Introduction: Immune checkpoint therapies (ICT) restore the immune system's ability to fight cancer by blocking inhibitory signals of T cell activation, allowing them to recognize and attack tumor cells. Representative examples include drugs targeting PD-1, PD-L1, and CTLA-4 proteins. Despite active research into ICT, many patients remain untreated. The response to ICT vary across cancer types and even within cancer types, depending on immune subtype. Therefore, we aimed to define a new immune signature to identify responders and non-responders in individual cancer types. We developed a model using data from 11 cancer types from the TCGA as a training dataset and validated this model with other data sets. Methods: Using TCGA data for 11 cancer types (BLCA, BRCA, COAD, UCEC, ESCA, HNSC, KIRC, LIHC, LUAD, SKCM, STAD), we divided them into immunologically hot group, cold group, and intermediate group, according to the expected effects of ICT. To divide the data into three groups, we used immune-related pathways, immune markers from literature, Cibersort, Ecotype, and Methylation data as features. Lasso was used to select important features, and Kmeans was used for clustering. To test the results, Silhouette score, CH index, and Dunn index were used. The results of these coefficients decreased further when methylation data and ecotype features were added, so these two features were finally removed and clustered. We then created a new gene set signature using leading edge genes/markers that overlapped across these features. We tested this gene set signature on melanoma data and confirmed that it effectively distinguished responders from non-responders. Results: A total of 5,089 samples corresponding to 11 cancer types were selected, and 38 pathways were selected as features, and it was confirmed through a heatmap that they were well divided into 3 groups. A new signature was discovered with a total of 84 genes, including leading edge genes that appear repeatedly in this pathway and markers corresponding to the signature. This confirmed that the discovered gene set was well-separated across each cancer type. For further validation, we used the GSE91061 dataset to examine whether the new gene set differed between the responder and non-responder groups using a NES plot. The positive enrichment score was biased toward the responder group, and the padj value was significant at 0.028, consistent with the pattern observed in TCGA. When ORA was used with Reactome and GOBP, immune-related pathways were also significantly identified. Conclusion: Given the close connection between ICT and immunity, we believe this new gene set signature will be helpful in distinguishing between ICT-responsive and non-responsive groups. We plan to conduct validation tests using other cancer type data and WSI on immunologically hot and cold tumor groups to further improve accuracy.