Methods of determining biological age

Abstract

Disclosed herein are methods and compositions for determining the age of a subject or cell or tissue.

Description

UC Ref: BK2023-047-3 BFF Ref: BERK-518WO METHODS OF DETERMINING BIOLOGICAL AGE CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] Pursuant to 35 U.S.C. § 119(e), this application claims priority to the filing date of United States Provisional Patent Application Serial No.63 / 512,243 filed July 6, 2023, the disclosure of each of which is herein incorporated by reference in their entirety. ACKNOWLEDGEMENT OF GOVERNMENT SUPPORT

[0002] This invention was made with government support under Grant Numbers EB023776, GM098218, and HL139605 awarded by the National Institutes of Health. The Government has certain rights in the invention. BACKGROUND OF THE INVENTION FIELD OF THE INVENTION

[0003] The field generally relates to methods and kits for determining the age of a subject or a biological sample. DESCRIPTION OF THE RELATED ART

[0004] Elastic Net (EN) DNA methylation (DNAme) clocks have low accuracy of predictions for individuals of the same age and a low resolution between healthy and disease cohorts; caveats inherent in applying linear model to non-linear processes. We found that change in methylation of CpGs with age is, interestingly, not the determinant for their selection into the clocks. Moreover, an EN clock’s selected CpGs change when non-clock CpGs are removed from the training data; as expected from optimization in a machine learning (ML) context, but inconsistently with the identification of health markers in a biological context. To address these limitations, we moved from predictions to measurement of biological age, focusing on the CpGs that on average remain invariable in their methylation through lifespan, postulated to be homeostatically vital. We established that dysregulation of such CpGs, measured as the sums of standard deviations of their methylation values, quantifies biological noise, which in our hypothesis is a biomarker of aging and disease. We term this approach a “noise barometer” - the pressure of aging and disease on an organism. These noise-detecting CpGs are particularly important as sums of SD on the entire 450K DNAme array data yield a random pattern through chronology. Testing how many CpGs of the 450K arrays become noisier with age, we found that the paradigm of DNAme noise as a biomarker of aging and disease remarkably manifests in ~1 / 4 of the total. In that large set even the CpGs that have on average constant methylation through age show increased SDs and can be used as noise detectors of the barometer.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO SUMMARY OF THE INVENTION

[0005] The invention provides methods for determining the age of a subject or a biological sample obtained from the subject based on the epigenome of the subject.

[0006] In one aspect, the method comprises (a) obtaining a biological sample of the subject; (b) detecting the methylation levels of a set of age-associated epigenetic markers described herein from the biological sample of the subject; and (c) comparing the methylation levels of the set of age-associated epigenetic markers described herein with the methylation levels of the same set of age-associated epigenetic markers described herein in a control. In an exemplary embodiment, step (b) comprises (b1) selecting a set of CpGs described herein, and (b2) measuring the methylation levels of each of the selected CpGs in a biological sample obtained from the subject.

[0007] In some embodiments, the present invention is directed to methods for determining the age of a subject or a biological sample obtained from the subject, which comprise selecting a set of CpGs comprising or consisting of 15 or more, 16 or more, 17 or more, 18 or more, 19 or more, 20 or more, 21 or more, 22 or more, 23 or more, 24 or more, 25 or more, 26 or more, 27 or more, 28 or more, 29 or more, 30 or more, 31 or more, 32 or more, 33 or more, 34 or more, 35 or more, 36 or more, 37 or more, or all 38 of the following CpGs: cg01767885, cg01869186, cg01900413, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg08499046, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg23417875, cg23983449, cg24027320, cg24112454, cg27621745, cg27642618, cg00479463, cg01681847, cg06899313, cg07710266, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, and cg27219748; measuring the methylation levels of each of the selected CpGs in a biological sample obtained from the subject; and comparing the methylation levels to a control. In some embodiments, the set of CpGs comprises or consists of 15 or more, 16 or more, 17 or more, 18 or more, 19 or more, 20 or more, 21 or more, 22 or more, 23 or more, 24 or more, 25 or more, 26 or more, 27 or more, 28 or more, 29 or more, 30 or more, 31 or more, 32 or more, 33 or more, 34 or more, 35 or more, 36 or more, 37 or more, 38 or more, 39 or more, 40 or more, 41 or more, 42 or more, 43 or more, 44 or more, 45 or more, 46 or more, 47 or more, 48 or more, 49 or more, or all 50 of the following CpGs: cg00868523, cg01767885, cg01869186, cg01900413, cg02449103, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439. In some embodiments, the set of CpGs additionally includes one or more CpGs selected from the group consisting of cg01974375, cg06784991, cg08128734, cg09809672, cg10501210, cg16054275, cg06639320, cg22454769, cg24079702, cg07553761, cg08262002, cg06493994, cg16867657, cg21572722, cg22736354, cg24724428, cg03473532, cg08097417, cg14361627, cg16419235, cg26290632, cg01812894, cg07080372, cg08160331, cg01820374, cg11649376, cg17436656, cg04875128, cg08090640, cg19283806, and cg07547549. In some embodiments, the age that is determined is the predicted chronological age of the subject or the biological sample. In some embodiments, the age that is determined is the predicted biological age of the subject or the biological sample. In some embodiments, the methods exclude machine learning steps, e.g., excludes the need to use of a training set of data to correlate methylation levels or values to chronological or biological ages. In some embodiments, the methods further comprise calculating the absolute deviation from the mean beta value of the measured methylation levels for each selected CpG to obtain a plurality of absolute deviations, summing the plurality of absolute deviations to obtain a sum of deviation, and comparing the sum of deviation to at least one standard to determine the age of the subject or the age of the biological sample. In some embodiments, the sum of deviation is a normalized sum of deviation. In some embodiments, the normalized sum of deviation is obtained by dividing the sum of deviation by an average sum of deviation obtained from a control group. In some embodiments, the control group is a group of healthy subjects having a chronological age of 25–27 years (i.e., at least 25 years and less than 28 years). In some embodiments, the at least one standard is at least one sum of deviation obtained from a control subject of a given age. In some embodiments, the at least one standard is at least one average of a plurality of sums of deviation obtained from a control population of a given age. In some embodiments, the at least one standard is at least one median of a plurality of sums of deviation obtained from a control population of a given age. In some embodiments, the at least one standard is normalized for the given age. In some embodiments, the at least one standard is a polynomial equation that correlates the at least one standard to at least one chronological age.

[0008] In some embodiments, the present invention is directed to methods for determining the age of a subject or a biological sample obtained from the subject, which comprise selecting a set of CpGs comprising or consisting of 15 or more, 16 or more, 17 or more, 18 or more, 19 or more, 20 or more, 21 or more, 22 or more, 23 or more, 24 or more, 25 or more, 26 or more, 27UC Ref: BK2023-047-3 BFF Ref: BERK-518WO or more, 28 or more, 29 or more, 30 or more, 31 or more, 32 or more, 33 or more, 34 or more, 35 or more, 36 or more, 37 or more, or all 38 of the following CpGs: cg01767885, cg01869186, cg01900413, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg08499046, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg23417875, cg23983449, cg24027320, cg24112454, cg27621745, cg27642618, cg00479463, cg01681847, cg06899313, cg07710266, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, and cg27219748; measuring the methylation levels of each of the selected CpGs in the biological sample; calculating the absolute deviation from the mean beta value of the measured methylation levels for each of the selected CpGs to obtain a plurality of absolute deviations; summing the plurality of absolute deviations to obtain a sum of deviation; normalizing the sum of deviation; and determining the age of the subject or the biological sample by correlating the normalized sum of deviation to a standard. In some embodiments, the set of CpGs comprises or consists of 15 or more, 16 or more, 17 or more, 18 or more, 19 or more, 20 or more, 21 or more, 22 or more, 23 or more, 24 or more, 25 or more, 26 or more, 27 or more, 28 or more, 29 or more, 30 or more, 31 or more, 32 or more, 33 or more, 34 or more, 35 or more, 36 or more, 37 or more, 38 or more, 39 or more, 40 or more, 41 or more, 42 or more, 43 or more, 44 or more, 45 or more, 46 or more, 47 or more, 48 or more, 49 or more, or all 50 of the following CpGs: cg00868523, cg01767885, cg01869186, cg01900413, cg02449103, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439. In some embodiments, the set of CpGs additionally includes one or more CpGs selected from the group consisting of cg01974375, cg06784991, cg08128734, cg09809672, cg10501210, cg16054275, cg06639320, cg22454769, cg24079702, cg07553761, cg08262002, cg06493994, cg16867657, cg21572722, cg22736354, cg24724428, cg03473532, cg08097417, cg14361627, cg16419235, cg26290632, cg01812894, cg07080372, cg08160331, cg01820374, cg11649376, cg17436656, cg04875128, cg08090640, cg19283806, and cg07547549. In some embodiments, the methods exclude machine learning steps. In some embodiments, the standard is (a) a best-fit polynomial curve of medians of normalized sums obtained from healthy control subjects of chronological ages ranging from 24 – 80 years fromUC Ref: BK2023-047-3 BFF Ref: BERK-518WO birth, (b) a polynomial equation representing the best-fit polynomial curve, or (c) a standard sum of deviation for a given biological age. In some embodiments, the polynomial equation is a 5th degree polynomial equation as follows: y = (5.240675828340669 * x) + (-0.2188129249734282 * x^2) + (0.004474781331195266 * x^3) + (-4.483740648065997e-05 * x^4) + (1.7658492552100836e-07 * x^5) + (-48.49899824287247), or an 8th degree polynomial equation as follows: y = (318.0470861000581 * x) + (-25.24397997218023 * x^2) + (1.1274410950264393 * x^3) + (-0.03097071704329002 * x^4) + (0.0005356837380989208 * x^5) + (-5.697293442369076e-06 * x^6) + (3.407386789366876e-08 * x^7) + (-8.777464346524329e-11 * x^8) + (-1727.276932311053); wherein x is age in years and y is the normalized sum of deviation from mean beta value of the given CpG of a given subject, and the root(X) is the biological age of the subject or the biological sample. In some embodiments, the standard sum of deviation and given biological age is selected from one of the following tables: Standard Sums of Deviation Given Biological Ages 040 280Standard Sums of Deviation Given Biological AgesUC Ref: BK2023-047-3 BFF Ref: BERK-518WO 0.60 30.7 0.65 32.6 .

[0009] In somds for determining the health of a subject relative to the subject’s chronological age, which comprise selecting a set of CpGs comprising or consisting of 15 or more, 16 or more, 17 or more, 18 or more, 19 or more, 20 or more, 21 or more, 22 or more, 23 or more, 24 or more, 25 or more, 26 or more, 27 or more, 28 or more, 29 or more, 30 or more, 31 or more, 32 or more, 33 or more, 34 or more, 35 or more, 36 or more, 37 or more, or all 38 of the following CpGs: cg01767885, cg01869186, cg01900413, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg08499046, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg23417875, cg23983449, cg24027320, cg24112454, cg27621745, cg27642618, cg00479463, cg01681847, cg06899313, cg07710266, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, and cg27219748; measuring the methylation levels of each of the selected CpGs in a biological sample obtained from the subject; calculating the absolute deviation from the mean beta value of the measured methylation levels for each of the selected CpGs to obtain a plurality of absolute deviations; summing the plurality of absolute deviations to obtain a sum of deviation; normalizing the sum of deviation; determining the biological age of the subject by correlating the normalized sum of deviation to a standard; and determining the subject’s as being (1) poor where the determined biological age is significantly or statistically greater than the subject’s chronological age, (2) average where the determined biological age and chronological age of the subject are substantially similar, or (3) superior where the determined biological age is significantly or statistically less than the subject’s chronological age. In some embodiments, the set of CpGs comprises or consists of 15 or more, 16 or more, 17 or more, 18 or more, 19 orUC Ref: BK2023-047-3 BFF Ref: BERK-518WO more, 20 or more, 21 or more, 22 or more, 23 or more, 24 or more, 25 or more, 26 or more, 27 or more, 28 or more, 29 or more, 30 or more, 31 or more, 32 or more, 33 or more, 34 or more, 35 or more, 36 or more, 37 or more, 38 or more, 39 or more, 40 or more, 41 or more, 42 or more, 43 or more, 44 or more, 45 or more, 46 or more, 47 or more, 48 or more, 49 or more, or all 50 of the following CpGs: cg00868523, cg01767885, cg01869186, cg01900413, cg02449103, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439. In some embodiments, the set of CpGs additionally includes one or more CpGs selected from the group consisting of cg01974375, cg06784991, cg08128734, cg09809672, cg10501210, cg16054275, cg06639320, cg22454769, cg24079702, cg07553761, cg08262002, cg06493994, cg16867657, cg21572722, cg22736354, cg24724428, cg03473532, cg08097417, cg14361627, cg16419235, cg26290632, cg01812894, cg07080372, cg08160331, cg01820374, cg11649376, cg17436656, cg04875128, cg08090640, cg19283806, and cg07547549. In some embodiments, the methods exclude machine learning steps. In some embodiments, the standard is (i) a best-fit polynomial curve of medians of normalized sums obtained from healthy control subjects of chronological ages ranging from 24 – 80 years from birth, (ii) a polynomial equation representing the best-fit polynomial curve, or (iii) a standard sum of deviation for a given biological age. In some embodiments, the polynomial equation is a 5th degree polynomial equation as follows: y = (5.240675828340669 * x) + (-0.2188129249734282 * x^2) + (0.004474781331195266 * x^3) + (-4.483740648065997e-05 * x^4) + (1.7658492552100836e-07 * x^5) + (-48.49899824287247), or an 8th degree polynomial equation as follows: y = (318.0470861000581 * x) + (-25.24397997218023 * x^2) + (1.1274410950264393 * x^3) + (-0.03097071704329002 * x^4) + (0.0005356837380989208 * x^5) + (-5.697293442369076e-06 * x^6) + (3.407386789366876e-08 * x^7) + (-8.777464346524329e-11 * x^8) + (-1727.276932311053); wherein x is age in years and y is the normalized sum of deviation from mean beta value of the given CpG of a given subject, and the root(X) is the biological age of the subject or the biological sample. In some embodiments, the standard sum of deviation and given biological age is selected from one of the following tables:UC Ref: BK2023-047-3 BFF Ref: BERK-518WO Standard Sums of Deviation Given Biological Ages 0.40 28.0Standard Sums of Deviation Given Biological Ages .

[0010] Both the foregoing general description and the following detailed description are exemplary and explanatory only and are intended to provide further explanation of the inventionUC Ref: BK2023-047-3 BFF Ref: BERK-518WO as claimed. The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute part of this specification, illustrate several embodiments of the invention, and together with the description explain the principles of the invention. DESCRIPTION OF THE DRAWINGS

[0011] This invention is further understood by reference to the drawings wherein:

[0012] FIG 1 presents tests of EN models on new healthy and disease datasets. Standard EN modes were trained on 25K - small (Panel A) and 450K - medium (Panel B) DNAme arrays datasets; tests with the original and new datasets, Arthritis, Down Syndrome, etc. as indicated, are shown. Blue – healthy, Red – disease, unless indicated by additional colors. Both models are testable with not only the original, but all new datasets, and report age-acceleration v. deceleration for the same sample set and throughout chronology, depending on the training dataset. Both models give overlapping age predictions for the healthy individuals and patients. Panel C: The EN model, which was trained on the small dataset, was tested on the datasets from BRCA-1 studies. The predictions of chronological age fail for the healthy control in tests but are successful for the less healthy cohorts. Panel D: Tests of small and medium models on the datasets of the Werner Syndrome studies. The medium model gives more accurate predictions of chronological age than the small EN model, and both overlap the ages of healthy subjects and people with Werner syndrome. Panel E: Test of small model on medium dataset yields age- acceleration for most of the samples.

