Tracking performance. a. Cumulative error accumulation along tracks over time. Mean and variation is shown over embryos in cumulative accuracy, specifically the fraction of AB (from AB8 to AB256) cells correctly tracked. Experimental results are in blue and show the fraction of cells tracked correctly from the beginning through each generation. Each gray line is the result of simulation of cumulative error over the same time period. A uniform average error rate and average cell cycle length is assumed. Errors are introduced at each cell and frame in the simulated lineage at the instantaneous error rate, and a cell is considered wrong if any errors have been introduced during its history. b. Initial cell detection error counts (FN + FP), and counts after bifurcation resolution as a function of cell density. Density is measured as the inverse of space between nuclei. For cell c
density = (radius(c
) + radius(NN(c
,t)). A density of one means no gap between the nuclei is present. c. Tracking error count (FN + FP) in the tentative bifurcation set and in the final results of our method. This is contrasted with tracking errors present in a more naive link likelihood maximization scheme. As above both error counts are displayed as a function of cell density. d. Difficult cases. Multiple detection errors in the same area presents difficulty for our method. In the first case the confluence of a FN and FP make it impossible to generate the exact correct answer. In the second example a division and FN together violate our methods assumption that both ends of a FN will be marked by a change in network topology, making the correct FN resolution impossible.