Background: Gastrointestinal stromal tumor (GIST) is the most common sarcoma, often originate in the stomach or the small intestine. Although the application of imatinib has brought a breakthrough in the clinical treatment of GIST, not all GIST patients are effective, and most GIST patients who receive imatinib sensitivity will develop drug resistance and eventually malignant progression. The tumor microenvironment (TME) is a key mediator of cancer progression and treatment outcome, and plays an important role in predicting cancer prognosis and treatment response. This study aims to investigate the role of TME in the treatment of GIST and the response to immunotherapy. Methods: Raw microarray data of 199 GIST patients were downloaded from GEO, including 159 KIT-mutant, 21 PDGFRA-mutant and 19 WT. All affymetrix datasets were re-normalized using the gcRMA package with default parameters. We integrated 275 TME and malignant signaling pathways related signatures from different database and previous researches, including CIBERSORT, MCP-counter, TIMER, and others. Fisher exact test's was used to identify signatures that differential between KIT-mutant, PDGFRA-mutant, and WT (p < 0.05), and 85 signatures were finally obtained. To classify the molecular profiles, the expression patterns were assessed in all 199 GIST patients using unsupervised clustering based on their ssGSEA scores, and can be clustered into three distinct microenvironments. Results: Using GSEA with all TME signatures genes ranked by the log2(fold change) between the threee clustes, we found that gene sets were considered immune favorable in cluster A, including gene sets related to activated CD8+ T cells (p = 0.0003)and activated CD4 T cell signaling pathway (p = 0.009). Cluster B enriched to the epithelial mesenchymal transformation pathway (p = 0.04) and TGFb signaling pathway (p = 0.003), cluster C enriched into DNA-repair pathway (p = 0.02). In an independent skin cutaneous melanoma cohorts (n = 86) treated with anti-PD-1 therapy, patients were classified into three molecular clusters, which significantly corrleated with response to ipilimumab (p = 0.007). The percentage of responders to anti-PD-1 therapy in cluster A was 75% in contrast to only 10% in cluster B. The three molecular clusters discern a potential responder versus non-responder (AUC = 0.87; p = 0.0015), with essentially a 80% chance of correctly predicting the response, which supporting the utilization of the classification method. Conclusions: In this study, the immuno-infiltrating landscape was characterized based on pan-immune signatures using transcriptomic profile in GIST. The molecular subtypes derived from pan-immune signatures have the potential to predict the efficacy of immunotherapy.