CpG feature count

MOSurvivor decreases CpG feature count in silico (-78.8% feature set, +0.054 years MAE).

“Relative to full-feature XGBoost (MAE 4.095 {+/-} 0.285 years), MOSurvivor reduced the feature set by 78.8% at an MAE increase of only 0.054 years (1.3%).”

In silicoon 30 held-out test splitswhole bloodn = 656

MOSurvivor-Guided Joint CpG Selection and XGBoost Hyperparameter Optimization for Compact Epigenetic Age Prediction

bioRxiv·29 Aug 2026Preprint