Fig. 3: Comparison of NeuronChat, CellPhoneDB, and CellChat in predicting VISp and ALM projection networks.
From: Inferring neuron-neuron communications from single-cell transcriptomics through NeuronChat
a Typical ROC curves (left panel) and PR curves for the three methods. b The boxplots of AUROC (left panel) and AUPRC (right panel) values for 100 repeats of the aggregated VISp projection networks inferred by the three methods. Each boxplot represents 100 independent repeated computations. Boxplot elements: center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range; points, outliers. Note that no variation in each boxplot is observed because the aggregated method “thresholded_weight” reduces the fluctuation caused by finite sampling in the permutation test (see also Supplementary Fig. 9). c The number of detected interaction pairs for the three methods. d, e The boxplots of AUROC (d) and AUPRC (e) values for the individual VISp projection networks inferred by the three methods. Sample size (i.e., the number of detected interaction pairs) for boxplot: n = 50, 28, and 28 for NeuronChat, CellPhoneDB and CellChat, respectively. Boxplot elements: center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range; points, outliers. f–j Repeat analysis for ALM projection network, analogous to (a–e). Sample size for boxplot: n = 47, 25, and 25 for NeuronChat, CellPhoneDB and CellChat, respectively.