[0013] FIG 2 shows distinct clustering of DNAme of healthy vs. disease cohorts. UMAP on the 450K CpG dataset for Arthritis, Down Syndrome, and BRCA1 (GSE42861, GSE52588, GSE57285) disease studies, which are performed with the random state number of 5 for all and the n neighbor of 26, 15, and 6 correspondingly. Each point represents an individual. The clustering of healthy, cancer, and BRCA1 mutation yet no cancer cohorts is distinct for each dataset.

[0014] FIG 3 shows assaying the CpG selection and ranking by EN and the new hypothesis- driven OLS Age Predictor. Panel A: DNAme clock was constructed with EN regression on a 450K array dataset (GSE40279, N=656), and the test set prediction performance is shown (Model 0); The selected clock CpGs were removed, the model was retrained, and model performance on the test set is shown (Model 1). Panel B: Scatter plot of the coefficient of variation for each clock CpG individually regressed on age plotted against their rank by their absolute weights. Model 0, top. Model 1, bottom. There is no preference for biologicallyUC Ref: BK2023-047-3 BFF Ref: BERK-518WO relevant CpGs, e.g., those changing in their methylation with age. The mean coefficient of variation of the clock CpGs are shown as red dashed lines. The CpGs of Model 1, e.g., those which were minimized to zero in Model 0, have a higher mean R2, despite being slightly less accurate. Panel C: Simple schematic of modified forward step-wise selection algorithm. Panel D: CpGs were ranked by the absolute value of the Pearson correlation of their methylation state with age. Line plots show Pearson’s r (top) and median absolute error (MAE) (bottom) of the test set of the OLS regression on the step-wise addition of the CpGs, starting with the most linearly changing in DNAme with age, Panel E. The tests of the hypothesis-driven OLS model on four out-of-sample data sets (GSE106648: multiple sclerosis, GSE52588: Down syndrome, GSE87640: IBS, GSE42861: rheumatoid arthritis) which include a disease case are shown, as indicated. Blue = healthy, red = disease.

[0015] FIG 4 shows removal of non-clock CpGs from a dataset and retraining an EN model leads to a different set of clock CpGs. First, an EN clock was trained with a 450K dataset (GSE40279), then non-clock CpGs were randomly removed independently and in a stepwise fashion. After each iteration of removal a new EN model was trained and the selected set of CpGs was compared to the set of clock CpGs when the whole dataset was used. Panel A: Percent of non-clock CpGs removed vs percent of original clock CpGs left out of the new model. Non-clock CpGs were removed in one percent increments from 0−10%, and 10% increments from 10−100%. At 2% of removal of the non-clock CpGs, the CpG set selected by EN began significantly changing. The percentage of unused original CpGs plateau at 17% in both gradual (1% at a time) and rapid (10% at a time) removal of the non-clock CpGs. Panel B: Percent original DeepmAge clock CpGs unused in the model with between 1−10% of non-clock CpGs removed and with 10−100% of non-clock CpGs removed. The set of Clock CpGs selected by deep learning are not impervious to removal when non-clock CpGs are removed. Fifteen of the 1000 CpGs selected by DeepmAge were included in the models’ original 847 Clock CpGs. The percentage of unused original DeepmAge CpGs plateaus at around 35% in both gradual and rapid removals. Panel C: Tests of the EN model after training on the original 450K dataset and after training on a dataset with 30% of non-clock CpGs removed. Panel D: OLS regression on the EN selected CpGs from the original 450K dataset and from the dataset with 30% of non-clock CpGs removed. The original and new EN models were roughly equivalent in predicting chronological age, though the original model was slightly favored. A simple OLS using the clock CpGs of these models was created, and see that the models are again roughly equivalent, but that the new model slightly outperforms the old model, and that both are comparable to the EN models. Panel E: A demonstration of the effect of removing theUC Ref: BK2023-047-3 BFF Ref: BERK-518WO non-significant CpGs has on the Elastic-Net energy landscape. Here, the coefficients assigned to 10 middle value clock CpGs are seen as a function of the alpha hyperparameter. There are significant differences between the coefficients in both the original EN model and a model with 30% of non-significant CpGs removed across alpha. In particular, there are significant differences between the coefficients the models assign both at the alphas the models eventually settle to and between these alphas.

[0016] FIG 5 shows physiologic polynomial curves of relative biological age. Panel A: Representative CpG that is on average invariable in its methylation throughout lifespan but becomes visibly noisier, showing higher absolute deviation from the mean in older individuals. Panel B: Normalized Sums of the absolute deviations from the mean of 50 CpGs with 0.01<r<0.05 for their beta values regressed on age and r > 0.2 for their absolute difference from the mean when regressed on age; all data was normalized by the Sums of 25−27 year of age samples; each dot is an individual sample. 5n (Panel C) and 8n (Panel D) polynomial functions of the Median of the Normalized Sum of Standard Deviations v. age of the 50 CpGs are shown in blue for healthy control of Arthritis dataset with overlays of the data from Arthritis patients (red dots). Each dot is the Median for each age. Panel E: 5n polynomial correlation curves of IBS on its own healthy control and HIV on the average of healthy controls of Arthritis and IBS; HIV dataset does not have a parallel healthy control. Panel F: Median biological age curves are shown for IBS, HIV and Arthritis as per the average of the healthy controls of Arthritis and IBS. Panel G. The 50 CpG based direct measurement of biological age was conducted as above, using the 450K DNAme array datasets on healthy controls (blue line) from the studies of IBS, Arthritis, Parkinson’s and Multiple Sclerosis (Combined Health) v. the patients’ datasets from the studies on HIV, IBS, Arthritis and Multiple Sclerosis (Combined Disease). Combined disease has higher relative biological age as compared to Combined Health. Panel H. Venn diagrams demonstrate sharing of the CpGs, which correlate with chronological age in their methylation and independence of the CpGs that serve as detectors of epigenetic noise and measure biological age.

[0017] FIG 6 shows the age-region linearity of the EN models. A standard DNAme clock was trained on the beta values and sample ages from GSE40279. The data were split 80 / 20 into training and test sets respectively. The predictions and their correlations (Pearson’s r) on the test data over the full chronological range and on specific age ranges (italicized) are shown. Panel A: The sample numbers of all age regions were balanced, the DNAme clock was trained and tested (Panel B). The most linear age regions of 18−30 years of age and 80−90 years of age areUC Ref: BK2023-047-3 BFF Ref: BERK-518WO shown for the standard (Panel C) and the age-even (Panel D) models. Interestingly, in the age- even EN model for the very old age region, all samples are under-predicted. See FIG 1, Panel D. The Pearson’s r of 0.64 of the 76−100 year region is potentially lower if not for the 80−95 year of age transition (outlined in red in FIG 1, Panel A). These age-ranges were selected because they might represent biological plateaus.

[0018] FIG 7 shows tSNE analyses for clustering of DNAme array values with chronological age of the blood donors. t-SNE on the PCA-initialized (300 components) GSE40279, which is performed with the perplexity value of 48. Each point represents an individual, and actual chronological age values of each individual are color-coded (scale on the right). There does not appear to be any significant relationship between t-SNE clustering and chronological age.

[0019] FIG 8 shows testability of Small and Medium EN Models on additional disease / health datasets and FSW OLS clock predictions on Arthritis, Down Syndrome and HIV. Panel A: Tests of the small model (GSE41037, GSE20067, GSE20236, GSE19711) and medium model (GSE40279) on out-of-sample datasets (GSE106648: MS, GSE32148: IBS) with disease cases, as indicated. The ages for healthy donors (blue) and patients (red) overlap. Panel B: The FSW OLS model (trained on GSE40279) yields good performance on the out-of-sample test data; yet it continues to overlap health and disease.

[0020] FIG 9 shows Hannum 71 Clock CpGs are not resilient to the removal of non-Clock CpGs. Standard EN clock was trained on the Hannum dataset (GSE40279), after which non- clock CpGs were independently stepwise removed and the alterations to the putatively relevant features of Hannum 71 Clock were analyzed. Percent clock CpGs removed is shown at the removal of 1−10% and 10−100% of non-clock CpGs. At 4% of removal of the non-clock CpGs the clock CpGs begin changing and the percentage of unused original CpGs plateau at 17% in a gradual (1% at a time) and rapid (10% at a time) removal of the non-clock CpGs.

[0021] FIG 10 shows Standard Deviations – based measurement of relative biological age. Standard Deviations from Mean of beta values were calculated for all 450K array CpGs with (r) < 0.05 (e.g., the 20898 CpGs) and plotted in relation to the chronological age, in five-year bins. Age-imposed increase in the methylation noise of these typically unchanging CpGs is evident.

[0022] FIG 11 shows Standard Deviations – based measurement of relative biological age. Standard Deviations from Mean of beta values were calculated for all 450K array CpGs with (r) < 0.05 (e.g., the 20898 CpGs) and plotted in relation to the chronological age, in five-year bins. Age-imposed increase in the methylation noise of these typically unchanging CpGs is evident.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0023] FIG 12 shows basic principles of the 1st and next gen DNAme EN clocks. (A) A schematic of the fundamental processing of DNAme arrays data into relative beta values and using these as input for the 1st gen and next gen EN DNAme clocks. A large data set with some underlying pattern suffices for forming the least erroneous linear correlation with a measured parameter. (B, C) Comparison between the epigenetic analysis of gene expression in typical biological studies (B) and the CpGs picked by EN DNAme clocks (C). (D, E) comparison between typical biomedical studies that yield significant differences between cohorts in select parameters (D) and EN DNAme clock predictions (E). Of note, high throughput sequencing is broadly available, yet clock models are typically built from less-accurate DNA hybridization data, and parallel bisulfite sequencing controls are lacking for the putative changes in DNAme of the individuals who are predicted by the clocks to be biologically older or younger.

[0024] FIG 13 shows the age-region linearity of the EN models. (A) A standard DNAme clock was trained on the beta values and sample ages from GSE40279. The data were split 80 / 20 into training and test sets respectively. The predictions and their correlations (Pearson’s r) on the test data over the full chronological range and on specific age ranges (italicized) are shown. (B) The sample numbers of all age regions were balanced, the DNAme clock was trained and tested. The most linear age regions of 18-30 year of age and 80-90 years of age are shown for the standard (C) and the age-even (D) models. Interestingly, in the age-even EN model for the very old age region, all samples are under-predicted, FIG 12D. The Pearson’s r of 0.64 of the 76-100 yr. range is potentially lower if not for the 80-95ya transition (outlined in red in A). These age- ranges were selected to test for the presence of DNAme plateaus, which might be less visible at the low magnification of the total age range.

[0025] FIG 14 shows fail-tests of EN models that are trained on biological v. non-biological parameters. (A) Standard first generation EN model was trained on 450K DNAme arrays dataset; tests with the original and new datasets, as indicated, are shown (blue – healthy, red – disease). BRCA-1 studies were done on 27K dataset, thus, after training on 27K dataset. EN was tested on BRCA-1 data. EN models were testable with the original and new datasets, except for the BRCA-1, where only the subjects with BRCA1 mutation and cancer but not healthy controls received fairly accurate age predictions. All EN models overlapped predictions for healthy subjects and patients (see FIG 15 for the tests of EN models on additional disease v. healthy datasets). (B) PhenoAge scatter plots overlay for the patients with the studied datasets (red dots- Combined Disease, blue dots-Combined Healthy). (C) PhenoAge tests on the six 450K datasets are represented as the bar graphs on Mean and SDs of predicted age minus actual age (left) andUC Ref: BK2023-047-3 BFF Ref: BERK-518WO as the comparison in absolute residuals (right); patients – red bars n=806, healthy controls – blue bars; n=754, p-values are shown, NS- non-significant. (D) PhenoAge predictions of DNAme age for the indicated age-intervals are shown as bar graphs of Means with SDs for the patients with Down Syndrome and Parkinson’s Disease (red bars), and their healthy controls (blue bars). DS, n=26, Control, n= 58; PD, n=289, Control, n=219; p-values are shown. (E) EN model was trained on the US population numbers, at subjects’ birth years, using the same 450K DNAme array, as in A. The test shows excellent linearity and low MAE in predicting the numbers of people living in US from DNAme array, through 151 clock CpGs. (F) The US population EN model was then trained to predict persons’ age with successful tests (near perfect linearity and MAE of 2.6 years) by 149 clock CpGs. (G) This US population clock was also testable as an age predictor on the out-of-sample 450K datasets of indicated diseases (red dots) and their healthy controls (blue dots); health and disease predictions overlapped. (H) The age predictions made by the US population clock are shown as scatter plots overlay for Combined Disease (arthritis, Down Syndrome, HIV and Chron’s – red dots) and their healthy controls (Combined Healthy, blue dots). All GSEs are in Methods.

[0026] FIG 15 shows testability of 27K and 450K EN models on additional disease / health datasets, and a demonstration that simple OLS on clocks CpGs is accurate on out-of-sample datasets. (A, B) Tests of the 27K models on out-of-sample datasets with disease cases, as indicated. (C) Tests of the same 450K model as in FIG 12 on out-of-sample datasets with disease cases, as indicated. The ages for healthy subjects (blue) and patients (red) overlap. (D) Test of 27K model on 450K dataset yields age-acceleration for most of the samples. (E) Simple OLS on clock CpGs is functional for age predictions and overlaps health (blue dots) and disease (red dots). GSEs are provided in Methods.

[0027] FIG 16 shows robust epigenetic differences between health and disease are clear in the 450K datasets. Distinct clustering of DNAme of healthy vs. disease cohorts. UMAP on the 450Kcytosinedataset for Arthritis, Down Syndrome, Parkinson’s Disease and BRCA1 studies (see Methods for dataset identifiers). Each point represents an individual. The clustering of healthy, cancer, and BRCA1 mutation yet no cancer cohorts is distinct for each dataset.

[0028] FIG 17 shows next gen clocks are not resilient to the removal of non-clock CpGs, and cytosine selection likely reflects changes in the data landscape. (A) Standard EN clock was trained on the Hannum dataset (GSE40279), after which non-clock CpGs were independently stepwise removed and the alterations to the putatively relevant features of Hannum 71, DeepmAge and PhenoAge clocks were analyzed. Percent clock CpGs removed is shown at theUC Ref: BK2023-047-3 BFF Ref: BERK-518WO removal of 1-10% and 10-100% of non-clock CpGs. (B) A demonstration of the effect of removing the non-significant CpGs has on the Elastic-Net energy landscape. Coefficients assigned to 10 middle value clock CpGs as a function of the alpha hyperparameter significantly differ in both the original EN model and a model with 30% of non-significant CpGs removed across alpha. There are significant differences between the coefficients the models assign, and the models eventually settle to and between these alphas.

