Siyu He
Stanford University
siyuhe@stanford.edu
Bio
Dr. Siyu He is a Katharine McCormick postdoctoral fellow in the Department of Biomedical Data Science at Stanford University, co-advised by Dr. James Zou and Dr. Stephen Quake. Her research lies at the intersection of AI and biomedicine, with a focus on developing generative AI models to learn and predict the spatiotemporal dynamics of cells in both healthy and diseased states. She has created several computational tools, including Starfysh, CORAL, Squidiff, and Devo, and has published her work in Nature Biotechnology and Nature Methods. Siyu earned her PhD in Biomedical Engineering from Columbia University, where she was co-advised by Dr. Kam Leong and Dr. Elham Azizi, and her B.S. in Physics from Xi’an Jiaotong University in China. Her long-term vision is to advance AI for healthcare, including novel frameworks such as agentic AI and quantum AI.
Areas of Research
- AI for Healthcare and Life Sciences
A spatiotemporal virtual cell foundation model for development and diseases
Cell identity and fate are governed by the dynamic information flow encoded in cellular transcriptomes across time and space. While current high-throughput molecular profiling enables characterization of cell identity, limitations in spatiotemporal resolution restrict the ability to model various dynamic processes. Advances in generative AI combined with massive single-cell datasets across species and engineered tissues make it possible to build foundation models for virtual cells. Here we present Devo, a spatiotemporal, diffusion-based virtual cell foundation model that enables prediction and generation of cellular processes in healthy, diseased, and therapeutic contexts. Devo is pretrained on large-scale cross-species embryogenesis data, and fine-tuned with engineered tissue datasets, spanning normal, diseased, drug-treated, and spatial temporal data. Devo enables the prediction of cell transcriptomic states under a flexible design of conditions, including development, disease progression, and drug perturbation. Devo has shown the capability to generalize in human development across various cell types, and effectively indicated the developmental defect of the neurological diseases. When applied onto tumor cells, Devo serves as a powerful platform for drug screening. Furthermore, we applied Devo on tabula muris data and showed a powerful platform as a predicting spatial mouse aging atlas. Together, our results establish Devo as a foundation cell model for in silico exploration of cellular and molecular landscapes, and provides a powerful platform for precision and regenerative medicine.