Fig. 6.

Comparison of MERFISH data and synthetic data generated by scDesign2, its variant without copula, ZINB-WaVE, and SPARSim in 2D visualization. a t-SNE plots and b principal component (PC) plots of training data, test data, synthetic data generated by each simulator, and combinations of test data and each synthetic dataset. Note that in b, the coordinates are defined by projection to the PC space of the test data. Gene expression counts are transformed as log(1+count) before dimensionality reduction. miLISI is short for median integration local inverse Simpson’s index, a higher value of which indicates that the simulated data mix better with the test data in the 2D visualization plot. By visually inspecting the patterns in these plots as well as comparing the miLISI values, we find that the synthetic data generated by scDesign2 most resemble the test data