Simulation-Based Machine Learning Approach to Classify Accelerated Biological Aging Using Telomere and Epigenetic Markers

Accelerated biological aging characterized by loss of telomeric iterations, epigenetic modification and persistent inflammation underlies increased risks of age related diseases, but benchmarking with machine learning (ML) methods for the classification of accelerated biological aging has been hindered by the scarcity of standardized datasets. In this research study, a synthetic cross-sectional dataset was developed consisting of 1000 commonly representing suburb ages of individuals representing normal and accelerated age profiles features simulated according to medical guidelines and research distributions, including chronological age, telomere length, epigenetic age, inflammatory markers (CRP, IL-6), senescence marker (p16INK4a), white blood cell counts. Multiple ML classifiers such as Random Forest, Gradient Boosting, Extra Trees, XGBoost and LightGBM, were evaluated using 5-fold cross validation, hyper parameter tuning, and probability calibration. All models showed best predictive performance with Extra Trees giving the best overall accuracy (98.5%) with runtime 14 seconds respectively, whereas XGBoost had a slightly lower accuracy (97.5%) but was the fastest at 3.76 seconds. This simulation based framework has the potential to contribute in a reproducible platform for assessing ML approaches for aging research and additionally underscores the potential of computational models in identifying accelerated aging and the need to evaluate the findings on large scale biological cohorts.