An integrated analytical approach for biomarker discovery in esophageal cancer: Combining trace element and oxidative stress profiling with machine learning

Abstract

Background

Early detection of esophageal squamous cell carcinoma (ESCC) significantly improves survival rates, yet reliable biochemical biomarkers for early diagnosis remain limited. The aim of this study is to identify potential early diagnostic biomarkers by integrating trace element and oxidative stress profiling with machine learning. This study investigates alterations in trace elements and oxidative stress-related biomarkers in cancerous and adjacent healthy esophageal tissues (used as paired controls) using ICP-MS and spectrophotometric biochemical assays.

Methods

A total of 28 early-stage ESCC patients were included. Concentrations of 12 trace elements (Al, Cr, Mn, Fe, Co, Cu, Zn, Se, Sb, Hg and Pb) were measured via ICP-MS. Additionally, 11 oxidative stress and antioxidant markers were analyzed: SOD, CAT, GPx, PON, ARE, MPO, MDA, GSH, TAS, TOS and OSI.

Results

Statistical analysis revealed significant increases in Cu, Fe, and TOS levels and a marked decrease in Se in cancerous tissues. Strong correlations were observed among specific trace elements and antioxidant enzymes. Machine learning models, including XGBoost, Random Forest, LightGBM, SVM and Logistic Regression were employed to classify tissue types and identify key diagnostic markers. The XGBoost model achieved the highest performance (91.7 % accuracy, AUC = 0.97), and SHAP analysis highlighted Se and Zn as the most influential variables.

Conclusion

These findings demonstrate that the combined profiling of trace elements and oxidative stress biomarkers, enhanced with machine learning, can offer powerful tools for early ESCC diagnosis.