Fig. 4

Performance of deMULTIplex2 on multiplexed PDX of human breast cancer. A UMAP computed with gene expression (GEX) colored by sequencing batch, expression-based tumor ID, and classification results by deMULTIplex and deMULTIplex2. For major tumor clusters, we highlight the percentage of correctly classified singlets. B Fraction of cells correctly predicted by deMULTIplex2 and other methods for each tumor model. All methods were run with default parameters. Demuxmix was excluded from the comparison because it returned errors on all three batches. C UMAP of cells from batch 3 computed using raw tag UMI counts. Circles highlight two samples that were missed by deMULTIplex with default settings, but were recovered with deMULTIplex2. D Cosine similarity vs. tag count plot and the tag count vs. total tag count plot for the sample tagged with “Bar2” and from tumor “HCI011” recovered by deMULTIplex2. The y = x line is shown in black. E GLM-NB fit (grey line) and posterior probability of cells being positively tagged by tag “Bar2” calculated by deMULTIplex2 in the two modeling spaces. The y = x line is shown in black, and majority of negative cells fall on or near that line in the second space. N: negative cells, P: positive cells. F RQR plotted against log total tag count, colored by posterior probability. For some positive cells, its RQR is infinity. These values were capped to the maximum value of non-infinity RQRs plus 1 for visualization purposes. The Q-Q plot compares the distribution of RQRs of predicted negative cells to that of a normal distribution. G Heatmap summarizing the rank of F-score of deMULTIplex2 and other methods on both simulated and real-world datasets. Dataset abbreviations are same as those in Fig. 3A. Winkler is the multiplexed PDX dataset from Winkler et al. [33]. NA indicates the method cannot be run on the corresponding datasets due to the unavailability of mRNA count matrix or an error