Stanford AI Maps Cell Biology Across 12 Species

Researchers at Stanford Medicine, in collaboration with the Chan Zuckerberg Initiative, have developed two advanced artificial intelligence models that enable comprehensive cross-species mapping of cellular biology. The foundational system, Universal Cell Embedding, paved the way for its successor, TranscriptFormer, a second-generation architecture trained on 112 million cells representing 12 distinct species, including humans, yeast, mice, zebrafish, and sea sponges. The research spans 1.53 billion years of evolutionary history and was detailed in recent publications in Nature and Science. TranscriptFormer operates on a methodology analogous to large language models, processing gene expression profiles rather than text. By training algorithms to predict missing cellular data points, the system constructs a mathematical universal space where cells from any organism can be mapped and compared based on functional similarity. This architecture allows researchers to navigate complex genomic data that exceeds human analytical capacity, transforming how biological information is interpreted. The models deliver immediate utility in evolutionary biology, comparative genomics, and disease research. By quantifying cellular relationships, TranscriptFormer has already resolved longstanding biological questions, such as confirming that sponge choanocytes share a closer genetic relationship to neurons in roundworms and frogs than to previously identified sponge neuroid cells. The system can also accurately identify cell types in untrained species, differentiate between healthy and diseased states, and simulate functional cellular configurations for therapeutic development. Stephen Quake, professor of bioengineering at Stanford Medicine, and Jure Leskovec, professor of computer science, co-led the project alongside Theo Karaletsos of the Chan Zuckerberg Initiative. The work, supported by BioHub and the Arc Institute, represents a foundational shift toward AI-driven biological discovery. Researchers anticipate the platform will serve as a critical instrument for decoding complex cellular mechanisms, accelerating cross-species analysis, and informing next-generation cell-based therapies.