#14 · Single-cell

Inferring temporal cell-state dynamics from snapshot single-cell and spatial data

The dominant thread is inferring temporal dynamics from snapshot measurements in which time is not directly observed. MIOFlow 2.0 couples neural SDEs with unbalanced optimal transport growth rates across single-cell and spatial snapshots, while GOTFlow applies graph-constrained unbalanced transport in a learned latent space to cross-sectional data. VeloVAE provides a Bayesian RNA velocity model incorporating timepoints, lineage bifurcations, and count data, and a unified dynamical-systems and control framework recasts pseudotime, RNA velocity, optimal transport and Schrodinger-bridge methods within one stochastic formulation. What is new is explicit temporal anchoring, through the Nr4a3-Tocky timer reporter combined with scRNA-seq in mCanonicalTockySeq, and noise-limited evaluation: scTimeBench benchmarks nine time-aware trajectory methods, and a circadian study shows transcriptional noise bounds phase inference from snapshots, with robust estimates requiring averaging across about 70 cells.