Fig. 6: Validation of HighStem hub genes and evaluation of machine learning classifier performance.

A Expression density plots of five HighStem hub genes in all cells. B Expression density in malignant epithelial cells, showing spatial overlap with stemness scores. C ROC curves show the single-cell classification performance of each gene. D Benchmark comparison of eight machine learning models. E Precision–recall and ROC curves of top-performing models. F ROC and PRAUC of the SVM model in development and test cohorts. G Confusion matrices of the training and test sets. H Decision curve analysis of the final classifier. I SHAP summary plot showing the contribution of each gene. J Predicted probability plots based on normalized gene expression. K SHAP dependency plots illustrate the relationship between feature values and model output.