The paper introduces stVCR, a generative deep-learning framework designed to reconstruct the continuous spatiotemporal dynamics of single cells from discrete snapshots. Because traditional spatial transcriptomics involves destructive sequencing, it typically produces unpaired "pictures" of biological processes rather than a continuous "video." To bridge this gap, stVCR utilizes dynamic optimal transport and rigid-body transformation to align data from different time points into a single coordinate system. This allows researchers to track cell differentiation, migration, and proliferation simultaneously while accounting for changes in cell populations due to division or death. The study validates the model's accuracy and robustness using both simulated datasets and real-world biological applications, such as axolotl brain regeneration and Drosophila embryo development. Ultimately, the framework provides an interpretable method for analyzing how gene expression and physical location interact to drive complex developmental processes.

References:

• Peng Q, Zhou P, Li T. stVCR: spatiotemporal dynamics of single cells[J]. Nature Methods, 2026: 1-12.