Extended Data Fig. 10: Diagram illustrating the method section.

a, b, Dot plots of t-SNE dimension reduction of 10 ICB clinical treatment datasets. SRP128156 including two sub-datasets: G100556 (red circle) and G102647(blue circle), t-SNE dimension reduction shows SRP128156 has an obvious batch effect (a). t-SNE dimension reduction shows the results of batch effect removing by the ComBat function in SVA (R package) for SRP128156 dataset (b). c, Details on the identification of ICB-related gene sets. The Antigen processing and presentation pathway was taken as an example: In 7 of the 10 datasets, ‘Antigen processing and presentation pathway’ results are positively enriched significantly, and the corresponding 7 leading-edge subsets are obtained. If the gene appears in more than 70 % of leading-edge subsets (7 × 70% = 4.9, that is, more than or equal to 5 leading-edge subsets), the gene is retained. Otherwise, the gene is discarded. (P-values were calculated by two-tailed Kolmogorov Smirnov test, then FDR (BH) for correction of p-values.) d,e, Overview of the LASSO model test data. 23 positive samples for test data (samples from NCBI GEO database, GEO accession number: GSE218603, GSE173107, GSE186195, GSE223110, and GSE162935, detail in Table S7) (d). 2573 negative samples for test data (GSEA was conducted on the gene expression data perturbed by small-molecule compounds from the LINCS 2020 dataset (e). Samples exhibiting negatively significant enrichment (NES < 0, FDR < 0.05) in the Core & Minor gene sets following perturbation are considered ‘negative samples’, and their efficacy labels are set to ‘0’.).