[0029] FIG 18 shows cytosine ranking by EN is not based on the changes in methylation with age. (A) DNAme clock was constructed with EN regression on a 450K array dataset (GSE40279, N=656), and the test set prediction performance is shown (Model 0). The selected clock CpGs were removed, the model was retrained, and model performance on the test set is shown (Model 1). Scatter plot of the coefficient of variation for each clock cytosine individually regressed on age plotted against their rank by their absolute weights. The mean coefficient of variation of the clock CpGs are shown as red dashed lines. The CpGs of Model 1, e. g., those which were minimized to zero in Model 0, have a higher mean R2, despite being slightly less accurate. (B) Scatter plots of the coefficient of variation for the CpGs regressed on age are plotted against their rank / EN weights of the published 1st generation Hannum clock, PhenoAge and PACE. (C) An EN clock was trained with a 450K dataset (GSE40279), then non-clock CpGs were randomly removed: independently in a stepwise fashion. After each iteration of removal, a new EN model was trained and the selected set of CpGs was compared to the set of original clock CpGs. Non-clock CpGs were removed in one percent increments from 0-10% (left panel) and 10% increments from 10-100% (right panel). At 2% of removal of the non-clock CpGs, the cytosine set selected by EN began significantly changing. The percentage of unused original CpGs plateau at 17% in both gradual (1% at a time) and rapid (10% at a time) removal of the non-clock CpGs.

[0030] FIG 19 shows raw data on the 50-cytosine noise barometer and on Low-mid, high methylated and unmethylated CpGs from the 17-cytosine group. (A) The normalized by young age, summed absolute distance from Mean of beta values v. chronological ages are shown as the scatter plots, each dot is an individual. Blue – healthy controls, Red – patients with arthritis. Polynomial fit of the Median is in FIG 18B. (B) Shown are the beta values over age scatter plots of the additional FIG 18 CpGs that have near same Mean of beta values in young and old cohorts and are the least regulated in the young cohort (SD>0.3 of Mean). Each dot is a sample, each cytosine is color coded.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0031] FIG 20 shows direct quantification of biological age from DNAme array data through noise barometer. (A) Representative cytosine that is on average invariable in its methylation throughout lifespan but becomes visibly noisier, showing higher absolute deviation from the mean in older individuals. Each dot is an individual. (B) Polynomial curve was fitted to the dot- plot of Median of normalized by healthy young sums of SDs of the 50 CpGs for healthy individuals, left; the same polynomial curve is overlayed with the dot plot of the Median of normalized by healthy young sums of SDs of the 50 CpGs for arthritis patients, right. Biological ages were mapped onto the right Y axis, as described in Methods. Each dot is age range: blue – healthy, red, arthritis. (C) Polynomial curve was fitted to the 460 Summed SDs (of the 460 CpGs) v. chronological age ranges, using the six combined 450K DNAme datasets of healthy controls. Right Y axis shows mapping of the summed SDs into biological ages and the compression of specific age ranges. (D) Scatter plots on the changes in beta values over age for representative CpGs that are least regulated in young, with SD=0.3 of absolute Mean of beta values, and all 8 most regulated in young CpGs with SD=0.015 of absolute Mean of beta values. Each dot is a sample. (E) Dot plots of the summed SDs of the 17 CpGs and of the 8 CpGs. Each dot is 49 samples on average. Black arrows indicate transitions from low to high noise. (F) Same as (D), but for CpGs that have SD=0.3 and 0.015 of the relative Mean of beta values. (G) Same as (E), but for the summed CpGs with SD=0.3 and 0.015 of the relative Mean of beta values. (H) Venn diagrams of the CpGs of the published clocks and the 460, 50 and 5, 8 – most regulated, 13, 17 least regulated CpGs of the noise barometer that quantifies biological age; the text is color-coded per each cytosine set. In this Figure, ‘cytosine’ or ‘Cyt’ refers to a cytosine in a CpG.

[0032] FIG 21 shows biological noise curves for individual diseases, individual CpGs, and the combined health v. disease progression of biological aging with HIV dataset included. (A) The 50-cytosine based measurement of biological noise is shown for the indicated diseases. The shapes of the cures are relevant in this panel. (B) Individual highly regulated in young CpGs have different curves of biological aging. (C) The shown Health (blue) versus Disease (red) polynomial fit curves of biological aging are based on the 460 CpGs and include HIV dataset, which does not have its own healthy control. In (B, C), each dot is an age-interval.

[0033] FIG 22 shows fundamental paradigm of noise barometer; DNAme clocks do not rely on DNAme noise. (A) Shown are polynomial fit curves of the chronological age v. sums of 460-, 293-, 5 SDs, SD and the dot plot line-fit of a single cytosine from the most regulated in young group, and the sums of 143, 448 SDs of all heteroscedastic CpGs with 20% or larger SD in oldUC Ref: BK2023-047-3 BFF Ref: BERK-518WO than young; the unfiltered 450K summed SDs are the black dot scatter that is overlaid with the 143, 448 (143K) line graph. These data are on the six combined datasets of patients with various diseases (red lines) and their healthy controls (blue lines). All noise detectors, but not the unfiltered 450K DNAme CpGs, outline a similar progression of biological aging of the healthy subjects, clearly distinguishing it from the disease-influenced changes. Each dot is the age range. (B) Dashed blue line shows the Median of summed SDs of three DNAme array healthy controls datasets that have the most samples (Arthritis, MS, and PD); this Median line is overlayed with the dot plots of each of these individual datasets, color-coded. Each dot is the age range. (C) Venn Diagrams show the presence of the 143K CpGs in the 50 and 460 noise barometers and in the DNAme EN clocks, color-coding is the same, as in FIG 20. (D) Scatter plots of the coefficient of variation of the CpGs’ SD regressed on age are plotted against their rank / EN weights for the published 1st generation clock, PhenoAge and PACE, [Horvath (2013); Hannum (2013); Belsky (2022); Levine (2018); Lu (2019)]. In this Figure, ‘Cytosine’ refers to a cytosine in a CpG.

[0034] FIG 23 shows a set of age-associated epigenetic markers with 5 CpGs.

[0035] FIG 24 shows a set of age-associated epigenetic markers with 8 CpGs.

[0036] FIG 25 shows a set of age-associated epigenetic markers with 13 CpGs.

[0037] FIG 26 shows a set of age-associated epigenetic markers with 17 CpGs.

[0038] FIG 27 shows a set of age-associated epigenetic markers with 50 CpGs.

[0039] FIG 28 shows a set of age-associated epigenetic markers with 460 CpGs.

[0040] FIG 29 shows overlapping CpGs for 143k CpGs passed 2 filters, PhenoAge CpGs, PACE CpGs, hannum CpGs (published), 460 CpGs, and 50 CpGs.

[0041] FIG 30 shows overlapping CpGs.

[0042] FIG 31 shows overlapping CpGs between Hannum and PhenoAge, and between Hannum, PhenoAge, and 143k CpGs passed 2 filters.

[0043] FIG 32 shows batch effects. DETAILED DESCRIPTION OF THE INVENTION

[0044] Disclosed herein is an accurate DNAme clock that is based on stepwise inclusion into an ordinary least squares (OLS) regression model of the CpGs with a high correlation with chronological age. In certain instances, this accurate DNAme clock does not require machineUC Ref: BK2023-047-3 BFF Ref: BERK-518WO language. It outperforms EN in accuracy when predicting on out-of-sample datasets, but still does not discriminate between health and disease. Also disclosed herein is a “noise barometer”, which is a method for detecting DNAme noise and is based on 50 tightly regulated CpGs that exhibit age-invariant methylation. Use of the noise barometer allows accurate measurements of biological age, exhibits the natural curve of epigenetic aging, distinguishes health and disease in subjects of the same chronological age, and distinguishes one disease from another.

[0045] As used in this application, the biological age is determined using a set of CpGs that are correlated with DNA methylation noise of a subject or an organism. In the current invention, the biological age is determined from an analysis of the modification levels of specific CpGs and, in particular, e.g., the methylation status at the C-5 position of cytosine. I. Discussion of previous epigenetic clocks

[0046] Elastic net regression has been the primary machine learning method of choice for constructing chronological DNAme clocks, with a more recent addition of deep learning methods. The fundamentals and assumptions of EN are summarized herein, but the main principle of EN is to use regularization to perform feature selection and absolute shrinkage of feature weights, allowing for the fit of a linear model to a potentially non-linear relationship between the response variable and the predictors. The EN model is trained using stochastic gradient descent to minimize a cost function (Equation 1) and returns a sparse set of feature weights,^^(Equation 2). In the application of DNAme clocks, a chronological age prediction is given by taking a linear combination of the learned weights and a sample’s surrogate measure of CpG methylation, known as beta values (Equation 3).

[0047] The interpretation of this method from the standpoint of typical biological experiments is that most of the experimental values are discarded by filtering them down to an extremely small subset, which can yield a strong linear correlation with the sample donors’ chronological ages, but only if these remaining values are adjusted by arbitrary weights.

[0048] Robustly different epigenetics of young versus old cells and tissues, or upon ectopic expression of Yamanaka factors, serious illnesses, etc., can be detected with Student’s t-test or other simple statistical analyses of well-controlled experiments that are more recently performed on single cells. The advantage of EN and other machine learning methods is in processing large datasets; however, they assume that every signal is valid and equal, while the experimental artifacts increase with age and disease due to DNA damage, mutations (C to T and others). Additionally, aging and disease involve lymphoid / myeloid, fibrotic, inflammatory, and senescent cell shifts in composition of various tissues; in DNAme arrays these are mixed withUC Ref: BK2023-047-3 BFF Ref: BERK-518WO individual DNAme changes, and probe artifacts. Of note, cell-intrinsic and tissue composition changes are much more accurately determined by Flow cytometry, single cell RNA Seq, ATAC- Seq, and in situ omics.

[0049] In biological terms, each EN linear correlation is one independent study (N=1), in which all samples are simultaneously compared to each other once and the low p-values signify that the best-fit line could not have been generated by chance, but importantly, do not reflect reproducibility from one clock to another. When the number of independent experiments is N>1, the accuracy of DNAme clocks may falter, which is often attributed to batch effects, implying difficulties with blinded analysis of a new sample on a trained clock. And the same is true for the predictors of chronological age that utilize neural networks, and those based on large proteome data, such as iAge or Somascan approaches.

[0050] Batch effects could be due to the natural important differences in DNAme, which is known to be highly dynamic, and due to differences between environmental and lifestyle factors, as well as experimental differences in instruments, reagents stocks, researcher introduced differences, etc. These variabilities are inherent and largely uncontrolled for, meaning that after successful DNAme arrays, no EN input is known to be more or less accurate than another, and none normalizes for the other.

[0051] Typical biomedical approaches yield batch-effect resilient discovery because the data are not compared once. Instead, a statistically robust number of young and old samples are studied independently from start to finish in experiments for the candidate parameters and, moreover, through several redundant assays, for example, mass spectrometry corroborates Western blotting which corroborates immunofluorescence; then low p-values signify results which are robustly reproducible even with different reagents stocks, etc. I. Equations Equation 1:^^^, y, w, ^, ^^ =1−^^1 − ^^ ^ + + ^individual samples and the columns correspond to individual CpGs, “y” is a vector of ages that correspond to the samples in matrix X, “n” is the number of samples, “w” is a vector of regression weights that scale the CpG beta values, “α” is the penalty hyperparameter that is learned during EN training, “ρ” is the L1 ratio, which is used to balance between Ridge and Lasso regression (ρ=1 simplifies to pure Lasso regression, ρ=0 simplifies to Ridge regression), “||*||1” and “||*||2” are the L1 and L2 norms, respectively.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO Equation 2: w^ = m^in ^^^, y, w, ^, ^^ wherein “^^” is a sparse vector of weights learned by the EN training algorithm. Equation 3: prediction!= w^ • x!,∗wherein “prediction” is the age prediction for the ith sample, “xi,*” is the row vector in X for sample i. Equation 4:% =∑^'! − '̅^^)! − )*^−'̅^^ − *^^(a generalEquation 5:MedAE^), ) / ^= 01234^^|)^ − ) / ^|, … ,|)7 − ) / 7|^wherein “MedAE” is the general median absolute error function, The variables are general and defined ad hoc, “y” refers to an actual value, while “) / ” refers to the corresponding estimated or predicted value. Equation 6:81 9=:; <!9wherein “μj” is the mean jth CpG, “N” is the number of samples used“βij” refers to the beta value of the jth CpG for the ith sample. Equation 7:'̅!9 = =<!9 − 89=wherein “'̅!9” refers tobetween the mean of the jth beta value and the ith sample’s beta value at the jth beta value. Equation 8:> =1; ; '̅∗ ?9wherein “ζ” is the mean differences from the mean for a subset ofover a subset of CpGs, “'̅?∗9 ” is the absolute difference from the mean for the ith young sample at the jth CpG. Equation 9:@ =1 wherein “@!” is the young-of the absolute differences from the means for the ith I. Results

[0052] Upon closer examination, it was found that while the EN model yields high linearity (Pearson’s r approaching 1) over the complete chronological age range, weaker linear correlations between DNAme and chronological age manifest in tests on separated age rangesUC Ref: BK2023-047-3 BFF Ref: BERK-518WO (FIG 6), ranging from 0.16 ≤ r ≤ 0.8. In support of a low natural association between the age and different states of DNA methylation, t-SNE analysis on the database did not demonstrate data clustering on chronological age (FIG 7). This is to be expected from the genetic and environmental variability between people compounded by the high dynamics of epigenetics that is influenced by such parameters as sense of well-being, financial status, circadian rhythm, etc.

[0053] From the earliest EN-based age predictors, errors in chronological age prediction were attributed to age acceleration, and there have been studies where age acceleration was reported for individuals with a disease on EN DNAme clocks; while the trend was true for some age groups, it was less reliable for other age groups. Determinations of true age acceleration due to disease are difficult to uncouple from the random variation in an experimental assay, particularly considering a lack of statistical support from predictions for replicate samples of the same individual on a particular clock, or on multiple clocks, or for the same disease, or for diseases that change biological age in opposite directions. Moreover, EN errors in chronological age were not experimentally verified nor confirmed with N>1.

[0054] Conceptually, how can EN clocks accurately predict both chronological age and age acceleration caused by disease? For chronology, the features most similar between the individuals of the same age should be selected and ranked highly by weighting, and for a disease or environmental changes, the variable between the individuals of the same age features need to be selected and highly weighted. This question is particularly relevant for the older ages where increase in variability of DNAme between individuals rises due to prevalence of different diseases, yet EN aims to minimize the weights of features with high variability between the samples of the same age in its goal of minimizing the cost function.

[0055] To address these points and to examine whether and to what degree EN clocks distinguish health from disease, studies with multiple health and disease datasets were conducted. Namely, standard EN regression models were constructed with 27K (small) and 450K (medium) array datasets and these models were tested with the original and several additional DNAme datasets, in which healthy blood donors were compared to donors with disease: Rheumatoid arthritis, Down syndrome, BRCA-1 cancer, Werner syndrome, Multiple Sclerosis, Crohn’s Disease, Ulcerative Colitis, Irritable Bowel Disease, and an HIV dataset that did not have a parallel healthy control. See FIG 1, Panels A−D and FIG 8, Panel A. To measure model performance, the correlation coefficient (Pearson’s r) was used between the predicted and actual ages for the test data, as well as the median absolute error (MAE) for the test data (Equations 4,5).UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0056] As expected, the EN models were testable with all datasets, making relatively accurate predictions of the chronological age (FIG 1, Panels A, B and FIG 8, Panel A). Interestingly, the EN models which were trained on the small and medium datasets differed from each other in reporting age acceleration versus deceleration for the same samples and through the entire chronology. The EN model that was trained on the small dataset was often less accurate in tests than the medium EN model, suggesting that the number of CpG probes to choose from in training is important for the age predictability, as expected for a statistical method.

