drug development at scale.

The competitive field includes Recursion, Insitro, and Insilico Medicine. These companies are well-funded and further along in the drug development pipeline. But CellType’s approach, using biological foundation models and AI agents to simulate biology rather than just analyzing data, is architecturally different and potentially more powerful.

In 30 days, I want to see the pipeline of pharma engagements. How many simultaneous partnerships are generating revenue? In 60 days, the question is whether any CellType-identified drug candidate has entered preclinical development. In 90 days, I want to know about the model’s predictive accuracy on held-out data. How well do CellType’s biological simulations predict real experimental outcomes? That number determines whether this is a research tool or a drug discovery engine.