Fig. 2: Comparison of the doublet detection approaches on the ground-truth datasets. | Communications Biology

Fig. 2: Comparison of the doublet detection approaches on the ground-truth datasets.

From: Chord: an ensemble machine learning algorithm to identify doublets in single-cell RNA sequencing data

Fig. 2

a DM-A and HTO8 tests were evaluated by nine methods and their ROC curves. b The AUCs of the nine methods on seven datasets and the heatmap of the AUC results. The number in the heatmap indicates the rank of the method in the dataset (only the top three methods are marked). c The standard deviation of the rank values for each method, the mean of the rank values of each method, and the mean AUC of each method across seven datasets. d Heatmap of average pAUC800 pAUC900, pAUC950 pAUC975, AUC and PR in DM-A, DM-B, DM-C, HTO12, HTO8, DM-2.1 and DM-2.2 dataset. The number in the heatmap indicates the performance rank of methods in the dataset (only the top four methods are marked). e By random sampling from the real DM-2.1 dataset, simulated datasets with varying numbers of cell number (from 1000 to 12,000 cells with an interval of 1000) were constructed, and the runtime of each method on the simulated datasets was recorded using the same computer server (E5-2678v3 CPU processor and 256 GB memory).

Back to article page