[0057] Importantly, the EN age predictions by both small and medium models were overlapping for the disease and healthy cohorts (FIG 1, Panels A−D and FIG 8, Panel A). This is biologically inaccurate and demonstrates that these age predictions do not reflect biological age. Particularly, arthritis is known to significantly change the composition of PBMCs toward inflammaging; in contrast, in Down’s Syndrome there is a pathological juvenile effect on blood cells, with a prevalence of childhood leukemias and less mature circulating PBMCs.

[0058] Thus, DNAme clocks report linear chronological determinations rather well, but the specifics differ as per training on a particular dataset, and EN clocks do not distinguish health from diseases in general or in progeronic vs. pathologically juvenile disorders.

[0059] Interestingly, when the datasets from BRCA-1 studies that were done with the 27K DNAme array were tested, the small EN model failed to predict the chronological ages of the healthy donors but produced accurately linear chronological predictions for the people with the BRCA-1 mutation in the absence of cancer and with BCRA-1 cancer (FIG 1, Panel C). These results strengthen the point that either chronological age or disease, but not simultaneous determinations, are provided by the EN method.

[0060] To confirm and extrapolate the conclusion on the overlap of health and disease by the EN models, Werner Syndrome, which causes premature aging due to a mutation of DNA helicase, was studied. The samples of blood from progeric patients are naturally very rare and their chronology is not broad; nevertheless, there are published studies with six age- and sex- matched blood samples of Werner Syndrome patients and healthy controls. Analysis of these data confirmed that both small and medium EN models fail to identify differences between health and disease even in this case of progeria, and that medium EN model reports more accurate chronological ages than the small EN model. See FIG 1, Panel D.

[0061] Lastly, the results show that even though the small EN dataset was from the combination of samples with diseases and healthy samples, and the medium EN dataset was from relativelyUC Ref: BK2023-047-3 BFF Ref: BERK-518WO healthy individuals, the small EN model reported age acceleration for most of the samples when tested on the medium dataset. See FIG 1, Panel E. Of note, all CpG probes of the small dataset are present in the medium dataset.

[0062] It was also found that whenever the clocks were reporting linear chronological correlations with DNAme, a simple OLS regression on the set of clock CpGs also yielded linear correlations, and these also overlapped health and disease. See FIG 8, Panel B. OLS regression on the EN-selected CpGs yields more accurate results than the EN model, suggesting that experimental differences are not the only source of batch effects, and the EN process contributes to these effects. See FIG 8.

[0063] Expecting that there are differences in epigenetic states between health and disease from the body of published work, multidimensional UMAP studies demonstrated robustly distinct clustering of healthy vs. disease cohorts for all studied datasets. See FIG 2.

[0064] Thus, the input DNAme data accurately reflected the epigenetic differences between health and disease, but the search and capture of the best linear correlations with chronological age by EN training misses these biologically important differences. This is most likely due from feature selection algorithm for linearly accurate chronological prediction minimizing the weights of those features which vary among individuals of the same age due to a disease.

[0065] To follow with the examination of the biological significance of age predictions by EN, the beta values of the clock CpGs were analyzed in more detail. It was found that EN does not inherently select CpGs that change in their methylation with age, and hence are biologically meaningful in such a way. See FIG 3, Panel A. All the clock CpGs that had been selected in the initial model (Model 0) were removed from the dataset and repeated the EN training again. This yielded a new clock (Model 1) that was nearly as accurate as the original in predicting chronological age, and once again, had no preference for biologically meaningful CpGs. See FIG 3, Panel B.

[0066] Observing the R2 values for the individual age-correlations of the clock CpGs demonstrated that many of the highly weighted features that were selected have little-to-no correlation with age, while some selected CpGs that were given low weights have strong age- correlation. See FIG 3, Panels A, B. In fact, the mean coefficient of variation for the age correlation of the selected features (red dashed lines in FIG 3, Panels A, B) was lower in the Model 0 as compared to Model 1, which had the first set of selected features removed from itsUC Ref: BK2023-047-3 BFF Ref: BERK-518WO training data. Hence, on average, the CpGs that do not change in their methylation with age were more frequently selected and highly weighted by EN.

[0067] It is improbable that cytosines that do not have age-correlated methylation changes contribute to age-specific changes in DNA methylation (alone or by interaction with other cytosines).

[0068] Following the hypothesis that linear differences of DNAme between young and old are all that are necessary for constructing accurate chronological age predictors, a modified version of the forward step-wise selection (FSW) algorithm to compare the performance of a simple OLS regression clock, with hypothesis-driven feature selection, against EN regression was implemented. With such logical CpG selection, a simple OLS model would suffice for accurate predictions of chronological age and with low variance when predicting on new datasets. Thus, a 450K DNAme array dataset (GSE40279, N=656) was used and regressed chronological age on each CpG feature independently, recording the magnitude of the age correlation of each. The CpGs were ranked by the magnitude of their age correlation. The data set was split (80 / 20) into training and validation sets and a second dataset (GSE42861, N=689) was used for an out-of- sample test set. An OLS model was trained using the top ranked CpG from the training dataset as the only predictor, and the prediction accuracy (Pearson’s r and MAE) was recorded on the validation and test sets. This process was repeated iteratively, adding the next highest-ranked CpG into the model until the number of features which yielded the minimum MAE in the test set was reached. FIG 3, Panel C shows the FSW OLS model is optimized for predictive accuracy (per MAE) with the top 116 age correlated CpGs in the model. Even the top ranked CpG by itself yields a model with r > 0.8, and 80 such CpGs yield an OLS model with r > 0.94 and MAE of 3.65 years in the test set.

[0069] Importantly, the modified FSW OLS clock displays high accuracy and low variance on out-of-sample data. See FIG 3, Panel D. These results show that there are certain CpG sites have a strong enough linear correlation with age that black box machine learning methods, which are partly used to capture non-linear dependencies, are not necessary for constructing chronological age predictors. However, the health and disease cohorts overlapped in the FSW OLS clock model, which indicates that methylation at these age-correlating CpGs do not inherently detect the effects of disease. See FIG 3, Panel D. The hypothesis driven FSW OLS model was testable on all examined datasets, in which healthy cohorts were compared to patients (FIG 3, Panel D, FIG 8) and always overlapped healthy and diseased datapoints.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0070] Table 1 shows the CpGs, gene names, and their roles for the genes which have high linearity of their CpG beta values with age and overlap between the 450K datasets (see Methods). Table 1 CpGs having high chronological age-correlation and overlap between multiple datasets CpG / Probe Chromosome Map Location Strand Gene Gene Accession No. cg01974375 1 151298954 F PI4KB NM_002651 cg06784991 1 53308768 R ZYG11A NM_001004339cg22454769 2 106015767 R FHL2 NM_001039492 cg24079702 2 106015771 R FHL2 NM_001039492 cg07553761 3 160167977 F TRIM59 NM_173084 cg08262002 4 16575323 F LDB2 NM_001290 cg06493994 6 25652602 R SCGN NM_006998 cg16867657 6 11044877 F ELOVL2 NM_017770 cg21572722 6 11044894 F ELOVL2 NM_017770 cg22736354 6 18122719 F NHLRC1 NM_198586 cg24724428 6 11044888 F ELOVL2 NM_017770 cg03473532 7 131008743 R MKLN1 NM_001145354 cg08097417 7 130419133 R KLF14 NM_138693 cg14361627 7 130419116 R KLF14 NM_138693 cg16419235 8 57360613 R PENK NM_001135690 cg26290632 8 91094847 F CALB1 NM_004929 cg01812894 9 75568506 R ALDH1A1 NM_000689 cg07080372 11 796607 R SLC25A22 NM_024698 cg08160331 11 75140865 F KLHL35 NM_001039548 cg01820374 12 6882083 R LAG3 NM_002286 cg11649376 12 81473234 F ACSS3 NM_024560 cg17436656 12 53627106 F RARG NM_000966 cg04875128 15 31775895 R OTUD7A NM_130901 cg08090640 17 41159289 R IFI35 NM_005533 cg19283806 18 66389420 F CCDC102B NM_001093729 cg07547549 20 44658225 R SLC12A5 NM_020708 Genome Version = GRCh37.p13 (GCF_000001405.25, which the chromosome accession numbers are: 1=NC_00001.10, 2=NC_00002.11, 3=NC_00003.11, 4=NC_00004.11, 5=NC_00005.9,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0071] The biological relevance of the genes annotated to the CpGs used in the FSW OLS model is present a priori, their methylation changes chronologically. An analysis of the roles of the top CpG annotated genes of the FSW OLS that overlap between the 450K datasets, suggests that they mostly reflect already known aging changes, such as increase in body mass / fat, repeated instances of viral infections and inflammation, pre-cancerous growths, alcohol detoxification, changes in teeth / hair / skin, etc.. Other notable roles include protein degradation, maturation of the extracellular matrix and perception of pain.

[0072] A standard EN clock was constructed and the CpGs that were excluded from the model by EN were systematically removed instead of the clock CpGs. See FIG 4. This interestingly, also significantly changed the original clock CpGs that were picked, plateauing at 17−20% of the unused original set after removal of 3−4% of irrelevant CpGs. See FIG 4, Panel A. This behavior disagrees with the notion that EN clock CpGs are biomarkers of aging because biomarkers should not change when non-biomarkers are deleted from a dataset.

[0073] In a different approach, the property of the EN to yield different clocks was uncovered by simply running the model over 500 times and getting different clock CpGs, likely due to the stochasticity of the gradient descent algorithm. Namely, Hannum et al. (2013) used bootstrapping to find highly conserved clock CpGs. From this process they recovered a model consisting of 71 CpGs. Efforts to create more accurate DNAme clocks have recently expanded beyond EN models, including neural networks. For example, DeepMAge is a recent DNAme clock created via deep learning. DeepMage is highly accurate, but only 15 of its 1,000 CpGs are clock CpGs in the EN model, and these CpGs also become changed when non-significant non- clock CpGs are removed. See FIG 4, Panel B.

[0074] Similarly, the 71 Hannum clock CpGs are not resilient to non-clock feature removal. See FIG 9. Whether the original, versa new EN clock CpGs had better or worse linear predictability of chronological age was determined. This study demonstrated that the original, and the changed clocks perform nearly equally well. See FIG 4, Panel C. Simple OLS regression on each set: original and changed, performed about as well as the EN clocks. See FIG 4, Panel D.

[0075] Thus, it appears that the clock CpGs in the EN model, even those which are the same as Hannum’s 71, or most of DeepMAge 1000 CpGs, are not biomarkers. It also seems that the accuracy accomplished by complex methods, such as deep learning, can also be accomplished with simple OLS (FIG 1− FIG 4 and FIG 9) simply because there exist CpGs with methylationUC Ref: BK2023-047-3 BFF Ref: BERK-518WO which strongly correlate to chronological age, and not because of any non-linear, biological-age- relevant interactions are being captured. The lack of a defined biomarker capacity of the clocks is expected from the general blindness of ML to the physical meaning of the experimental values.

[0076] Additionally, ML methods that rely on stochastic gradient descent for optimization appear to suffer from the pitfall that removal or absence of certain features changes the abstract surface of the feature space which influences the local optima (i.e., feature sets, weights and hyperparameter values) that gradient descent can converge upon. This point was experimentally supported by studying the effects of removing the non-significant CpGs on the energy landscape of Elastic Net. See FIG 4, Panel E. Overall, predicting chronological age does not seem to depend on a unique set of CpGs, and there is a lack of evidence for significance of the selected CpGs as biomarkers of aging.

[0077] Finally, the biological age from the DNAme arrays data was determined from the variation in CpG methylation (“noise”) of those CpGs that are the most regulated and on average remain invariable in their methylation through lifespan. The age-imposed increase in the DNAme noise is visible when looking at the scatter plot of beta values of individual CpGs that are on average age-independent in their methylation across the lifespan; an example of such a CpG that remains on average unmethylated is shown in FIG 5, Panel A. Of note, a decrease in CpG noise from 20−30 years is biologically expected, during versus after maturation of the immune system.

[0078] To capture this visibly noticeable biological phenomenon, focus was on the CpGs that have Pearson’s r of beta values vs. age between 0.01 and 0.05, e.g., on average not changing in methylation with age, and excluding potential experimental errors of probes and arrays, e.g., zeros. This yielded 6,282 CpGs that were shared between different 450K array datasets. Next, these 6,282 CpGs were analyzed to identify those that become dysregulated (noisy) with age and this yielded only 50 CpGs as potential molecular detectors of age-related epigenetic noise (see Methods). The 50 CpGs are set forth in Table 2: Table 2 50 CpGs selected as "noise" biomarkers CpG / Probe Chromosome Map Location Strand Gene Gene Accession No. cg00868523 19 54387036 R PRKCG NM_002739 cg01767885 16 1147214 R C1QTNF8 NM_207419 cg01869186 7 100423987 F EPHB4 NM_004444 cg01900413 11 128419356 F ETS-1 NM_001143820UC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg02449103 6 31627711 F C6orf47 NM_021184 cg02247160 3 41243829 F CTNNB1 NM_001904 cg04153991 3 111577705 F PHLDB2 NM_001134438 cg04495670 11 69634296 F FGF3 NM_005247 cg05557449 17 46801809 R C17orf93 NR_024103 cg06362282 6 33245893 R B3GALT4 NM_003782 cg06494091 9 80646128 F GNAQ NM_002072 cg06918474 5 126205081 F 3-Mar NM_178450 cg07165260 16 85062881 F KIAA0513 NM_014732 cg08499046 11 31824327 F PAX6 NM_001604cg05207943 11 82095488 R * * cg06899313 6 117394045 F * * cg07710266 6 134436740 F * * cg12518535 13 24477318 F * * cg14400541 2 206692685 F * * cg14657277 2 123419093 R * * cg20037507 3 62363248 R * * cg20445245 6 45631799 R * * cg24201362 6 467681 F * * cg25734842 1 228657381 F * *UC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg27219748 16 84301329 R * * cg27317439 10 102976549 F * * Genome Version = GRCh37.p13 (GCF_000001405.25, which the chromosome accession numbers are: 1=NC_00001.10, 2=NC_00002.11, 3=NC_00003.11, 4=NC_00004.11, 5=NC_00005.9, 6=NC_00006.11, 7=NC_00007.13, 8=NC_00008.10, 9=NC_00009.11, 10=NC_000010.10, 11=NC_000011.9, 12=NC_000012.11, 13=NC_000013.10, 14=NC_000014.8, 15=NC_000015.9, 16=NC_000016.9, 17=NC_000017.10, 18=NC_000018.9, 19=NC_000019.9, 20=NC_000020.10, 21=NC_000021.8, 22=NC_000022.10, X=NC_000023.10, and Y=NC_000024.9) * Not characterized

[0079] The absolute differences from the mean (ADM) of the 50 CpGs were summed for each sample (individual), and these sums were normalized by the sum of the ADMs of the young healthy cohort (25−27 years). See FIG 5, Panel B, (Methods, Equations 6−9). An upward, but not strictly linear increase in DNAme noise was detected by this quantitative analysis. See FIG 5, Panel B.

