Fig. 2

Multi-cohorts validation and functional characterization of plasma EV-derived rs/tsRNA signature in gastric cancer. A qRT-PCR analysis of six candidate sncRNAs in validation cohort (119 GC patients vs 143 healthy controls). Data are shown as the mean ± SEMs. Significance determined by unpaired two-tailed Student's t-test. ****P < 0.0001. B ROC analyses of the three-rs/tsRNA signature across training and validation cohorts (143 GC patients vs 167 healthy controls). C Feature importance ranking by mean decrease Gini index of three rs/tsRNAs(S2/S7/S10), four clinical protein indices (CA19-9, CA125, CEA, AFP), and their combination, as determined by random forest analysis. D ROC analysis of the performance of three rs/tsRNAs, four clinical protein indices, and all index combinations in the validation set of machine learning model. E Confusion matrix showing the classification accuracy of three rs/tsRNAs, four clinical protein indices, and their combined use in distinguishing GC patients from healthy controls through machine learning. F Validation in two independent test cohorts (Test Cohort 1: 47 GC vs 47 controls; Test Cohort 2: 40 GC vs 43 controls). Data are shown as the mean ± SEMs. Statistical significance was assessed using unpaired two-tailed t tests, ***P < 0.001; ****P < 0.0001.G ROC curves demonstrating diagnostic values of three rs/tsRNAs in two test cohorts. H Functional validation of rs/tsRNA-target interactions via dual-luciferase reporter assay. Data are shown as the mean ± SEMs. Statistical significance was assessed using unpaired two-tailed t tests, *P < 0.05; **P < 0.01; I Western blot verification of ErbB/Hippo pathway modulation in GC cells. MKN-45 and AGS cells transfected with sncRNA mimics and scramble negative control RNA, respectively. GAPDH served as internal control. Quantification represents mean ± SEM from three independent experiments