MACHINE LEARNING-BASED IDENTIFICATION OF KEY GENES AND DEVELOPMENT OF A PREDICTIVE NOMOGRAM FOR PERIODONTITIS DIAGNOSIS

Periodontitis is a prevalent chronic inflammatory disease affecting tooth-supporting tissues. Identifying molecular features that distinguish diseased from clinically healthy periodontal tissue sites may support biomarker discovery and improve molecular characterization of the disease.

Gene expression datasets GSE16134 (development set) and GSE10334 (external validation set) were obtained from the Gene Expression Omnibus database. Differentially expressed genes (DEGs) were identified using |log₂fold change| > 2 and P<0.05. Nine machine learning algorithms-Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), LASSO, Gradient Boosting Machine (GBM), Generalized Linear Model (GLM), XGBoost (XGB), Decision Tree (DT), and Neural Network (NNET)-were applied to screen feature genes. The top five RF-ranked feature genes were used to construct a nomogram model. Model performance was evaluated using ROC curve analysis, decision curve analysis, and calibration curves.

We identified 12 DEGs, including 2 upregulated (AADACL2, DSC1) and 10 downregulated genes. RF and LASSO demonstrated the strongest and closely comparable discrimination among the nine algorithms. Five critical genes (IGHM, MZB1, DSC1, AADACL2, CXCL6) were identified. The integrated nomogram achieved an AUC of 0.900 (95% CI: 0.842-0.949) in the external validation dataset. Individual genes also showed satisfactory discriminative performance (AUC: 0.816-0.875).

This study established a machine learning framework for identifying periodontitis-associated genes and developed a five-gene nomogram with strong discriminative performance. The identified genes represent candidate molecular biomarkers, and the integrated signature may assist discrimination between diseased and clinically healthy periodontal tissue sites. Further patient-level and prospective validation is warranted before clinical implementation.