[0080] The median of these data, e.g., the average progression of epigenetic noise with chronological age of the samples from healthy people and patients with arthritis, which yielded natural polynomial curves of epigenetic aging with discernible plateaus, was identified. See FIG 5, Panels C, D. The shape of the natural healthy curve of biological DNAme aging was clearly different from that of arthritis-imposed when measured on the 5n polynomial curve or on the 8n polynomial curve. See FIG 5, Panels C & D. The transition timepoint between the relatively healthy adult plateau and an upward shift to a higher biological age fits well with known biology and points out important hot-spot ages. The late life decline in the biological age due to Arthritis, as compared to healthy control, is expected from the diminished complexity of PBMCs that become skewed toward the inflammatory myeloid subset.

[0081] Testing for broader applicability, the biological ages of samples from people with IBS and their healthy controls, were measured using the same 50 CpGs, e.g., the noise barometer. The results were similar in demonstrating the polynomial progression of biological age in healthy people and a relative increase of biological age in chronologically young-adult (but not old) patients with IBS. See FIG 5, Panel E. Further confirming the batch effect resilient approach, the 50 CpG noise barometer detected biological age increase in patients with HIV. See FIG 5, Panel E. Moreover, the shape of the 50 CpG biological age curve was different for arthritis, IBS and HIV. See FIG 5, Panel F, suggesting a novel population-based diagnostic capacity. The same general phenomenon could be observed, albeit with lower resolution, by simply calculating absolute difference from the mean of beta values for CpGs with the (r) of beta values v. age < 0.05 and plotting their correlation with chronological age in 5-year bins. See FIG 10.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0082] Expanding and validating the batch-effect resilience, this approach was applied to datasets of healthy and disease patient blood samples that were assayed for their DNAme in multiple independent studies unrelated to each other. The shape of the curve of biological aging in the 50 CpGs, was akin to individual healthy datasets that are shown above; and combined disease was clearly discernible from combined health. See FIG 5, Panel G.

[0083] With aging, the differences in the noise of what should be invariable CpGs, increased in a common upward trend for healthy individuals and those with illnesses. See FIG 5, Panel B−E. This is biologically expected from the broad increase of epigenetic changes that are caused by various diseases and the profound increase of disease with age (even if these are not specifically IBS, PD, Arthritis or HIV).

[0084] Interested to know if these noise detectors, the CpGs that are on average invariable in their methylation with age and are heteroscedastic are represented in the EN linear models and the stepwise OLS model, cross-referencing was performed, and the data established complete independence of the noise-based measurement of biological aging from the chronological correlations with adjusted by weighting data on CpG methylation. See FIG 5, Panel H.

[0085] These results suggest that biological age can be measured in a batch effect resilient method through quantification of the SD of the should be invariable CpGs, which become heteroscedastic with age. In this method, DNAme noise is the biomarker and particular set of CpGs is the detector of such biomarker. This method is different from and moreover, orthogonal to, the linear predictors of chronological age, e.g., population statistics. I. Discussion

[0086] As described herein, the 50 CpGs can be used to determine the biological age of a given subject. The 50 CpGs can be used to measure the chronological age of a subject as well as the subject’s biological (epigenetic noise) age. The 50 CpGs can be used to distinguish healthy from diseased, and can also be used to distinguish between different pathologies.

[0087] These measurements of epigenetic age are the quantification of primary data with no data adjustments, which is a significant improvement over predictions. Different shapes of different diseases might enable epidemiology of a specific disease in a population, based on the curve of epigenetic noise.

[0088] The 50 CpG noise barometer also reveals a common trend of elevated biological age in > 50 years old patients, yet either no change or a decline in biological age of PBMCs in the older patients. This is interesting and fits with the notions of age-imposed incline in multiple diseasesUC Ref: BK2023-047-3 BFF Ref: BERK-518WO (not just the one studied in a dataset) and complexity of PBMCs being lowered by inflamaging and autoimmune disorders, thus pathologically reducing the biological variability, i.e., noise.

[0089] An age-specific increase in biological noise has been noticed for many parameters, including DNA methylation, but it was not used for the derivation of a natural curve of biological or epigenetic aging, or for measuring it, alone or together with distinguishing health from disease. The noise detector CpGs are biologically relevant, and their future in-depth studies can improve the understanding of the key time points and the mechanisms of deregulation of gene expression, signal transduction, pertinent to the aging and pathology of tissues and organs.

[0090] The likely reason for the noise-detectors CpGs to be on average invariable in their methylation with age is that they are in the regulatory regions of genes that are vital at constant levels, such as coding for ribosomal proteins, core transcriptional factors, E3 ligases, regulators of cholesterol and glucose metabolism, MAPK signaling, beta-catenin, etc..

[0091] If the determination of chronological age is the task, the results herein show that OLS on the age-dependent CpGs is sufficient and biologically it is a more parsimonious alternative to EN or other black-box ML methods. The results demonstrate that for generating an accurate age predictor from DNAme array values, EN can be replaced by the much simpler OLS, constructed using forward stepwise selection. These data also establish that generation of a sparse set is not a unique benefit of EN. Lastly, summing the age independent CpGs with age-related CpGs is not needed, questioning the inclusion and particularly, high ranking by the EN clocks of the CpGs that do not change in their methylation with age. For the OLS model, there is also a user- friendly flexibility in focusing on the best linearity, the lowest MAE, or a balance between the two.

[0092] Methods in the art may be used to measure the methylation levels of CpGs. For example, reduced representation bisulphite sequencing (RRBS), methylation restriction enzyme sequencing (MRE-seq), methylation DNA immunoprecipitation sequencing (MeDIP-seq), methyl-CpG-binding domain protein sequencing (MBD-seq), methylation DNA capture sequencing (MethylCap-seq), short-read sequencing (using, e.g., Illumina's HiSeq / MiSeq or LifeTech's Ion Proton platforms), target methylation sequencing (e.g., Bisulfite Padlock Probes, BSPP, etc.), bisulfite sequencing, combined bisulfite-restriction analysis, pyrosequencing, methylation-sensitive single-strand conformation analysis (MS-SSCA), high resolution melting analysis (HRM), methylation-sensitive single nucleotide primer extension (MS-SnuPE), base- specific cleavage / MALDI-TOF, methylation-specific PCR (MSP), MspI cleavage, and theUC Ref: BK2023-047-3 BFF Ref: BERK-518WO methods described in US 6214556; US 5786146; US 6017704; US 6265171; US 6200756; US 6251594; US 5912147; US 6331393; US 6605432; US 5786146; US 6143504; US 6596493; US 6884586; US 6300071; US 7195870; US 20030148327; US 20030148326; US 20030143606; US 20050009059; US 20050196792; US 20060292564; US 20220136037, which are herein incorporated by reference in their entirety. II. Additional Discussion of previous epigenetic clocks

[0093] Elastic Net (EN) is the most used machine learning (ML) approach in constructing DNAme clocks [Horvath (2013); Hannum (2013); Belsky (2022); Levine (2018); Lu (2019)]. The main principle of EN is to use regularization on datasets where the number of features far exceeds the number of samples and perform feature selection toward an absolute shrinkage of feature weights, allowing for the fit of a linear model to a relationship between the response variable and the predictors [Zou (2005)]. The EN model is trained using stochastic gradient descent to minimize a cost function (Equation 1) and returns a sparse set of feature weights, wˆ (Equation 2). Prediction is given by taking a linear combination of the learned weights and a sample’s surrogate measure of methylation, e.g., the beta values (Equation 3) [Du (2010)]. It must be noted that a least-error linear relationship is found regardless of whether the relationship is in fact linear. It is also notably different from typical biological experiments, that after most of the data are filtered out the remaining values are multiplied by weights that may be biologically arbitrary, as long as they mathematically enforce the hypothesis of linear change, FIG 12A.

[0094] DNAme clocks do not reflect the overall epigenetic state of a cell or a tissue and do not predict the process of epigenetic regulation of gene expression by CpG islands. Instead, the clocks are based on a small subset of cytosines that tend to be isolated and scattered sparsely through the genome, FIG 12B, 12C. Such disconnect of the clocks from the current understanding of gene regulation and purely correlative nature of predictions questions the relationship to the underlying biology. The DNAme array data are not derived from exact and well-controlled sequencing of methylated vs. unmethylated DNA. They are estimates of hybridization strength of 480-900K probes, read as optical signals of different intensity, FIG 12A. Probe hybridizations are largely based on the methylation levels of a single end-of-probe cytosine. All signals are normalized and represented as beta values from 0 (unmethylated) to 1 (methylated), based on the maximum and minimum fluorescent intensities of a given cytosine read assay or Illumina run. Random differences in the hybridization efficiency of a probe and in the maximum and minimum fluorescent intensities of an assay or run are expected, and thisUC Ref: BK2023-047-3 BFF Ref: BERK-518WO influences the beta values and hence the predictions. After training on measured parameters, for example age or health scores, an EN model returns weights by which the beta values are multiplied before being summed; the sums are the correlate-predictor. There is no biological reason for weighting these beta values, but since the task is to correlate the measured axis linearly, EN returns the data-adjustment weights for a mathematical solution to the least- squares linear fit, FIG 12A [Belsky (2022); Levine (2018); Lu (2019)].

[0095] The measured and then predicted axes are conceptually always a regular set of numbers (1, 2, 3… 10) be these inputted as chronological age (first generation clocks), levels of a few proteins followed by years-to-death (GrimAge), a composite of clinical parameters associated with age (PhenoAge), a health scale for people in ~26-50 age range (PACE), or any measurable parameter. In EN, no physical meaning is implied to these numbers.

[0096] As expected, EN regression can confirm well-known, robust differences of age, disease, ectopic epigenetic determinants such as Yamanaka factors, trisomy, etc., cases where DNAme arrays and bisulfate sequencing yield similar data [Kantlehner (2011); Lake (2018); Cheung (2018); Kim (2015); Li (2011)]. However, artifacts of non-specific probe hybridizations are expected to increase with age and disease due to DNA damage and mutations (C to T and others), skewing the data distribution in the outputs of DNAme arrays [Zhou (2018); Hop (2020)]. Aging and disease also involve lymphoid to myeloid, fibrotic, inflammatory, senescent, and other shifts; these may be accurately determined by Flow cytometry, single cell RNA Seq, ATAC-Seq, and in situ omics, but are unresolvable in bulk DNAme arrays [Quach (2017); Mann (2018); Kirschner (2017)].

[0097] In typical biomedical research, a statistically robust number of samples is studied independently from start to finish, however, the metrics of clock performance (e.g., mean squared error, median absolute error, p- value) are often correlations based on an experimental n=1, FIG 12D, 12E. Low p-values mean that the best-fit line could not have been generated by chance in this one experiment, but do not signify accuracy from one experiment (dataset) to another. The natural dynamics of DNAme and experimental differences in instruments, reagents stocks, researcher introduced, etc. [Bell (2019); Jones (2015); Unnikrishnan (2019); Bergsma (2020)] are inherent and largely uncontrolled for, meaning that after a successful DNAme array assay, no input / output of EN is known to be more or less accurate than another, and none normalizes for the other. The dataset (split into a training and test subsets) is a subject to above- mentioned batch effects, and thus, the possibility of misinterpreting random experimental variation for biologically relevant changes. The reliance of clocks solely on hybridization arraysUC Ref: BK2023-047-3 BFF Ref: BERK-518WO also contrasts with typical molecular biology approaches that study each parameter through several redundant assays [Mehdipour (2020)].

[0098] The idea of a biological age clock has been explored over the past decades through studies of the Hayflick number, telomere length, DNAme and proteomics, all under the hypotheses that each of these methods reflects a life-long linear change of a biomedically important parameter. Clock’s CpGs have been profiled as biomarkers but such bioinformatics seems to be currently de-emphasized [Belsky (2022); Salameh (2020)]. With respect to the notion of linearity, many studies demonstrate a healthy young-adult plateau with stably controlled epigenetics, gene expression and protein levels [Ashraf (1998); Lai (2019); Simon (2015); Gibney (2010)]. And not all patterns of aging are determinants, for example, hair greying, or loss might be age and health-linked and can be used in EN-based predictions, but hair restoration does not yield health or youth and hair is not a determinant of mortality [Rhodes (1998); Krupa (2009)].

[0099] Diseases contribute to health decline and shorten life span; the diseases of age can be detected by many well controlled analyses, including genetics, epigenetics, and clinical tests without clocks that, in some cases, differ from clock predictions [Drewelies (2022); Janssen (2005); Reichert (2012); French National Institute for Demographic Studies (2018)]. The question is not whether epigenetics can be used to detect changes in health or if health declines with age, compounded by diseases (both are given), but whether ML clocks predict these changes with high resolution or as early as other methods.

[0100] Here, we reveal the capacities and limitations of DNAme clocks of all generations, by expanding the n=1 experimental to a more systematic examination of clock performance on many different DNAme array datasets of healthy controls and patients with various diseases. Additionally, we show that even though EN regression models linearly predict age and risk of mortality from DNAme array data, the clocks are not based on any specific changes in DNA methylation with age, or dysregulation of such methylation, e.g., a change in SD of beta values. Lastly, we describe a conceptually different measurement (not a prediction) of persons’ biological age, which we term a noise barometer, reflecting the concept of pressure of aging and disease. II. Results II. a. The robustness of linear correlations varies over the age ranges

[0101] We found that while EN-trained DNAme models yield high linear correlation (r > 0.9) between age and DNAme over the complete age range, weaker linear correlations manifest inUC Ref: BK2023-047-3 BFF Ref: BERK-518WO tests on binned age ranges (FIG 13), from 0.16 ≤ r ≤ 0.8. Specifically, bins of 18-30yr (during the maturation of the immune system) and 80-100yr have the highest correlation of r=0.8, with the rest of the age ranges are more moderately correlated (0.16 ≤ r ≤ 0.5). The uneven correlations of different age bins to DNAme changes are consistent with the notion of non- linear changes in biological age over the lifespan, e.g., the healthy plateaus and steeper health decline after combinatorial tissue disrepair [Ashraf (1998); Lai (2019); Simon (2015); Gibney (2010)].

[0102] A non-linearity of epigenetic changes through the lifespan should be expected, due to genetic and environmental variability between people, compounded by the high dynamics of epigenetics that is influenced by such parameters as sense of well-being, financial status, circadian rhythm, etc. [Kanherkar (2014); The dynamic epigenome (2013)].

[0103] The non-linearity of biological age, the high variation of the assay and, hence, of the data, which is used for EN training, might result in diminished accuracy of correlations, particularly for some age groups. On the other hand, the overall estimates of being young, middle-aged, or old are reliably well predicted by the EN DNAme clock. II. b. Tests of accuracy and biological relevance of predictions

[0104] Two claims are made by studies of EN-based predictors, first that the errors in prediction by DNAme clocks are attributed to age acceleration in particular individuals. In support for this attribution, a similar acceleration was seen for some diseases, albeit only for some age groups [Horvath (2013); Hannum (2013); Belsky (2022); Levine (2018); Lu (2019); Dugue (2018)]. A second claim is that because clocks trained on one sample set can work on out-of-sample set data, this indicates some biological relevance of EN predictions and of CpG selection.

