Fig. 2

Correlated lineage marker onsets of differentiating HSPCs in time-lapse microscopy allow to infer the time point of lineage choice. a The tracking of dividing single cells in a time-lapse experiment gives rise to a genealogy. Expression of the lineage marker CD16/32 (indicated in green) is detected along the genealogy. b A simple differentiation model with an exponentially decreasing differentiation probability and a delay due to stochastic gene expression up to a detection limit induces marker correlations. After the first division, both cells independently differentiate at time t₁ and t₂ and start expressing the marker, but divide before reaching the detection threshold (gray line). Inheriting the state of the mother, the daughter cells will reach the threshold at similar, but due to stochasticity in gene expression, not identical times. c As the underlying dynamics are unknown, the observed marker onsets in a genealogy can originate from different possible differentiation scenarios termed hidden trees. The hidden tree with the highest probability is the most likely scenario of differentiation. d For one observed genealogy T, all hidden trees H ∈ H(T) are constructed. A particular hidden tree can be decomposed into a single tree U that contains only undifferentiated cells, and the set of subtrees D_i whose roots are differentiating at unknown time points (t′, t″). To obtain its likelihood, each subtree D_i is represented as a graphical model and message passing is performed (see Methods)