Fig. 1

The profile of cell free immune-related circulating miRNAs (cf-miRNAs) between malignancies and non-malignancies. (a) Workflow of the study. (b) Distribution of the number of samples, histological type, age, and sex in 10 GEO datasets. Clinical samples include lung cancer (LUCA, n = 1606), esophageal cancer (ESCA, n = 601), gastric cancer (STAD, n = 1447), Liver hepatocellular carcinoma (LIHC, n = 466), colorectal cancer (COADREAD, n = 272), breast cancer (BRCA, n = 1285), prostate cancer (PRAD, n = 809), pancreatic cancer (PAAD, n = 227), ovarian cancer (OV, n = 327), bladder Cancer (BLCA, n = 392), sarcoma (SARC, n = 591), glioma (n = 212), biliary tract cancer (CHOL, n = 81), and 7516 non-cancer individuals (health, other diseases, and benign tumors). (c) Principal component analysis (PCA) analysis of malignancies and non-malignancies based on differentially expressed miRNAs. (d) Circos plot showing the differentially expressed miRNAs immune pathway among malignancies. The inner heatmap showed the expression of miRNAs across cancer types. (e) Heatmap showed a significant difference between malignancies and non-malignancies based on 39 cf-miRNAs in the validation set. (f) Youden index of each classifier in the validation set. The X-axis is five types of machine learning algorithms and the Y-axis is Youden value. The redder means a higher value. (g) Youden index (left) and area under curve (AUC) (right) performance for each classifier