[0105] To examine these claims, we performed systematic fail-tests of DNAme clocks with large DNAme datasets from nine independent studies. First, we reconstructed EN regression models that replicate the published first-generation clocks, and tested them with the original and out-of-sample DNAme array datasets, in which healthy subjects were compared to those with disease: Rheumatoid arthritis, Down syndrome, BRCA-1 cancer, Werner syndrome, Multiple Sclerosis, Crohn’s Disease, Ulcerative Colitis, Irritable Bowel Disease, and an HIV dataset that did not have a parallel healthy control, FIG 14A and FIG 15. Model performance was assessed by correlation coefficient (Pearson’s r) between the predicted and actual ages and median absolute error (MAE) (Equations 4,5).UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0106] The models fairly accurately predicted the age of samples, but with an error of age acceleration throughout chronology (shifted above the line) and with overlap in age predictions for the patients and healthy subjects, FIG 14A and FIG 15. This is biologically inconsistent, as arthritis significantly increases PBMC inflammaging [Karouzakis (2011); Nemtsova (2019)], Werner Syndrome is a disease of premature aging (caused by a mutation of DNA helicase) [Oshima (2017); Guastafierro (2017)], and Down syndrome has a pathological juvenile blood phenotype with a prevalence of childhood leukemias and less mature circulating PBMCs [Webb (2007)]. In tests on BRCA-1 studies, the EN model had very high MAE and poor correlation with the age of the healthy subjects but produced fairly accurate predictions for the people with the BRCA-1 mutation, and cancer, FIG 14A.

[0107] The 27K EN model that was from the combination of disease and healthy samples surprisingly reported age acceleration when tested on the 450K dataset that was from relatively healthy individuals, FIG 15.27K dataset CpG probes are present in the 450K dataset. Lastly, we found that a simple ordinary least squares regression (OLS) on the set of clock’s CpGs also yielded linear correlations, and these were somewhat more accurate in predicting age (lesser age acceleration through chronology) than the EN model, FIG 15.

[0108] In testing PhenoAge, thought to be biologically more relevant, we expanded the approach to multiple independent studies with 450K DNAme arrays on Arthritis, Multiple Sclerosis, Down Syndrome, Parkinson’s, IBS, Werner Syndrome. We analyzed whether predicted age is on average higher for disease samples as compared to their healthy controls. Additionally, instead of MAE on one 450K correlation, which is the Median from an experimental n=1, we quantified the errors as the arithmetic Means of predicted minus actual age, with Standard Deviation (SD) and p-values, in these six large data studies, experimental n=6. In other words, we applied the typical data analysis that is used in biomedical research, when testing for a relevant change.

[0109] When applied to the DNAme clock, these analyses demonstrated that PhenoAge generally predicted that both healthy subjects and patients with various diseases were younger than their chronological age (shifted below the line) and these predictions had on average 12-17 years of error, which for some age groups grew to over 20 years, FIG 14B, 14C. The accuracy of PhenoAge was below statistical significance when it was analyzed with n=6, based on the errors for the same age and when comparing health v. disease. This agrees with batch effects causing a lack of reproducibility, and / or with the normally high variation in the underlying biology of epigenetics. PhenoAge predicted Down Syndrome patients to be biologically olderUC Ref: BK2023-047-3 BFF Ref: BERK-518WO than controls, even though their blood cells are reported to be pathologically juvenile [Webb (2007)]. Parkinson’s disease patients surprisingly had a prediction trend of being younger than their healthy controls at 30–40- year range and not statistically different from their healthy controls at all age intervals after 51 years, FIG 14D.

[0110] The second-generation clocks are trained on, and can thus successfully predict, various biological and health parameters and also age, in both test and out-of-sample datasets. As mentioned above, this property of EN does not in itself require a biological meaning or even a physical connection between the correlated and the measured parameter. To demonstrate this point, we constructed a EN clock that accurately predicted the numbers of people living in US from DNAme array data, and like PhenoAge and PACE, it also predicted people’s age on both the original and out-of-sample datasets, FIG 14E, 14F, 14G. In fact, the US population clock needed only 149 CpGs for predicting age with 2.6-year MAE and Pearson’s r of 0.97, e.g., outperforming many conventional DNAme clocks, FIG 14E. The DNAme predictor of the size of US population also illustrated that the ML EN process tends to overlap age predictions for health and disease, regardless of the training variable, FIG 14G, 14H. The increase in US population over time is a proxy for chronology, but this is also true for all DNAme clocks, be they trained on age, levels of inflammatory cytokines, packs of cigarettes smoked, composite health scores, or any metric which tends to have a progressive change with time. The meaning of the values that are composed together to create a latent variable are not preserved in EN training. EN regression only requires that there are some features which have at least some moderate correlations with time progression.

[0111] As expected from the body of published literature, inflammaging, trisomy and diseases are robustly different from health, and this can be detected with much higher resolution than EN by analyzing the primary, un-adjusted, non-linearized data of the DNAme array. For example, Uniform Manifold Approximation and Projection (UMAP), which preserves the topological structure of the data (with all 450k+ CpGs) while embedding it in multi- dimensional computational space, robustly distinguishes between health and disease in the studied here 450K array datasets, FIG 16 [Becht (2018)].

[0112] Thus, the input DNAme data accurately reflects the expected significant epigenetic differences between healthy and disease populations. However, under the training of a least erroneous linear correlation EN is expected to minimize the weights of CpGs-beta values, which vary among individuals of the same age (or its proxy) due to a disease. Even when a combination of health parameters is used for training, as with the next generation EN clocks, theUC Ref: BK2023-047-3 BFF Ref: BERK-518WO disease-imposed DNAme differences that vary between age groups, might be overlooked in search for the best linear predictors of age or age-related outcomes for the whole population.

[0113] Summarily, if chronology or its proxy is the response variable, and finding the best linear correlation is the task, the relevance of ML predictors is not obvious. Additionally, random experimental variation of the arrays and / or EN training might be misinterpreted as biologically meaningful change in all generations of clock models. II. c) EN predictors are not based on change in DNA methylation with age

[0114] To explore a possible mechanism by which clock features predict age or health scores we studied the assigned weights of CpGs beta values of several published DNAme clock models. The weight assigned by a model to a predictor is a surrogate measure of its importance in the model, as the predictions returned by a model are most sensitive to changes in predictors with the highest weights. Interestingly, we found that EN does not necessarily select CpGs whose methylation either change or correlate strongly with age, and hence are biologically meaningful in such a way, FIG 18A.

[0115] In fact, the ranking of CpGs as per their methylation change with age became better, yet MAE became slightly worse, after all the original clock picked CpGs were removed from the dataset, and as expected, EN selected the next weighted best fit linear predictor set, FIG 18A. The redundancy of CpG sets is known, e.g., EN is “resilient” to specific CpGs, or their combination, and it is the EN process on large data that gives the linear correlation. In support of these finding, the changes of DNAme with age were not the determinant of the ranking of CpGs by the published 1stgeneration clock trained on age or PhenoAge, and PACE trained on health / disease parameters [Horvath (2013); Hannum (2013); Belsky (2022); Levine (2018); Lu (2019)], (FIG 18B). The red dashed lines are the mean coefficient of variation and that the line remains flat with age signifies the irrelevance of age-specific changes in DNAme for the ranking of clock’s CpGs.

[0116] Exploring the mechanisms of CpGs selection further, we next removed not the clock picked CpGs, but the CpGs that were excluded from the model by EN, and then retrained EN on the remaining set, and we did this repeatedly, FIG 18B. Removal of unpicked CpGs significantly changed the picked CpGs from the original training, plateauing at 17-20% of the unused original set after removal of 3-4% of irrelevant CpGs. DeepmAge, PhenoAge and Hannum’s 71 clock CpGs were similarly deselected upon the non-clock CpG removal, FIG 17A, 17B, 17C.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0117] Such behavior does not agree with a hypothesis that biomarkers should not change when non-biomarkers are deleted from a dataset and supports the null hypothesis that picked CpGs are not biomarkers. ML methods that rely on stochastic gradient descent for optimization appear to suffer from the pitfall that removal or absence of certain features changes the hyper surface of the feature space which influences the local optima (i.e., feature sets, weights and hyperparameter values) that the gradient descent can converge upon. This point was experimentally supported by studying the effects of removing the non-significant CpGs on the energy landscape of EN, FIG 17D.

[0118] Summarily, the biomedical significance of clock CpGs is not clear, and DNAme clocks are not directly based on the changes in DNA methylation with age or disease. II. d) It is not absolute beta values, but the noise (SD) of certain CpGs that biomarks biological age

[0119] In the final part of our study, we quantify biological age from the DNAme arrays data, using six independent studies with 450K DNAme arrays that were performed on 1806 samples. We postulated that homeostatically vital CpGs would serve well as the detectors of biological noise, which in our hypothesis is a biomarker of aging and disease. Furthermore, we postulated that such noise detecting CpGs can be characterized and thus identified by a nearly equivalent Mean of their beta values between young and old, but a higher SD about the Mean with age. Our goal was to measure this indicator of biological age, not to predict it.

[0120] The age-imposed increase in the DNAme noise is visible when looking at a scatter plot of the beta values of individual CpGs that are on average invariable in their methylation across the lifespan, as exemplified in FIG 20A (each dot is an individual). To capture this biological phenomenon from the DNAme array data, we first specifically identified the CpGs that remain on average invariable in their methylation through age (Pearson’s (r) 0.02-0.05 of beta values over age) and become deregulated or “noisy” with age (change in absolute distance of their beta values from the Mean, ADM); this yielded 50 CpGs out of the 450K DNAme array, GSE42861 dataset, FIG 20B. Consistent with the hypothesis of biological relevance, all these CpGs turned out to be in vital genes known to be homeostatically regulated, Table 3 and FIGs 23-32, which shows genome loci annotations of the 50 noise detector CpGs. The biological age (Right Y axis) was then quantified based on the Median of the normalized by the young age sums of SDs of these 50 CpGs (Left Y axis). This unadjusted primary data had a good fit with polynomial curve, it displayed age-imposed increase in biological noise with a plateau at ~35-45 years of age, in healthy subjects, and a clear increase in the noise for the young and middle-aged patientsUC Ref: BK2023-047-3 BFF Ref: BERK-518WO with arthritis, FIG 20B (each dot represents the indicated age; the raw data with dots representing individuals per each age is shown in FIG 19A).

[0121] Next, we examined if DNAme noise, which we define as the age-imposed increase in SD of beta values, would self-manifest when we select CpGs solely based on their nearly-same methylation through chronology. We identified all the CpGs with less than 0.1% difference in the Means of their beta values between the young (25-28years) and the old (67+ years) individuals, in the healthy controls’ datasets of the studies on Down Syndrome, Arthritis, Parkinson’s, Multiple Sclerosis, IBS, and Werner Syndrome. A healthy change in CpG noise is biologically expected as the immune system matures [Georgountzou (2017); Kumar (2018)], thus our young range is 25-28 years. These CpGs that are well-regulated through age were also cross-checked by having Pearson’s (r) 0.02-0.05 of beta values over age, e.g., excluding the zeroes of experimental errors; this approach yielded 460 CpGs that had minimal change in their Mean methylation with age. Interestingly and in agreement with our hypothesis, every one of these CpGs, whose methylation on average was age-invariable, manifested an increase in SD – DNAme noise with age, FIG 20C and FIGs 23-32. This unadjusted primary data had a good fit with a polynomial curve of chronological ages v. summed SD (Left Y Axis); and the non- linearity of the pattern compressed certain biological ages, which were mapped onto the Right Y axis by entering the summed SDs into the polynomial equation, FIG 20C.

[0122] Next, we narrowed down the 460-CpG set of the noise detectors to those which are most regulated in the young cohort and those least regulated in the young cohort; once again, identifying the CpGs with these properties that are shared between all studied six healthy controls’ datasets. Out of 450K array data, only 460 CpGs had nearly the same Mean of beta values in young and old cohorts, and out of these only 8 had SD of >0.015 of the absolute Mean and only 5 had SD of >0.015 of a relative (set to 1) Mean of their beta values, e.g., they were the most regulated. Very interestingly, all the 8 and 4 out of the 5 were highly methylated (beta values ~1) and the remaining one CpG had beta value near 0 – unmethylated; none were in- between or partially methylated, even though the samples analyzed for CpG methylation were of a mixed population of peripheral blood cells, FIG 20D, 20F.

[0123] It is possible that these CpGs reflect loci that should be always completely silenced (say, controlling a different cell fate) or open (say, for housekeeping genes), regardless of the cell types of the circulating cells. In contrast, the CpGs that were the least regulated in the young cohort, 17 CpGs with SD of <0.3 of the absolute Mean and 13 CpGs with SD of <0.3 of a relative=1 Mean, had a large range of methylation (spanning the 0-1 range of beta values), FIGUC Ref: BK2023-047-3 BFF Ref: BERK-518WO 20D, 20F. Other least regulated CpGs are shown in FIG 19B. Annotations for the 5, 8, 13 and 17 CpGs – noise detectors are in FIGs 23-32.

[0124] Summing the 8, or the 5, or the 17, or the 13 Means of the SDs for each 3-year age interval between 25 and over 80 years, established that each set detected a natural pattern of human biological aging with a progression of epigenetic noise that is low in young, high in old, and has healthy, low noise plateaus, FIG 20E, 20G (each dot is 49 individuals on average, black arrows point to the 3-year interval before the upward transitions in noise). Interestingly, both the least and most regulated CpGs, which were identified as such by the two approaches in the young samples, became noisier with age, FIG 20E, 20G.64-67 years and 49-52 years demarcated the upward shifts in biological noise (black arrows) for the 8 most- and 17 least- regulated CpGs, respectively, (the SD of the absolute Mean of beta values), FIG 20E.41-44 years and the very early 29-32 years transitions from low to high biological noise (black arrows) were detected for the 5 most- and 13 least-regulated CpGs, respectively (SD of the relative Mean of beta values), FIG 20G.

[0125] We think of this measurement of noise increase with age as a barometer, implying epigenetic pressure on homeostatic gene regulation.

[0126] The 460 CpGs that had constant Means with age and the 50 CpGs found on average invariant yet noisier with age by linear regressions, were independent of the first-gen, PhenoAge and PACE EN CpG sets, FIG 20H. The young-to-adult focused PACE CpG set was also expectedly different from the full age-range focused first generation and PhenoAge models.

[0127] Next, we examined differences between health and disease, using the combined DNAme array data from the six studies by different laboratories. As mentioned, all 460 CpGs of noise barometer self-manifested higher SD of beta values in the old as compared to young, and 293 of these CpGs had at least 20% increase in their SDs with age. Interestingly and consistently with the fundamental paradigm of DNAme noise as a biomarker of aging and disease, summing the SDs of 460, 293 or 5 CpGs yielded similar patterns of healthy biological aging and the disease-imposed shifts, FIG 22A. Moreover, this pattern of biological aging and difference between health and disease were also yielded by SD of a single noise barometer cytosine (one of the most regulated in young), FIG 22A. The overlap between health and disease at certain ages is expected from the overall elevation of diseases with age (not just those studied), and from the possibility of pathologically diminished biological noise in someUC Ref: BK2023-047-3 BFF Ref: BERK-518WO disease stages due to reduced complexity of blood cells (less lymphocytes and greater prevalence of inflammatory macrophages).

