Development of a computational method to identify T cells that had deleted Pdcd1. (A) Schematic for hypothetical cell types (based on PD-1 protein and transcript expression) present in tumors from WT and inducible PD-1^del. mice, and the development of a computational method to distinguish true PD-1^KO (e.g., cells that have no PD-1 protein and a deletion of the Pdcd1 gene) from PD-1^Lo (e.g., cells that have no PD-1 protein but an intact Pdcd1 gene) in the single-cell dataset. (B) Integrative Genomics Viewer tracks showing the mapping of reads to the individual exons of the Pdcd1 gene in the scRNA-seq dataset. Samples divided by genotype and PD-1 protein status as indicated in this figure. The locations of the five exons and loxP sites are indicated. (C) Contour plots showing detection of exon 2 and exon 5 in CD8⁺ TILs in the scRNA-seq dataset. Plots split up by genotype (WT versus inducible PD-1^del.) and PD-1 protein status. (D) Logistic regression plot indicating the trained classification model with a solid blue line, with the 95% confidence interval indicated with translucent green shading. Horizontal axis indicates the linear predictor function of the logistic regression, and vertical axis indicates the PD-1 protein status (based on FACS-based sorting or CITE-seq) with random vertical noise added for visibility of overlapping points. (E) ROC curve plot indicating the logistic regression model accuracy for predicting KO status, where PD-1 status (based on FACS-based sorting or CITE-seq) is taken as ground truth. AUC/ROC is indicated (AUC = 0.93). (F) Heat map for the confusion matrix showing the number of cells predicted to be KO based on transcript versus non-KO, compared with whether the cell was PD-1 protein–positive or protein–negative (based on FACS-based sorting or CITE-seq). Shown are only cells from the inducible PD-1^del. TME. (G) UMAP plot showing the distribution of CD8⁺ TILs from the inducible PD-1^del. mice that were low for the PD-1 protein (e.g., PD-1^Lo based on FACS-based sorting or CITE-seq) and either predicted to be KO based on the distribution of reads across the five exons (left, shown in black) or predicted to be non-KO (also referred to as WT) (right, shown in black). Gray dots indicate the rest of the cells in the dataset that do not fit that definition. (H) Volcano plot showing differential gene expression between PD-1^Lo CD8⁺ TILs from inducible PD-1^del. mice (based on protein expression) that were predicted to be Pdcd1^KO or Pdcd1^WT based on transcript information. Select genes are highlighted. Full list of differentially expressed genes can be found in Table S9. For B–H, shown are cells from two independent experiments representing four mice per genotype. The schematic in A was created with https://BioRender.com.