DS-MCGR: A Dempster-Shafer Combination Method Integrating Molecular Characterization and Graph Representation for Identifying Active Compounds Related to Esophageal Cancer

Introduction: Esophageal cancer is among the most lethal malignancies of the gastrointestinal tract, posing a serious threat to human health. Network pharmacology has provided novel insights into the pathogenic mechanisms and therapeutic strategies for esophageal cancer. However, accurately and efficiently identifying active compounds using network pharmacology remains challenging, particularly when analyzing complex herbal prescriptions. Methods: To address the limitations in current virtual screening algorithms for identifying the active compounds related to esophageal cancer from traditional Chinese medicine (TCM) prescriptions, this study proposes a new algorithm named DS-MCGR, which integrates Dempster-Shafer evidence theory with Molecular Characterization and Graph Representation. This virtual screening algorithm comprises three steps: building a standard database, training the model, and screening active compounds in prescriptions. Results: Extensive experiments were conducted using 3-fold, 5-fold, 8-fold, and 10-fold crossvalidation on the collected datasets. The results show that DS-MCGR consistently outperforms 11 classical machine learning algorithms and 7 deep learning models across most evaluation metrics. Under 3-fold cross-validation, DS-MCGR achieved the best performance with a Specificity of 0.9717, a Kappa of 0.9167, an MCC of 0.9173, an F1 score of 0.938, a PR-AUC of 0.9746, and an ROC-AUC of 0.9889. The sensitivity was 0.958, ranking second. Discussion: Case studies on pomegranate and Salvia miltiorrhiza further validated the practical utility of DS-MCGR. The model could accurately identify the active compounds for treating esophageal cancer. Conclusion: DS-MCGR significantly improves prediction accuracy by integrating chemical and structural features, offering a promising tool for discovering anticancer components from TCM and advancing drug discovery for esophageal cancer.