Fig. 3: COMPOSITE multiplet detection performance in single-cell multiomics setting.
From: A unified model-based framework for doublet or multiplet detection in single-cell multiomics data
a Scatter plot displaying the relationship between the overall modality weight and the prediction performance in terms of the F1 score. Gray lines connect the three modalities from the same dataset. The upward trends indicate that the modalities that can provide better prediction performances are associated with higher overall modality weights in general. This suggests that the COMPOSITE model effectively upweights the modalities with better prediction performances. b Prediction performances (in terms of F1 score) of each modality at different ADT consistency levels. The data are from the Ileum-1 dataset. c and d Boxplots showing the performances (in terms of F1 score) of each multiplet detection method on the in-house DOGMA-seq datasets (n = 10) from peripheral blood samples (c) and DOGMA-seq datasets (n = 7) from ileum biopsy samples (d). In the boxplots, the box spans from the first to the third quartile, with the median depicted as a line in the middle. The whiskers extend to 1.5 times the interquartile range (IQR). In the labels of the x-axis, the texts within the parenthesis indicate the modalities of data that were used as input into the corresponding method. Source data are provided as a Source Data file.
