LineageFlow: Growth-Corrected State Dynamics for Probabilistic Cell-Fate Prediction

Abstract

Single-cell lineage tracing links early transcriptional states to the fates of their descendants, but forecasting clonal fate remains challenging because differentiation and unequal cell expansion jointly shape descendant populations. Here we introduce LineageFlow, a computational framework that combines continuous-time cell-state transitions, relative net growth and state-conditioned generative flows to predict future fate distributions and transcriptional profiles from early single-cell measurements. In clone-disjoint evaluations of lineage-traced hematopoiesis, LineageFlow improves fate prediction and expression-distribution accuracy over matched computational baselines. A separately retrained analysis of cord-blood lineage data shows a similar advantage at a longer prediction horizon. Controlled simulations show that accounting for differential growth improves fate prediction when descendant states expand at different rates, while offering little benefit under equal growth. These findings highlight the importance of modeling state transitions and relative expansion together when predicting the fate of unseen clones.