Development of a precise saliva-based epigenetic clock using a six-CpG-marker panel.

Saliva, a frequently encountered biological specimen in forensic investigations, has great potential to provide critical leads if the donor's age can be accurately estimated. This study aimed to develop a precise epigenetic clock for predicting age by identifying the most highly informative CpG sites derived from saliva. Using genome-wide 850K methylation array data from saliva samples of 76 Han Chinese individuals, we initially selected 33 candidate CpG sites based on correlation analyses and an extensive literature review. We then applied a multi-stage feature selection process, performing 250 iterations across five machine learning algorithms running in parallel: Elastic Net, Lasso, Ridge, Support Vector Regression (SVR), and robust Lasso (rLasso). By using recursive feature elimination (RFE) optimized to reduce mean absolute error (MAE), we identified an optimal set of six core CpG sites (cg09809672, cg10501210, cg26290632, cg04875128, cg06639320, and cg07553761) that showed strong associations with age. Among the models tested, rLasso and SVR delivered the best and comparable predictive accuracy, with MAEs of 3.240 and 3.267 years, respectively, which were significantly better than the other linear models. Validation on three independent datasets and comparison with seven established epigenetic clocks showed that our models achieved accuracy comparable to leading methods, like the Hannum and Horvath clocks, while substantially outperforming others such as the Levine, PedBE, and Wu clocks. Moreover, the MAEs of our models were lower than those reported for 85% of the 60 previously published saliva-based epigenetic clocks. In summary, we developed a highly accurate epigenetic age prediction panel for saliva that includes only six CpG sites. The resulting rLasso and SVR models provide an excellent balance of high predictive accuracy and practical applicability, making them valuable tools for forensic age prediction and aging-related research.