Fig. 4
KPNN analysis of T cell receptor (TCR) stimulation. a Receiver operating characteristic (ROC) curves for the TCR KPNN, predicting TCR stimulation based on single-cell RNA-seq profiles with different levels of dropout. The inset shows the mean ROC area under curve (AUC) values at different dropout rates (measuring prediction performance) as well as the mean correlation across replicates (measuring network robustness). b Differential node weights at a dropout rate of 10%, comparing networks trained on actual data (x-axis) and on control inputs (y-axis). Nodes with padj below 0.05 are shown. c Trained TCR KPNN with the subnetwork of significantly differential nodes (padj < 0.05, dropout rate = 10%) highlighted in red. d Log fold change (LogFC) of gene expression for TCR regulators identified by the KPNN. Commonly used thresholds for differential expression (fold change = 1.5 and fold change = 0.66) are indicated in purple