Table 1.
Overview of predictor consistency and best performing data processing pipelines
| Predictor information | Reliability (ICC statistics) | ||||||
|---|---|---|---|---|---|---|---|
| Name | Phenotype | Array | Probes | Median | Min | Max | Best analytical pipeline |
| GrimAge [15] | Mortality | EPIC/450 K | 1030 | 0.990 | 0.921 | 0.994 | ENmix: bg = oob, dye = mean, norm = q3, probe = rcp |
| ZhangAge [8] | Chronological age | EPIC/450 K | 514 | 0.991 | 0.987 | 0.992 | ENmix: bg = neg, dye = mean, norm = q2, probe = rcp |
| TIMP_1 [15] | TIMP-1 serum protein | EPIC/450 K | 42 | 0.988 | 0.973 | 0.992 | ENmix: bg = oob, dye = relic, norm = q2, probe = rcp |
| Bcell [5] | B-lymphocyte cell fraction | EPIC | 50 | 0.980 | 0.881 | 0.988 | Minfi: no bg correction with control normalization |
| Neu [5] | Neutrophil cell fraction | EPIC | 50 | 0.984 | 0.973 | 0.987 | ENmix: bg = oob, dye = mean, norm = q2, probe = rcp |
| B2M [15] | B2M serum protein | EPIC/450 K | 91 | 0.973 | 0.759 | 0.985 | ENmix: bg = oob, dye = relic, norm = q1, probe = rcp |
| SkinBloodAge [9] | Chronological age | EPIC/450 K | 391 | 0.979 | 0.908 | 0.982 | ENmix: bg = neg, dye = relic, norm = q1, probe = rcp |
| Smoking_Lu [15] | Smoking pack years | EPIC/450 K | 172 | 0.971 | 0.889 | 0.981 | ENmix: bg = oob, dye = no, norm = no, probe = rcp |
| Smoking_McCartney [20] | Smoking pack years | EPIC | 233 | 0.975 | 0.942 | 0.979 | Minfi: noob with dye correction |
| HannumAge [7] | Chronological age | 450 K | 71 | 0.972 | 0.834 | 0.978 | ENmix: bg = est, dye = relic, norm = no, probe = rcp |
| CD8T [5] | CD8 + T-cell fraction | EPIC | 50 | 0.969 | 0.881 | 0.978 | ENmix: bg = neg, dye = mean, norm = q1, probe = rcp |
| NK [5] | Natural killer cell fraction | EPIC | 50 | 0.952 | 0.883 | 0.977 | ENmix: bg = neg, dye = relic, norm = q3, probe = rcp |
| BioAge4HAStatic [17] | Chronological age | 450 K | - | 0.966 | 0.826 | 0.975 | ENmix: bg = oob, dye = relic, norm = no, probe = rcp |
| Cystatin_C [15] | Cystatin-C serum protein | EPIC/450 K | 87 | 0.954 | 0.829 | 0.973 | ENmix: bg = oob, dye = no, norm = q2, probe = rcp |
| PhenoAge [14] | Mortality | EPIC/450 K/27 K | 513 | 0.954 | 0.926 | 0.97 | ENmix: bg = neg, dye = relic, norm = q1, probe = rcp |
| Mono [5] | Monocyte cell fraction | EPIC | 50 | 0.953 | 0.865 | 0.968 | Minfi: illumine bg correction with control normalization |
| DNAmTL [12] | Telomere length | EPIC/450 K | 140 | 0.952 | 0.912 | 0.965 | ENmix: bg = oob, dye = relic, norm = q1, probe = rcp |
| HorvathAge [6] | Chronological age | 450 K/27 K | 353 | 0.950 | 0.867 | 0.964 | WateRmelon: naten |
| CD4T [5] | CD4 + T-cell fraction | EPIC | 50 | 0.959 | 0.951 | 0.964 | ENmix: bg = neg, dye = no, norm = no, probe = rcp |
| epiTOC [18] | Mitotic divisions | 450 K | 385 | 0.911 | 0.498 | 0.962 | ENmix: bg = oob, dye = mean, norm = q2, probe = rcp |
| Leptin [15] | Leptin serum protein | EPIC/450 K | 187 | 0.896 | 0.447 | 0.953 | ENmix: bg = oob, dye = relic, norm = q3, probe = rcp |
| VidalBraloAge [13] | Chronological age | 27 K | 8 | 0.945 | 0.922 | 0.952 | ENmix: bg = neg, dye = mean, norm = no, probe = rcp |
| MiAge [19] | Mitotic divisions | 450 K | 268 | 0.884 | 0.348 | 0.947 | WateRmelon: nanes |
