Fig 1: Validation of proteomic signatures and risk prediction models for gastric lesion progression. a Associations of four proteins with early GC in combined validation sets and association of risk score with progression of gastric lesions in prospective follow-up subjects. b. Receiver operating characteristic curve of random forest classifier prediction model for progression of gastric lesions during the follow-up. Performance of the models was estimated by leave-one-out cross-validation method. Model-1 includes age, sex, H.pylori infection and baseline pathology. Model-2 includes variables in model-1 and risk score of four proteins. Model-3 includes all variables in model-2 and molecular subtypes. c. Immunohistochemistry staining of four proteins in formalin-fixed, paraffin-embedded tissues (×40 magnification). All four proteins are predominantly stained in cytoplasm, consistent with the Human Protein Atlas report (https://www.proteinatlas.org/ENSG00000110169-HPX/pathology/stomach+cancer#Location for HPX, ENSG00000096088 for PGC, ENSG00000099977 for DDT, and ENSG00000163382 for APOA1BP). Wilcoxon rank-sum test, ns=non-significant, *P < 0.05, **P < 0.01, ***P < 0.001. AUC, area under the curve; CAG, chronic atrophic gastritis; CI, confidence interval; GC, gastric cancer; HGIN, high-grade intraepithelial neoplasia; IM, intestinal metaplasia; LGIN, low-grade intraepithelial neoplasia; LOOCV, leave-one-out cross-validation; OR, odds ratio; SD, standard deviation; SG, superficial gastritis.