At the core is the idea that: 👉The local neighbourhood of a cell (which cell types are nearby) affects its probability to divide or die. 👉By learning how neighbourhood composition correlates with division and removal markers across many cells, the authors reconstruct a dynamical model.

👉 From that model, they can simulate how the tissue would evolve from the snapshot over time. From a statistical perspective, the validity of the model is based on many implicit assumptions about the spatio-temporal dynamics, that I believe are difficult to assess in practice.

In spatio-temporal statistics, one usually expects replicates over time to obtain meaningful inferences. Cross-sectional variation does not substitute for temporal variation. And of course, the analysis neglects heterogeneity within the datasets. Some of these are recognized in the limitations.