DEPH_Collins/Ajaz Female Saliva Methylation Clock: a 1,000-CpG chronological age predictor for female saliva
Authors/Creators
- 1. Muhdo Health Ltd
- 2. deph labs
- 3. King's College Hospital
Description
DEPH_Collins/Ajaz Female Saliva Methylation Clock (version 1.1.0)
A first-generation chronological DNA methylation clock for female saliva, trained at DEPH labs on a Muhdo Health Ltd. kit extract. Authors: Christopher Collins and Saima Ajaz. Licence: CC BY 4.0.
Formula. From 2,500 kit CpGs, the 1,000 sites with the largest absolute Pearson correlation with age were refit with ridge regression (alpha = 2):
DNAm_age = 40.539696 + sum(coefficient_i × beta_i)
Training. 660 QC-pass females, chronological age 11.7–81.3 years. Tissue is saliva, typically about 70% immune cells and 30% buccal epithelium. Ancestry from a coarse 6-SNP AIM panel: 65% Eurasian, 15% East Asian, 10% African, 5% Northern European, 5% unclassified. Arrays with kit mean beta outside 0.30–0.40 were not used to fit the clock.
Internal performance: Held-out test, n = 132: r = 0.973, R2 = 0.944, MAE = 2.52 years. 10-fold cross-validation with site ranking repeated inside each fold, n = 660: r = 0.963, R2 = 0.927, MAE = 2.39 years. In-sample R2 = 0.975.
External performance on public GEO saliva (Illumina 450K; 945 of 1,000 clock sites present; missing sites filled with training-set means): GSE92767, Korean men aged 18–73, n = 54: r = 0.980, raw R2 = 0.835, raw MAE = 5.22 years. GSE99029, mixed sample aged 21–91, n = 57: r = 0.953, raw R2 = 0.785, raw MAE = 6.46 years. GSE78874, PEG cohort, mean age 69, n = 259: r = 0.878, raw R2 = 0.668, raw MAE = 4.30 years.
Raw R2 falls on 450K data because of slope compression and a dataset offset. Pearson r, and a one-time intercept-and-slope recalibration in a new lab, are the fair transfer metrics.
This record contains the 1,000 coefficients, training-column means for missing-site imputation, a Python scorer, methods, and public GEO scores. It does not contain individual-level Muhdo Health Ltd. training betas or participant identifiers.
The clock estimates chronological age and was trained on females.