| LinAge [10] | Chronological age | 450 K | 99 | 0.930 | 0.878 | 0.939 | ENmix: bg = est, dye = relic, norm = no, probe = no_rcp |
| ADM [15] | ADM serum protein | EPIC/450 K | 186 | 0.900 | 0.756 | 0.938 | ENmix: bg = neg, dye = mean, norm = q3, probe = rcp |
| WHR [20] | Waist-to-hip ratio | EPIC | 226 | 0.878 | 0.634 | 0.925 | ENmix: bg = oob, dye = relic, norm = q2, probe = rcp |
| ZhangMortality [16] | Mortality | 450 K | 10 | 0.877 | 0.807 | 0.92 | Minfi: no bg correction with control normalization |
| BodyFat [20] | Body fat | EPIC | 968 | 0.893 | 0.843 | 0.918 | ENmix: bg = est, dye = relic, norm = no, probe = rcp |
| Cholesterol [20] | Total cholesterol | EPIC | 204 | 0.888 | 0.762 | 0.917 | ENmix: bg = oob, dye = no, norm = q2, probe = rcp |
| BMI [20] | BMI | EPIC | 1109 | 0.904 | 0.877 | 0.914 | ENmix: bg = neg, dye = mean, norm = no, probe = rcp |
| GDF_15 [20] | GDF-15 serum protein | EPIC/450 K | 137 | 0.819 | 0.502 | 0.903 | ENmix: bg = est, dye = mean, norm = q1, probe = rcp |
| LDL [20] | LDL | EPIC | 233 | 0.846 | 0.732 | 0.901 | ENmix: bg = oob, dye = relic, norm = no, probe = rcp |
| HDLratio [20] | Total to HDL cholesterol ratio | EPIC | 412 | 0.848 | 0.643 | 0.890 | ENmix: bg = oob, dye = relic, norm = q1, probe = rcp |
| Alcohol [20] | Alcohol | EPIC | 450 | 0.807 | 0.551 | 0.878 | ENmix: bg = neg, dye = relic, norm = no, probe = rcp |
| WeidnerAge [11] | Chronological age | 27 K | 3 | 0.826 | 0.583 | 0.865 | ENmix: bg = neg, dye = relic, norm = no, probe = rcp |
| Education [20] | Educational attainment | EPIC | 373 | 0.774 | 0.506 | 0.865 | Cross: noob with dye correction + BMIQ |
| HDL [20] | HDL cholesterol | EPIC | 737 | 0.835 | 0.694 | 0.853 | ENmix: bg = est, dye = relic, norm = q1, probe = rcp |
| CD8pCD28nCD45Ran [6] | Specific T-cell fraction | 27 K | - | 0.814 | 0.756 | 0.845 | ENmix: bg = oob, dye = relic, norm = no, probe = rcp |
| PlasmaBlast [6] | Plasma B cell fraction | 27 K | - | 0.718 | 0.638 | 0.840 | Cross: noob with dye correction + BMIQ |
| PAI_1 [15] | PAI-1 serum protein | EPIC/450 K | 211 | 0.744 | 0.22 | 0.838 | ENmix: bg = neg, dye = relic, norm = q3, probe = rcp |
| CD8naive [6] | CD8 T-cell fraction | 27 K | - | 0.777 | 0.659 | 0.830 | WateRmelon: danen |
Shown is general information on each DNAm-based predictor alongside their corresponding ICC statistics. The name of the predictor, the phenotype it is trained on, the array platform it can be applied on, and the number of predictor probes (if available) are listed on the left side of the table. ICC statistics are listed on the right side of the table. The ICC quantifies the degree of absolute agreement between estimator values of a pair of technical replicates. For each predictor, across 101 pipelines, the median, minimum, and maximum ICC are listed. Predictors are ranked by the maximum ICC. The final column reports methodological details of the best performing data processing pipelines (i.e., the pipeline with the highest consistency). Bg background correction, dye dye-bias correction, norm normalization method, probe probe-type bias correction. Full details on analytical pipelines and how they were implemented are available in Additional file 4.