Inferring Cell Differentiation Dynamics with Unobserved Progenitors

A cell differentiation map describes the transitions between cell types during a developmental process, and determining this map is a key challenge in developmental biology. Recent single-cell lineage tracing technologies generate lineage trees that describe the history of cell divisions during a developmental process but do not directly measure differentiation events between cell types. Current approaches to infer cell differentiation maps from cell lineage trees make unrealistic assumptions about the developmental process, do not allow for unobserved progenitor cell types, or do not infer cell-type specific rates of growth and transitions. To address these issues, we introduce TROUPE, a likelihood-based framework that infers differentiation and growth dynamics directly from leaf-labeled cell lineage trees while allowing for unobserved progenitor types via biologically motivated potency constraints. We provide an efficient algorithm for computing the maximum likelihood transition and growth rates, as well as a simple model-selection scheme to choose the number of unobserved types. On simulations, TROUPE recovers transition and growth rates better than previous approaches even when there are unobserved types. We apply TROUPE to a model of mammalian development, Trunk-Like Structures (TLS), and show that TROUPE computes reasonable rates of self-renewal and differentiation in both standard and perturbed conditions.