Fig. 3

Results of the application of EmptyDropsMultiome on sample A after quality control. A, B Histograms of ATAC and RNA counts of all the droplets. We profile the soup using only the ambient cluster in ATAC and RNA. The lower and upper bound of the cluster are marked by the purple and blue line and are deduced via a Gaussian mixture model. C Venn diagram showcasing that the droplets selected by CellRanger-arc (cR) are almost a subset of those selected by EmptyDropsMultiome (eD). D Histogram of the FRiP for all droplets selected by CellRanger-arc or EmptyDropsMultiome after quality controls. A very large number of droplets retained by EmptyDropsMultiome has FRiP below the threshold imposed by CellRanger-arc. E The distribution of droplets selected by EmptyDropsMultiome in the space of logarithmic counts. A very small number of droplets selected by EmptyDropsMultiome lies below the k-means line (middle line shown) and is likewise excluded due to CellRanger-arc’s hard cut-off on library size. For upper and lower lines, see the “Methods” section. F The histogram of the linear combination of logarithmic ATAC and RNA counts that is constant along the k-means line. Again, we see a small number of droplets that CellRanger-arc misses due to this hard cut-off