12-Gene Signature for Prediction of Chemotherapy Response in Gastric Cancer

Background Gastric cancer (GC) remains a major global health burden with high rates of recurrence following standard treatment. Fluoropyrimidine-based chemotherapy constitutes the therapeutic backbone for GC; However, inter-patient heterogeneity in treatment response and recurrence risk limits the effectiveness of uniform management strategies. Robust molecular markers capable of stratifying recurrence risk and treatment benefit are therefore needed. Objective To develop and validate a computational gene expression–based predictive model for recurrence risk and chemotherapy response in GC patients treated with fluoropyrimidine-based regimens. Design: A retrospective, bioinformatics-driven study using publicly available datasets from the Gene Expression Omnibus (GEO). Methods Gene expression profiles from 815 chemotherapy-treated GC patients were analyzed and divided into a training cohort (n = 123) and independent validation cohorts (n = 654). Genes associated with both disease recurrence and chemotherapy response were identified across multiple datasets. Based on the expression levels of these genes, a predictive index (PI) representing the gene signature was calculated for each patient in the training cohort using linear dimensionality reduction. Patients were subsequently stratified into low- and high-risk recurrence groups according to whether their PI was below or above the cohort median, respectively. Model performance was further refined using a Support Vector Machine (SVM) classifier and evaluated through leave-one-out cross-validation (LOOCV) in validation datasets. Results We identified 12 common genes that associated with both responsiveness and recurrence risk. The 12-gene signature emerged as an independent prognostic factor for disease-free survival. In the combined validation cohort, the model demonstrated strong discriminative performance, achieving a sensitivity of 1.00 and a specificity of 0.967 in stratifying patients into low- and high-risk recurrence groups. Subgroup analyses showed consistent predictive utility in clinically relevant subsets, including patients with distant metastasis (stage IV), diffuse histological subtype, tumors located in the middle or distal stomach, and patients younger than 65 years. Conclusion This study presents a computationally derived 12-gene signature that enables robust risk stratification of GC patients receiving fluoropyrimidine-based chemotherapy. The model provides a reproducible, data-driven framework for identifying patient subgroups with elevated recurrence risk and may serve as a hypothesis-generating tool to inform future prospective validation and personalized treatment strategies.