Calibrating developmental dynamics from single-cell atlases: lineage paths outrun least-action routes

In the lineage-resolved Caenorhabditis elegans atlas, observed lineage-centroid trajectories have an action ratio of 3.79 versus optimal paths, exceeding the learned model’s own forward-simulation score of 2.07.

• Why it matters: Single-cell atlases infer developmental dynamics from snapshots, but the magnitude of deviation between realized lineage paths and inferred least-action routes has lacked controlled measurement.
• What they did: LADD learns a stochastic developmental vector field from time-series single-cell RNA-seq, computes minimum-action and minimum-time paths, and uses a three-layer calibration to separate true path deviation from centroid-measurement cost.
• The result: In Danio rerio, the observed ratio matches the model’s floor, while in C. elegans residual deviation is linked to speed, with lineages moving 30–43% faster along the learned flow.

AI summary of the abstract · may contain errors · not medical advice · read the paper.