RESUMO
【Objective】 To select and identify miRNA signatures to predict TMB level in gastric cancer based on The Cancer Genome Atlas (TCGA) database and machine learning methods. 【Methods】 MiRNA expression and somatic mutation profiles of gastric cancer (GC) were downloaded from TCGA database. R "limma" package was performed to select differentially expressed miRNAs between high-TMB and low-TMB groups. Two machine learning algorisms, random forest (RF), and Support Vector Machine-Recursive Feature Elimination were utilized to identify miRNAs with the highest discriminative ability. ROC was used to test the predictive ability of these signatures in multiple datasets. Besides, immune cells of different TMB levels were compared by the CIBERSORT method. 【Results】 A total of 56 differentially expressed miRNAs (DE-miRNAs) were filtered. Functional enrichment analysis showed that these DE miRNAs are mainly enriched in signaling pathways related to tumor occurrence and development as well as immunity-related biological processes. The RF and SVM-RFE algorithms jointly identified 10 diagnostic features of miRNAs, among which only hsa-miR-210-3p is considered the most relevant predictive biomarker for TMB classification. The AUC value of hsa-miR-210-3p in the training, testing, and total sets is 0.822, 0.721, and 0.793, respectively, and has been validated in other cancer types. Besides, CIBERSORT analysis suggests differences in immune cell infiltration between high- and low-TMB groups. Meanwhile, there is a significant positive correlation between the expression of immune checkpoint related genes and mismatch repair related genes and hsa-miR-210-3p. 【Conclusion】 This study successfully identified hsa-miR-210-3p as a predictive biomarker for TMB classification, which can effectively predict TMB values in gastric cancer and other cancer patients and may provide some guidance for immunotherapy.