Fig. 6: TF knockout experiments and their effect on cell type generation.

A The UMAP shows the distribution of n = 2000 randomly selected cells from the experimental PBMC-All dataset. The color shows the iLISI value of each cell (miLISI = 1.94) calculated from a UMAP embedding, jointly obtained from the experimental cells and the same number of GRouNdGAN-generated cells. TFs were omitted as features when generating UMAP plots. In the simulated data, each gene was regulated by 15 TFs, identified using GRNBoost2. B Each plot shows the density of cells in the embedding space: the left plot (blue) corresponds to the real cells and the right plot (red) shows the simulated cells. C The boxplots show the distribution of iLISI values of n = 2000 real cells, calculated along with n = 2000 unperturbed simulated cells (blue) and with n = 2000 perturbed simulated cells (orange). For each cell type, top three most differentially expressed TFs were knocked out. The p-values reported in this figure were calculated using one-sided Wilcoxon signed rank tests. The center horizontal line corresponds to the median. Each box spans the interquartile range, and its limits show the upper (Q3) and lower quartiles (Q1). Whiskers extending from the boxes reach 1.5 times above and below the interquartile range. Outliers are shown individually if they fall outside the whiskers. D The scatter plot shows the iLISI values of CD19 + B cells calculated along with unperturbed simulated cells (x-axis) and along with perturbed simulated cells (y-axis). The circles correspond to real cells and their colors reflect the density of datapoints in that region. The p-value reported in this figure was calculated using a one-sided Wilcoxon signed rank test. E The UMAP embedding of real cells (left), simulated cells after knockout of top TFs of CD19 + B cells (middle), and all together (right). Source data are provided as a Source Data file.