(Gorodenkoff/Shutterstock)
What if we could stop cancer, heart disease, or Alzheimer’s before the first symptoms ever appear? That’s the question guiding ongoing efforts at the Chan Zuckerberg Initiative to reshape how we understand and treat disease at the cellular level. The driving force in this initiative isn’t medication or surgery, but artificial intelligence.
The Chan Zuckerberg Initiative (CZI) is a philanthropic organization founded in 2015 by Priscilla Chan and Meta CEO Mark Zuckerberg. Backed by a pledge to donate 99% of their Facebook shares, the initiative has become a major force in funding long-term science and technology projects focused on using AI to understand and prevent diseases.
Studying human cells is one of biology’s toughest problems. There are trillions of cells in the body, each driven by complex interactions between genes, proteins, and other molecular components. However, experiments in this field have traditionally been slow and often unpredictable. For years, researchers believed there were only a few hundred cell types, but newer methods like single-cell sequencing have revealed thousands.
A recent article in Nature profiled the Chan Zuckerberg Initiative’s efforts to build virtual cells, which are essentially AI-powered models designed to simulate how biological cells function and respond to changes, whether from drugs, mutations, or disease. This holds great potential in helping researchers and scientists generate better hypotheses even before entering the lab or doing any experiment.
(mechichi/Shutterstock)
The CZI is developing various open datasets and tools to support research on human cells, and plans to invest hundreds of millions of dollars over the next decade in building AI-powered virtual cell models.
At SXSW 2025, CZI co-founder Priscilla Chan explained how AI could be used to model the inner workings of human cells. “What if you showed an AI model three images of the human heart at an atomic level? What if you showed it videos of individual cells interacting with other cells? What if you taught it the molecular code and all the structures inside every one of your cells? What you would get is a powerful simulation of how the human cell works,” she said.
“We call that the virtual cell model. And we think a virtual cell model would completely change the way we understand health and medicine.”
Stephen Quake, the head of science at the Chan Zuckerberg Initiative (CZI) in Redwood City, California, is among the top researchers working on creating virtual cells. He believes the future of biology should look very different from today.
Instead of spending most of their time running lab experiments, he says researchers could rely on powerful AI models to simulate how cells work, using the lab mainly to confirm what the software predicts. “Our goal,” he says, “is to create computational tools so that cell biology goes from being 90% experimental and 10% computational to the other way around.”
Computational models of cells aren’t new, but early efforts relied on fixed, rule-based systems that often fell short. Projects like the 2012 simulation of Mycoplasma genitalium attempted to hard-code cellular processes but lacked the scalability or flexibility to model more complex organisms.
CZI’s approach is data-driven. Its models are trained on large single-cell RNA sequencing datasets, which provide snapshots of gene activity across thousands of individual cells. These data are used to build AI systems that can capture cellular states and transitions, with the goal of predicting how a cell might respond to genetic changes, drugs, or disease. Unlike older models, these systems don’t require a full mechanistic map; they infer behavior statistically from patterns in the data.
To support this approach, CZI has built one of the world’s largest nonprofit GPU clusters for life sciences. This includes 1000+ NVIDIA H100s arranged in a DGX SuperPOD. This infrastructure powers CZI-developed projects like Tabula Sapiens, a comprehensive cell atlas mapping the types and states of human cells, and TranscriptFormer, a deep learning model built by CZI researchers to compare gene expression across species. These tools provide the foundation for training virtual cell models.
As part of its broader efforts to reshape science by leveraging the power of AI, the Chan Zuckerberg Initiative, in partnership with the Sloan Foundation, has awarded eight new grants to support the development of AI tools that make research more efficient, transparent, and accessible. These range from improving peer review and figure captions to helping researchers organize knowledge and collaborate more effectively.
CZI is not alone in this space. The Arc Institute recently launched its own virtual cell model and a public challenge to benchmark predictions. Google DeepMind has also confirmed work on a virtual cell project.
Despite the progress and breakthroughs, not everyone in the scientific community is fully convinced about virtual cells. Some researchers view the current momentum around virtual cells as more aspirational than practical, noting that while the concept holds long-term promise, real results are still limited.
“It’s primarily being used as a rallying cry and a funding mechanism, and it’s working,” says Anshul Kundaje, a computational biologist at Stanford. “Investors are putting a huge amount of funding into this space.”
Whether virtual cells live up to their promise remains to be seen. For now, they represent a bold new direction in the intersection of AI and biology.
