Fig. 1: Immune landscape of different GC subtypes.

a Violin plots of immune infiltration in EBV⁺MSS (n = 26), MSI-H (n = 45), and EBV-MSS (n = 190) patients in TCGA-STAD cohort estimated by cibersortX. Two-tailed Mann-Whitney U test was used to evaluate statistical significance between every two subtypes. b Multiplex IHC images of an MSI-H GC patient (GC-44), an immune-ignorant MSS GC patient (GC-60), and an immune-hot MSS GC patient (GC-66). Red denotes CD4⁺ T cells; yellow denotes CD8⁺ T cells; cyan denotes tumor cells, DAPI was used as the counterstain for cellular nuclei. c UMAP plot of 118,887 cells in 24 gastric tumors (GSE183904). Endo. is the abbreviation for endothelial cells; DC for dendritic cells; and epi. for epithelial cells. d Stacked bar plots of cell fractions in MSI-H (n = 6) and MSS GC tumors (n = 18). Each color denotes a cell type, as indicated in Fig. 1c. e Volcano plot of differential expression genes between CD8⁺ T cells of MSI-H GCs and those of MSS GCs. The y-axis denoted the adjusted P value, and the x-axis denoted the log2 fold change. Genes with a |log2 fold change | ≥ 1.5 and a P value ≤ 1 × e-5 were shown in red. P values were calculated with Wald test using DESeq2 and adjusted using Benjamini-Hochberg method for multiple comparisons. f Schematic graph of the theory of positive-unlabeled learning. Original positive and unlabeled data points were shown as red and gray dots in the original image. Unlabeled positive (blue dots) and unlabeled negative (yellow dots) data points were identified using positive-unlabeled learning. g Pipeline of constructing the positive-unlabeled learning models for RNA-seq data (TCGA-STAD, PRJEB25780, and PRJEB40416). The model uses MSS patients in TCGA-STAD as training dataset, and patients in PRJEB25780 and PRJEB40416 as validation datasets. Detailed methods were described in the method section. h The receiver operating characteristic (ROC) curve presents false-positive rates against true-positive rates of predicting responders versus non-responders in MSS patients of PRJEB25780 cohort (validation dataset) based on PUL scores (AUC = 0.924). i T-SNE plot of MSI-H&TMB-H&TILs-H (n = 40), MSS-PR (n = 46), and MSS-NR (n = 238) patients in TCGA-STAD cohort. Geographical proximity between data points denotes their similarity in immune features. Proximity between MSI-H&TMB-H&TILs-H subtype and MSS-PR subtype was observed. On the right, heatmap of PUL scores of MSI-H&TMB-H&TILs-H (n = 40), MSS-PR (n = 46), and MSS-NR (n = 238) patients in TCGA-STAD cohort. *P  <  0.05; **P  <  0.01; ***P  <  0.001; ****P  <  0.0001; ns is abbreviation for not significant.