[0128] Confirming that selection of noise detecting CpGs is important, sums of SDs of all CpGs from these six combined 450K DNAme array healthy controls’ datasets showed a random scatter through chronological age, FIG 22A, each black dot is an age range. At the same time, in agreement with the fundamental nature of the noise / SD approach, 1 / 4thof 450K DNAme array CpGs (143,448 CpGs) was found to be significantly heteroscedastic and having 20% or more increase of the SDs of their beta values with age. Summing the 143,448 SDs yielded a similar pattern of biological aging and disease-imposed shifts, as when using 460, 293, 5, or 1 cytosine(s), FIG 22A.

[0129] Similar shapes of the healthy curves and good resolution between health and disease in the combined data from six independent studies by different laboratories suggested batch effect resilience of our noise barometer. This was confirmed and extrapolated by comparing the three most numerous healthy control datasets (from the studies of Arthritis, Parkinson’s, and Multiple Sclerosis) in their biological aging curves, FIG 22B and FIGs 23-32. Low batch effect was seen for the three individual healthy datasets, except in the 49-64 years age range, which could be due to a natural increase in DNAme dynamics from cumulatively diverging environments, habits, and diseases (not just those specifically studied), FIG 22B. Based on Venn Diagrams, 460 cytosine noise detectors have the highest overlap with the global 143 thousand subset, followed by the 50 cytosine noise detectors and various EN DNAme clocks, FIG 22C and FIGs 23-32.

[0130] The pattern of healthy biological aging based on the 460 and 5 CpGs was also similar to that yielded by the 50 CpGs, FIGs 20B, 22A. And the 50 CpGs noise barometer demonstrated different curve shapes for different diseases, FIG 21A. At the single levels, the most regulated CpGs – noise detectors had different patterns of their age-specific noise increase, FIG 21B and FIG 22A. The natural biological aging curves of the combined health vs. combined disease that include the HIV dataset, are shown in FIG 21C.

[0131] Since SD is the main parameter in our noise barometer, we also tested if EN clocks might be based on selection of CpGs that have higher SD of their beta values with age and / or disease. The regression analysis demonstrated that this is not the case for either the 1stgeneration EN clock that is trained on age or for PhenoAge or for PACE that are trained on health / disease parameters [Horvath (2013); Hannum (2013); Belsky (2022); Levine (2018); LuUC Ref: BK2023-047-3 BFF Ref: BERK-518WO (2019)]. Namely, there was a lack of correlation between the ranking (weighting) of the clock’s CpGs by the ENs and the changes of the SDs of their beta values with age, FIG 22D.

[0132] These results suggest that biological age can be measured in a batch effect resilient approach through quantification of the SD of the methylation of CpGs that should-be-invariable, yet are age deregulated. The natural curve of biological aging defined by this approach is not linear; it has healthy low-noise plateaus and upward shifts at specific ages. In this method, DNAme noise is the biomarker and a particular set of CpGs is the detector of such biomarker. This method is different from and moreover, orthogonal to, the linear predictors that are based on population statistics. Importantly, regardless of the selection of all the CpGs that should-be- invariable (50, 460) or their subsets most or least regulated in young (1, 5, 8, 13, 17), an increase of biological noise with age was always observed, and there was never a high in young to low in old progression for any cytosine of the noise barometer. II. DISCUSSION

[0133] In this work, we fail-tested the key concepts of ML, EN DNAme clocks, and as such, our conclusions should be broadly applicable to most / all variants of these approaches. Our data confirms that the general phenomenon of change with age, e.g., the distinction between young, middle age and the old, is accurately determined by ML population statistics. With respect to overall implications and unmet need, changes in health are routinely quantified by Student’s t- test, etc. statistics, of well-controlled, independently analyzed from start to finish, epigenetic, proteomic and other molecular and cellular analyses of blood samples. These allow single cell resolution and are confirmed through redundancy of experimental methods, guarding against misinterpretation of random variability or typical dynamics for the changes in health.

[0134] Standard biomedical analyses giving Mean, SD, and p- values from several independent start-to-finish studies do not permit the erroneous assumption that a sample from each person is an experiment in itself when a population was studied by aligning all samples, once. Through such standard analyses, we illustrate how batch effects and dynamics of DNAme might reduce the reproducibility and accuracy of the clocks, making these, at times, below statistical significance. Of note, small, poorly reproducible changes might appear significant, and the healthy group cannot be reliably used as a standard for a disease group, when p-values are calculated to diminish from the sheer number of samples in an n=1 study.

[0135] Regarding fundamental conclusions, it is well known that a 75-year-old person with high levels of C-reactive protein, or diminished albumin, or with obesity, or high cholesterol, is unhealthy and has an increased risk of mortality. EN regression finds correlation, not causality,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO and as we show here, it does not rely on the changes of cytosine methylation or dysregulation with age or disease. Thus, even though DNAme array data is used, the biological meaning of predictions remains uncertain. And as we illustrate with the size of US population, non- biological changes can be successfully predicted by the DNAme array datasets. Summarily, any set of regularly changed numerical values can be linearly predicted through EN training on any large data, biological or not, physically connected to the measured axis or not. If anybody is interested, an expansion of the Universe could be predicted from evolutionary DNAme array data. And a person’s health and age could be predicted with relatively good accuracy from expanding distance between the Galaxies, or anything that changes progressively over time.

[0136] Since the physical meaning of the training parameters is lost in EN process, any assumption that specific health scores, C-reactive protein levels, etc., are being predicted, is ambiguous, because in the model training, a decline in health, reaching mortality and the increase in age all have linear progressions designated as a regular series of numbers.

[0137] In these regards, high throughput sequencing is routine, yet clock models are typically built from less-accurate DNA hybridization data. For reliable scientific or consumer information from EN and other ML approaches, several key controls are important, such as, 1) performing independent start to finish experiments with the same or near-same-time samples, that are analyzed on independently run DNAme arrays, and compared with different samples of the same chronological span, by independently trained EN models; 2) sequencing-normalized bisulfite sequencing controls on whether a clock’s CpGs change in their methylation when a person is predicted to be age / disease accelerated or at 2-4 years age intervals; 3) confirming the chromatin change near clock CpGs by ATAC- Seq, for example; 4) testing whether genes of clock CpGs change their expression (mRNA and protein levels) when a shift in biological age is predicted for a person; 5) confirming that the levels of specific proteins (GrimAge) or health scale are statistically different in 4- 2-years ranges, or at the point of age acceleration.

[0138] Through our investigation, we describe a new quantification of biological age from DNAme, based on the noise of the most-regulated, age invariable CpGs. This approach fits well with the importance of age-specific increase in biological noise [Eling (2019)] and it is the quantification of primary data without numerical adjustments, which improves on ML predictions. The points of increased DNAme noise turned out to be 49-52 and 64-67 years of age, and it would be very interesting to probe global omics at these transitional ages. Our noise barometer distinguishes health from disease and can potentially distinguish one pathology from another completely different pathology. The different time-shapes of different diseases mightUC Ref: BK2023-047-3 BFF Ref: BERK-518WO enable epidemiology of a specific disease in a population, based on the curve of epigenetic noise. The biological significance of noise-detector CpGs is clear and the effects of their deregulation with time and disease are expected to be many and deleterious. Namely, the likely reason for the noise-detectors CpGs to be on average invariable in their methylation with age is that they are in the regulatory regions of genes that are vital at constant levels, such as coding for ribosomal proteins, core transcriptional factors, E3 ligases, regulators of cholesterol and glucose metabolism, MAPK signaling, beta-catenin, Table 3. The less-characterized loci of noise detector CpGs, Table 3 and FIGs 23-32, would be interesting to study for their involvement in biological aging. The paradigm shift is that no specific cytosine or gene is biomarks biological age, but it is the noise and dysregulation, e.g., it is not the expression levels or beta values, but the SD that are the basis of quantifying biological age and risk of disease. Accordingly, summing SD of either 143 thousand or 460, or 5 CpGs. or even SD of one cytosine noise detector, yields overall the same non-linear curve of DNAme aging and distinguishes health from disease. The noticeable plateaus of the DNAme aging emphasize that biological age is progressing non-linearly, in contrast to the chronological age, making accurate linear predictions problematic for certain age ranges. Quantification of the deviation from the healthy range seems to be more relevant and it is enabled by the noise barometer. Table 3. Families and functions of the genes annotated to the 50 CpGs noise biomarkers. Probe Gene Family Effects PRKCG Protein Kinase C -UC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg23801965 GSC2 Goosecoid-like homeodomain Axis formation - Pleiotropic cg23983449 GABARAP GABA R associated proteinsRegulates ligand-gated chloride channels andneurotransmitter signaling

[0139] Reasonably accurate predictability of chronological age from DNA samples positions EN models as potential forensics tools, to inform on the age-ranges of suspects, as was indeed suggested. DNA methylation arrays enable rapid high throughput analysis and generally agree with bisulfate sequencing when the big changes are assayed, such as differences between cancer vs. cancer-free cohorts [Wilhelm-Benartzi (2013); Yong (2016); Sugden (2020)]. Yet, for a law enforcement degree of accuracy, it might be prudent to also perform higher resolution bisulfite sequencing, controlled by non-bisulfite sequencing, for testing the precision of the age predictions.

[0140] Lastly, the assignments of human biological age are typically based on blood samples, and while some patterns might be shared in different organs (for example, fibrosis and inflammation generally increase with age, which is not healthy) the overall biological age is determined by the integrated dynamics of divergent cell-fate gene regulation in the organism, which varies between the tissues, collectively influencing organismal metabolism, proteostasis, organelles, fibrosis, regeneration, etc.

[0141] The above should not be confused with predicting ages of different tissues and species [Lu (2023)]; in a similar fashion, changes in human health, population, and temperature can beUC Ref: BK2023-047-3 BFF Ref: BERK-518WO collectively predicted, or any set of regularly changed numbers, through training EN on DNAme array or any large data. Methods

[0142] The invention provides for methods for determining the age of a subject based on the epigenome of the subject. The subject may be human, mammal, animal, plant, or any multicellular organism. Examples of suitable mammals include but are not limited human, monkey, ape, dog, cat, cow, horse, goat, pig, rabbit, mouse and rat. The epigenome may be deoxyribonucleic acid (DNA) in which the DNA may be subjected to epigenetic modification. The epigenetic modification may be methylation of CpG residues. In one embodiment, the methylation is the covalent attachment of a methyl group at the carbon-5 (C-5) position of cytosine.

[0143] In one embodiment, the method comprises obtaining a biological sample of the subject. Additionally, the method comprises detecting the methylation levels of a set of age-associated epigenetic markers in the epigenome of the subject. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises CpGs described herein. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises CpGs described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of CpGs described herein. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of CpGs described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of CpGs described herein. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of CpGs described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. Further, the method comprises comparing the methylation levels of the set of age-associated epigenetic markers of the subject with the methylation levels of the same set of age-associated epigenetic markers in a control, so as to obtain a value or a range of values for the age of the subject, thereby determining the age of a subject or a biological sample obtained from the subject.

[0144] As used herein, the term "biological sample" can include any cell, tissue, or secretory fluid (whether host or pathogen related) from which DNA can be analyzed. This can include, for example, a human cell sample such as skin. The term "biological sample" is also understood to include fluid samples such as blood, urine, saliva, and cerebrospinal fluid and extends to other biological samples including, for example, mucus from the nasopharyngeal region and the lowerUC Ref: BK2023-047-3 BFF Ref: BERK-518WO respiratory tract (i.e., sputum). Examples of biological samples include, but are not limited to, saliva, sputum, spit, blood, perspiratory fluid, sweat, pus, tear, mucosal excretion, vomit, urine, stool, semen, vaginal fluids, other type of bodily fluid, cell-free samples, cheek swabs, swabs of a different bodily part, homogenous samples, heterogeneous samples, tumor samples, plasma, or serum samples. In an exemplary embodiment, the biological sample is a blood sample. In an exemplary embodiment, the biological sample comprises genomic DNA extracted from saliva of the subject.

[0145] In certain embodiments, the obtaining a biological sample of the subject occurs by the subject providing a biological sample described herein. In certain embodiments, the obtaining a biological sample of the subject occurs by the subject providing a biological sample comprising the saliva of the subject. In certain embodiments, the biological sample is provided by the subject via a container (such as a vial). In certain embodiments, the biological sample comprising the saliva of the subject is provided via a container (such as a vial). Sets of Age-Associated Epigenetic Markers

[0146] In accordance with the practice of the invention, the set of age-associated epigenetic markers of the subject consists of five or more CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; ten or more CpGs, of which at least ten are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; fifteen or more CpGs, of which at least fifteen are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; twenty or more CpGs, of which at least twenty are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; twenty-five or more CpGs, of which at least twenty-five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; thirty or more CpGs, of which at least thirty are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; thirty-five or more CpGs, of which at least thirty-five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; forty or more CpGs, of which at least forty are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; forty-five or more CpGs, of which at least forty-five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; fifty or more CpGs, of which at least fifty are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; fifty-five or more CpGs, of which at least fifty-five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; sixty or more CpGs, of which at least sixty are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; sixty-five or more CpGs, of which at least sixty-five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; or seventy or more CpGs, of which at least seventy are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0147] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to sixty CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to fifty-five CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to fifty CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to forty-five CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age- associated epigenetic markers of the subject consists of five to forty CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to thirty- five CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to thirty CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to twenty-five CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to twenty CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to fifteen CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to ten CpGs, of which at least five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28.

[0148] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of fifteen to sixty CpGs, of which at least fifteen are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of fifteen to fifty-five CpGs, of which at least fifteen are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of fifteen to fifty CpGs, of which at least fifteen are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0149] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of twenty-five to thirty-five CpGs, of which at least twenty-five are described in FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28.

[0150] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of three to seven CpGs, of which at least three are described in FIG 23. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four to six CpGs, of which at least four are described in FIG 23. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five CpGs described in FIG 23. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of five CpGs described in FIG 23. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises five CpGs described in FIG 23. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of cg01614020, cg19028499, cg06626168, cg05184256, and cg24366087. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg01614020, cg19028499, cg06626168, cg05184256, and cg24366087. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises cg01614020, cg19028499, cg06626168, cg05184256, and cg24366087.

[0151] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to thirteen CpGs, of which at least five are described in FIG 24. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of five to ten CpGs, of which at least five are described in FIG 24. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of six to nine CpGs, of which at least six are described in FIG 24. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of seven to nine CpGs, of which at least seven are described in FIG 24. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of eight CpGs described in FIG 24. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of eight CpGs described in FIG 24. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises eight CpGs described in FIG 24. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of cg09521490, cg20398123, cg19028499, cg06626168, cg23183348, cg05184256, cg24366087, and cg24979233. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg09521490, cg20398123, cg19028499, cg06626168, cg23183348, cg05184256,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg24366087, and cg24979233. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises cg09521490, cg20398123, cg19028499, cg06626168, cg23183348, cg05184256, cg24366087, and cg24979233.

