Table 2.

Main characteristics of the estimators based on molecular and genetic markers.

Reference	Methodology	Accuracy Reported	Sample Types	Number of Samples (Overall)	Age Range (Years)	Comments	Associations with Age-Related Conditions and Diseases
Fleischer J. et al. [30]	Linear regression10-fold cross-validation	MAE = 7.7 years MAE (median) = 4 years	Human dermal fibroblasts	133 (people) 10 Hutchinson-Gilford progeria syndrome patients	1–94 2–9	Predicts progeria patients as 15–24 years older than age-matched controls; hence, provides an accurate estimation of biological age The only method that predicts accelerated aging in HGPS patients	No associations reported
Van den Akker E. et al. [41]	Linear regression 5-Fold-cross-validation	R2 = 0.65 MAE = 7.3 years	Blood metabolome	25,000	-	Only a biological sample required; no additional metadata needed Participants with current metabolic syndrome or diabetes mellitus type 2 were estimated older than healthy counterparts	Cardiometabolic health; increased risk of hospitalization due to heart failure, cognitive decline and cardiovascular and all-cause mortality; in nonagenarians, lower instrumental activities of daily living and increased risk of all-cause mortality during 10 years of follow-up
Ren X. et al. [32]	Elastic net	Multiple	Multiple	9662	-	Transcriptional age is significantly impacted by race The first model to perform RNA-Seq-based identification of differential gene expression for each individual tissue type	Significant correlation between the transcriptional age acceleration and mutation burden, mortality risk, and cancer stage in several types of cancer; Complementary information to DNA methylation age
Meyer D. et al. [33]	Temporal scaling and binarization	R2 = 0.92 MAE = 6.63 years MAD = 5.24	Blood	1020	-	Universal applicability, no methylation analysis required; Improved accuracy for HGPS patients compared with Fleischer’s transcriptome-based model; no DNA methylation in C. elegans, hence the effect of the epigenetic clocks in gene expression is unclear.	No associations reported
Wang J. et al. [35]	Multivariate linear regression (MLR) Regression tree (best performing) Bagging regressionRandom forest regression (RFR, best performing) Support vector regression (SVR)	MAE = 8.767 years (S.rho = 0.6983) MAE = 9.126 years (S.rho = 0.660)	Blood	100	19–73	Significantly smaller prediction MAE values for males than females (MAE = 6.133 years for males and 10.923 years for females in the regression tree model)	No associations reported
Peters M. et al. [29]	Meta-analysis	Multiple	Blood	14,983 (individuals)	-	Lower predictive accuracy compared to epigenetic clocks	Higher systolic and diastolic blood pressure, total cholesterol, HDL cholesterol, fasting glucose levels and body mass index (BMI)
Zubakov D. et al. [42]	Linear regression with sjTREC as a single predictor	R2 = 0.835, p = 8.16 × 10−215 Standard error of the estimate ± 8.9 years;	Blood	195 (individuals)	0–80	Storage time analysis showed no statistically significant difference between the sjTREC quantifications in fresh and 1.5-year-old blood samples of the same individuals	No associations reported
Galkin F. et al. [43]	Elastic Net (EN) Random Forest (RF) Gradient Boosting (XGB) Deep Neural Networks (DNNs)	MAE = 5.91 years	Stool	4000	18–90	Accuracy comparable to the existing DNAm solutions (MAE < 5 years) The microbiome composition (such as Akkermansia muciniphila, a marker of obesity, glucose metabolism, and overall intestinal health) could be used in diagnosing gut metabolism disorders; further research needed due to inconsistent results	No associations reported
Lahallier B. et al. [40]	SomaScan assay	Multiple	Plasma	2925	18–95		At peaks 2 and 3 (at the ages of 60 and 78), the proteins were associated with cardiovascular diseases, as well as Alzheimer’s disease and Down syndrome