# AI Model Maps Cells Across the Tree of Life
Researchers at Stanford Medicine have built an artificial intelligence system that can analyze cells from yeast to humans within a single unified framework. The team developed two models trained on gene expression data from multiple species.
Stanford Medicine researchers first developed a model called universal cell embedding, which established the initial approach to cross-species cellular biology mapping. Building on that foundation, the team expanded the work to create a second-generation model with broader capabilities.
The training dataset included humans, mice, rabbits, chickens, zebrafish, fruit flies, the African clawed frog, the malaria-causing parasite, a sea urchin, a sea sponge, the roundworm Caenorhabditis elegans, and yeast. The system processes these diverse organisms by analyzing which genes each cell uses—a property called gene expression that defines what a cell actually does, rather than simply what its genome contains.
Stephen Quake, a Stanford professor of bioengineering and co-lead author of the research published in Science and Nature, explained the technical challenge. "How do we think about all these genes at once? The human brain can't," he said. "These models help us do that. The algorithms are ways for us to understand very complex data that humans can't wrap our minds around."
The second-generation model learns through a process analogous to large language models like ChatGPT. Rather than predicting missing words in text, the system repeatedly examines gene expression patterns with some values hidden, adjusting its internal parameters until it can accurately predict the missing information. This creates what Quake describes as a "universal space"—a mathematical framework where any cell from any organism on Earth can be positioned and compared to others.
The model has already yielded specific biological insights. Scientists studying sponges have long debated which sponge cell type most closely resembles neurons found in other animals. The AI analysis showed that the sponge cell called a choanocyte exhibits gene expression patterns quite similar to neurons in roundworms and frogs, helping clarify the evolutionary origins of nervous systems.
The model also resolved another mystery about sponges. The so-called neuroid cell was named for its neuron-like features, but the AI revealed its gene expression profile more closely matches a frog gland, suggesting the cell likely plays a role in digestion rather than neural signaling.
Beyond comparing known organisms, the model demonstrated an ability to process species it had never encountered during training. When presented with gene expression data from an entirely new species, the system successfully identified cell types. It can also distinguish healthy cells from diseased ones.
The research team, which included scientists from the Chan Zuckerberg Initiative and BioHub, sees future applications in cellular medicine. The model could potentially guide the design of new cell-based treatments by computationally predicting functional cells.
The work spans hundreds of millions of cells drawn from many species, creating what researchers call an unprecedented view of cellular biology across the evolutionary spectrum. Quake characterized the potential: "We hope it's going to be a very powerful tool for discovery. This is a tool to help skilled biologists think about complex problems."
remove this sentence Co-first authors of the Nature paper were Yanay Rosen, a doctoral student in computer science at Stanford, and Yusuf Roohani, now an associate director of machine learning at the Arc Institute. Jure Leskovec, professor of computer science, and Theo Karaletsos of the Chan Zuckerberg Initiative served as co-lead authors alongside Quake.
**Sources**
https://phys.org/news/2026-09-ai-enable-species-cell-biology.html
https://doi.org/10.1038/s41586-026-10689-z
https://doi.org/10.1126/science.aec8514