Fig. 2: Benchmarking NeuronChat on projection networks of two mouse cortex regions: VISp and ALM.
From: Inferring neuron-neuron communications from single-cell transcriptomics through NeuronChat
a Projections from seven cell types of VISp to their cortical target regions, identified by retrograde labeling3. b Aggregated and binarized intercellular communication network for VISp. The false-positive links represent those predicted by NeuronChat but not identified by retrograde labeling. The false-negative links are those identified by retrograde labeling but missed by NeuronChat prediction. Threshold for binarization is 0.028 (normalized by the maximum) in this plot. c ROC curve for the NeuonChat-inferred aggregated intercellular communication network for VISp, with AUROC indicated on the top. The gray dashed line represents the random classifier with an AUROC of 0.5. The color bar indicates the binarization threshold. d PR curve for NeuonChat-inferred aggregated intercellular communication network for VISp, with AUPRC indicated on the top. The gray dashed line represents the random classifier with an AUPRC equal to the fraction of links identified by retrograde labeling (i.e., 14/21). e AUROC (left panel) and AUPRC (right panel) for the inferred VISp projection network of each individual interaction pair. f–j Repeat analysis for ALM, analogous to (a–e). Threshold for binarization is 0.165 for (g).
