Validation of the quasi-likelihood estimation procedure with artificial datasets. We generated simulated CFSE datasets using numerical realisations of three different branching processes models of cell kinetics, and tested our estimation procedure by using these datasets to estimate the model parameters. As in Figure 1, division probabilities are represented by γ, survival without division as δ, and death as α = 1 - γ - δ. Model 1 – division and death probabilities change after the first division. Changes in parameters are indicated by different shading of cells. Model 2 – Probabilities of division and death change after one timestep. Model 3 – Resolving two subpopulations. We generated artificial CFSE profiles by adding the contributions from two branching processes – one with cell type A, in which division and death probabilities changed after first division, and one with cell type B, with constant probabilities of division and death. Type A cells were present at initial frequency fA. For each Model (1, 2, 3) we generated time series of simulated CFSE data sets by running three independent branching processes (each starting with 104 cells) and used the counts in each generation after 2, 4 and 6 timesteps as independent timepoints. This ensured that the data at each timestep were uncorrelated measurements and so would contribute additively to the log likelihood. 104 replicate timeseries were generated for each model and used with the QL procedure to estimate the parameters.