Fig. 2

Performance of deMULTIplex2 on simulated datasets. A Five simulations with different data size, complexity and noise, as listed in Additional file 2: Table S1. The top row shows UMAPs computed with raw tag UMI count. The middle row shows UMAPs based on deMULTIplex2-computed RQRs. The bottom row shows the cell distribution in the log(tag umi) vs log(total tag count) space of a randomly selected tag in each simulation, colored by ground-truth identity and deMULTIplex2 computed posterior probability of being positive for that tag. The y = x line (grey) and the GLM-NB fit for the negative cells (steel blue line) are also plotted. B Heatmap summarizing the F-score of deMULTIplex2 and other methods on the five simulated datasets. Methods that require mRNA count matrix as input were excluded from this comparison. NA indicates the method failed to run. C Per-tag performance of deMULTIplex2 compared to other methods. Mean values are highlighted with the diamond points. D Performance of all methods on the simulated datasets with down-sampled reads. Median total tag count per cell is listed for each down-sampling rate. E Performance of all methods on the simulated datasets with down-sampled cells. Median cell number per sample is listed for each down-sampling rate