Virtual dynamics of spatiotemporal single-cell interactions

Abstract

Building virtual cells and tissues requires reconstructing how interacting cell populations evolve in space and time, yet spatial omics experiments provide only sparse snapshots of this process. Here, we introduce CytoBridge, a computational method that reconstructs continuous virtual cell dynamics from discrete spatial transcriptomic data. Starting from a joint transcriptomic-spatial space, CytoBridge learns stochastic trajectories of cell states, positions, and population sizes through an unbalanced mean-field Schrodinger bridge formulation that explicitly incorporates cell-cell interaction as a dynamical driver. A time-varying interaction graph built from spatial proximity and ligand-receptor priors allows the velocity to be decomposed into an interaction component estimated from observed neighborhoods and ligand-receptor features, and an intrinsic-context component that captures intrinsic regulatory processes and unmeasured contextual influences. Applied to five datasets across three spatial transcriptomics platforms and spanning development, regeneration, and neurodegeneration, CytoBridge accurately generates tissue dynamics at unmeasured time points, follows ligand-receptor signaling through measured and unmeasured stages, and supports hypothesis-generating in silico perturbation. CytoBridge provides a general method for turning heterogeneous spatiotemporal transcriptomic measurements into continuously evolving virtual tissues.