[0152] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of ten to twenty CpGs, of which at least ten are described in FIG 25. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of ten to fifteen CpGs, of which at least ten are described in FIG 25. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of eleven to fourteen CpGs, of which at least eleven are described in FIG 25. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of twelve to fourteen CpGs, of which at least twelve are described in FIG 25. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of thirteen CpGs described in FIG 25. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of thirteen CpGs described in FIG 25. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises thirteen CpGs described in FIG 25. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of cg12088417, cg22796025, cg21735741, cg11222703, cg20823529, cg13322072, cg14485877, cg22757362, cg19528830, cg19077165, cg04613452, cg14752227, and cg03873518. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg12088417, cg22796025, cg21735741, cg11222703, cg20823529, cg13322072, cg14485877, cg22757362, cg19528830, cg19077165, cg04613452, cg14752227, and cg03873518. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises cg12088417, cg22796025, cg21735741, cg11222703, cg20823529, cg13322072, cg14485877, cg22757362, cg19528830, cg19077165, cg04613452, cg14752227, and cg03873518.

[0153] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of fifteen to twenty CpGs, of which at least fifteen are described in FIG 26. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of fifteen to nineteen CpGs, of which at least fifteen are described in FIG 26. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of sixteen to eighteen CpGs, of which at least sixteen are described in FIG 26. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of seventeen CpGs described in FIG 26. In an exemplary embodiment, the set of age-associated epigenetic markers of theUC Ref: BK2023-047-3 BFF Ref: BERK-518WO subject consists essentially of seventeen CpGs described in FIG 26. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises seventeen CpGs described in FIG 26. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of cg09606941, cg17942639, cg16239482, cg01614020, cg19456540, cg13727849, cg22996004, cg05082965, cg11222703, cg26570901, cg19528830, cg08137948, cg25311271, cg25600933, cg22757362, cg19001789, and cg25997306. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg09606941, cg17942639, cg16239482, cg01614020, cg19456540, cg13727849, cg22996004, cg05082965, cg11222703, cg26570901, cg19528830, cg08137948, cg25311271, cg25600933, cg22757362, cg19001789, and cg25997306. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises cg09606941, cg17942639, cg16239482, cg01614020, cg19456540, cg13727849, cg22996004, cg05082965, cg11222703, cg26570901, cg19528830, cg08137948, cg25311271, cg25600933, cg22757362, cg19001789, and cg25997306.

[0154] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of forty-five to sixty CpGs, of which at least forty-five are described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of forty-five to fifty-five CpGs, of which at least forty-five are described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of forty-six to fifty-four CpGs, of which at least forty-six are described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of forty-seven to fifty-three CpGs, of which at least forty-seven are described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of forty-eight to fifty-two CpGs, of which at least forty-eight are described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of forty-nine to fifty-one CpGs, of which at least forty-nine are described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of fifty CpGs described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of fifty CpGs described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises fifty CpGs described in FIG 27. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of cg00479463, cg00868523, cg01681847, cg01767885, cg01869186, cg01900413, cg02247160, cg02449103, cg04153991, cg04495670, cg05207943, cg05557449, cg06362282, cg06494091, cg06899313, cg06918474, cg07165260,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg07710266, cg08499046, cg12518535, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14400541, cg14657277, cg14811319, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20037507, cg20261915, cg20283716, cg20445245, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg24201362, cg25007283, cg25734842, cg27219748, cg27317439, cg27621745, and cg27642618. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg00479463, cg00868523, cg01681847, cg01767885, cg01869186, cg01900413, cg02247160, cg02449103, cg04153991, cg04495670, cg05207943, cg05557449, cg06362282, cg06494091, cg06899313, cg06918474, cg07165260, cg07710266, cg08499046, cg12518535, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14400541, cg14657277, cg14811319, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20037507, cg20261915, cg20283716, cg20445245, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg24201362, cg25007283, cg25734842, cg27219748, cg27317439, cg27621745, and cg27642618. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises cg00479463, cg00868523, cg01681847, cg01767885, cg01869186, cg01900413, cg02247160, cg02449103, cg04153991, cg04495670, cg05207943, cg05557449, cg06362282, cg06494091, cg06899313, cg06918474, cg07165260, cg07710266, cg08499046, cg12518535, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14400541, cg14657277, cg14811319, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20037507, cg20261915, cg20283716, cg20445245, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg24201362, cg25007283, cg25734842, cg27219748, cg27317439, cg27621745, and cg27642618.

[0155] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred fifty to four hundred seventy-five CpGs, of which at least four hundred fifty are described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred fifty-two to four hundred seventy-three CpGs, of which at least four hundred fifty-two are described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred fifty-four to four hundred seventy CpGs, of which at least four hundred fifty-four are described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred fifty-four to four hundred sixty-eight CpGs, of which at least four hundred fifty-four are described in FIG 28. In an exemplary embodiment, the set of age- associated epigenetic markers of the subject consists of four hundred fifty-four to four hundred sixty-six CpGs, of which at least four hundred fifty-four are described in FIG 28. In anUC Ref: BK2023-047-3 BFF Ref: BERK-518WO exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred fifty-six to four hundred sixty-four CpGs, of which at least four hundred fifty-six are described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred fifty-eight to four hundred sixty-two CpGs, of which at least four hundred fifty-eight are described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred fifty-nine to four hundred sixty-one CpGs, of which at least four hundred fifty-nine are described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of four hundred sixty CpGs described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of four hundred sixty CpGs described in FIG 28. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises four hundred sixty CpGs described in FIG 28.

[0156] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of thirty to fifty CpGs, of which at least thirty are selected from the group consisting of cg01767885, cg01869186, cg01900413, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg08499046, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg23417875, cg23983449, cg24027320, cg24112454, cg27621745, cg27642618, cg00479463, cg01681847, cg06899313, cg07710266, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, and cg27219748. In an exemplary embodiment, the set of age- associated epigenetic markers of the subject consists of cg01767885, cg01869186, cg01900413, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg08499046, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg23417875, cg23983449, cg24027320, cg24112454, cg27621745, cg27642618, cg00479463, cg01681847, cg06899313, cg07710266, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, and cg27219748. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg01767885, cg01869186, cg01900413, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg08499046, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg23417875, cg23983449, cg24027320, cg24112454, cg27621745, cg27642618, cg00479463, cg01681847, cg06899313, cg07710266, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, and cg27219748. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprisesUC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg01767885, cg01869186, cg01900413, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg08499046, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg23417875, cg23983449, cg24027320, cg24112454, cg27621745, cg27642618, cg00479463, cg01681847, cg06899313, cg07710266, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, and cg27219748.

[0157] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of forty to fifty-five CpGs, of which at least thirty are selected from the group consisting of: cg00868523, cg01767885, cg01869186, cg01900413, cg02449103, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of cg00868523, cg01767885, cg01869186, cg01900413, cg02449103, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg00868523, cg01767885, cg01869186, cg01900413, cg02449103, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises cg00868523, cg01767885, cg01869186, cg01900413, cg02449103,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439.

[0158] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of twenty-five to thirty-five CpGs, of which at least twenty-five are selected from the group consisting of: cg01974375, cg06784991, cg08128734, cg09809672, cg10501210, cg16054275, cg06639320, cg22454769, cg24079702, cg07553761, cg08262002, cg06493994, cg16867657, cg21572722, cg22736354, cg24724428, cg03473532, cg08097417, cg14361627, cg16419235, cg26290632, cg01812894, cg07080372, cg08160331, cg01820374, cg11649376, cg17436656, cg04875128, cg08090640, cg19283806, and cg07547549. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of cg01974375, cg06784991, cg08128734, cg09809672, cg10501210, cg16054275, cg06639320, cg22454769, cg24079702, cg07553761, cg08262002, cg06493994, cg16867657, cg21572722, cg22736354, cg24724428, cg03473532, cg08097417, cg14361627, cg16419235, cg26290632, cg01812894, cg07080372, cg08160331, cg01820374, cg11649376, cg17436656, cg04875128, cg08090640, cg19283806, and cg07547549. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists essentially of cg01974375, cg06784991, cg08128734, cg09809672, cg10501210, cg16054275, cg06639320, cg22454769, cg24079702, cg07553761, cg08262002, cg06493994, cg16867657, cg21572722, cg22736354, cg24724428, cg03473532, cg08097417, cg14361627, cg16419235, cg26290632, cg01812894, cg07080372, cg08160331, cg01820374, cg11649376, cg17436656, cg04875128, cg08090640, cg19283806, and cg07547549. In an exemplary embodiment, the set of age-associated epigenetic markers of the subject comprises cg01974375, cg06784991, cg08128734, cg09809672, cg10501210, cg16054275, cg06639320, cg22454769, cg24079702, cg07553761, cg08262002, cg06493994, cg16867657, cg21572722, cg22736354, cg24724428, cg03473532, cg08097417, cg14361627, cg16419235, cg26290632, cg01812894, cg07080372, cg08160331, cg01820374, cg11649376, cg17436656, cg04875128, cg08090640, cg19283806, and cg07547549.

[0159] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of CpGs which are on average age-invariant but are age-deregulated. In an exemplaryUC Ref: BK2023-047-3 BFF Ref: BERK-518WO embodiment, such age-invariant / age-deregulated CpGs have an individual correlation with age of 0.02 < r < 0.05. In an exemplary embodiment, such age-invariant / age-deregulated CpGs have an absolute difference from the mean (ADM) of 0.2 < r. In an exemplary embodiment, CpGs which are on average age-invariant but are age-deregulated are determined by: performing regression analysis (such as described herein) on the CpG correlating with age; calculating the mean beta value for each CpG; calculating the absolute difference from the mean (ADM) for each CpG; performing regression analysis on the ADM of the CpG correlating with age. In an exemplary embodiment, the on average age-invariant but age-deregulated CpGs are further determined by applying a normalizing factor as described herein the ADMs. In an exemplary embodiment, the normalizing factor is created by summing the ADMs and taking the average of the sums. In an exemplary embodiment, a noise score σ can be calculated by summing the ADMs and then normalizing by ζ (Equation 9). In an exemplary embodiment, Y axis biological age numbers can be calculated through the polynomial fit for the median value of normalized sum for each age in the healthy control dataset. In an exemplary embodiment, the noise score can be found by setting the noise score to be Y and find root, or the X values, from the polynomial equation as described herein.

[0160] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of CpGs which are on average age-invariant. In an exemplary embodiment, such age- invariant CpGs have >0.1% difference in the mean beta value from a group of healthy subjects having a chronological age of 25–27 years and in the mean beta value from a group of healthy subjects having a chronological age of more than 67 years. In an exemplary embodiment, such age-invariant CpGs have Pearson’s (r) of from 0.02 to 0.05 of beta values over age. In an exemplary embodiment, such age-invariant CpGs have a standard deviation of about 0.015 of the absolute and relative means in a group, such as healthy subjects having a chronological age of 25–27 years. In an exemplary embodiment, such age-invariant CpGs have a standard deviation of from 0.0005 to 0.05 of the absolute and relative means in a group, such as healthy subjects having a chronological age of 25–27 years. In an exemplary embodiment, such age- invariant CpGs have a standard deviation of about 0.3 of the absolute and relative means in a group, such as healthy subjects having a chronological age of 25–27 years. In an exemplary embodiment, such age-invariant CpGs have a standard deviation of from 0.1 to 0.5 of the absolute and relative means in a group, such as healthy subjects having a chronological age of 25–27 years.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0161] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of CpGs which have significant unequal variance. In an exemplary embodiment, such CpGs are determined by performing a heteroskedasticity test (White’s test) for a CpG in a group of healthy subjects having a chronological age of 25–27 years and in the same CpG in a group of healthy subjects having a chronological age of more than 67 years and finding unequal variance with p < 0.05.

[0162] In an exemplary embodiment, the set of age-associated epigenetic markers of the subject consists of CpGs in which the difference is greater than 20% between a standard deviation for a CpG in a group of healthy subjects having a chronological age of more than 67 years and a standard deviation for the same CpG in a group of healthy subjects having a chronological age of 25–27 years (“SD CpGs”). In an exemplary embodiment, the set of age- associated epigenetic markers of the subject consists of CpGs which are SD CpGs and CpGs which have significant unequal variance. Additional Details for the Methods

[0163] In one embodiment, the method comprises use of a statistical method to compare the methylation levels of the set of age-associated epigenetic markers of the subject with the methylation levels of the same set of age-associated epigenetic markers in a control. Examples of suitable statistical methods include but are not limited to multivariate regression method, linear regression analysis, tabular method or graphical method. In an exemplary embodiment, the statistical method comprises ordinary least squares regression. In an exemplary embodiment, the statistical method comprises modified forward stepwise selection. In a preferred embodiment, the statistical method comprises modified forward stepwise selection ordinary least squares.

[0164] In an exemplary embodiment, the biological sample of the subject comprises DNA. In an exemplary embodiment, the biological sample of the subject comprises genomic DNA or nuclear DNA. In accordance with the practice of the invention, detecting the methylation levels may comprise isolating genomic DNA or nuclear DNA from the biological sample, reacting the isolated genomic DNA or nuclear DNA with one or more probe / agent (e.g., a chemical probe / agent) which differentially reacts with unmodified cytosine so that the cytosine is converted to uracil. The step may also comprise determining or analyzing the methylation levels at the cytosine position (also referred to herein as the C position) of a CpG in the isolated genomic DNA or nuclear DNA of the biological sample by detecting the presence of a cytosine or uracil. The presence of cytosine or uracil indicates the presence of a 5-methylcytosine orUC Ref: BK2023-047-3 BFF Ref: BERK-518WO unmodified cytosine, respectively, in the original CpG. Alternatively, resistance to cleavage by a restriction enzyme may indicate the presence of 5-methylcytosine at the original CpG. Sensitivity to cleavage by the restriction enzyme may indicate presence of unmodified cytosine at the original CpG. Further, the step may further comprise determining the proportion of 5- methylcytosine or unmodified cytosine initially present at each CpG in the set of CpGs described herein from the biological sample of the subject; or alternatively, determining the ratio of 5-methylcytosine to unmodified cytosine or the ratio of unmodified cytosine to 5-methyl- cytosine cytosine initially present at each CpG in the set of CpGs described herein from the biological sample of the subject based on characterizing outcome of probing the isolated genomic DNA or nuclear DNA.

[0165] In accordance with the practice of the invention, detecting the methylation levels may comprise isolating genomic DNA or nuclear DNA from the biological sample, incubating the isolated genomic DNA or nuclear DNA with one or more restriction enzyme which recognizes a specific DNA sequence, is affected by a CpG, within or adjacent to the restriction enzyme recognition or cleavage site, and differentially cleaves the DNA based on the presence or absence of a methyl group at C-5 position of cytosine of the CpG. The step may also comprise determining or analyzing the methylation levels at the C position of a CpG in the isolated genomic DNA or nuclear DNA of the biological sample by its resistance to cleavage at a potential cleavage site by the restriction enzyme indicating presence of 5-methylcytosine at the original CpG within or adjacent to the restriction enzyme recognition or cleavage site. Sensitivity to cleavage by the restriction enzyme may indicate presence of unmodified cytosine. Further, the step may further comprise determining the proportion of 5-methylcytosine or unmodified cytosine initially present at each CpG in the set of CpGs described herein from the biological sample of the subject; or alternatively, determining the ratio of 5-methylcytosine to unmodified cytosine or the ratio of unmodified cytosine to 5-methylcytosine initially present at each CpG in the set of CpGs described herein from the biological sample of the subject.

[0166] In accordance with the practice of the invention, the methods of the invention may be automated.

[0167] In accordance with the practice of the invention, the biological sample may be any of blood, lymphocyte, monocyte, neutrophil, basophil, eosinophil, myeloid lineage cell, lymphoid lineage cell, bone marrow, saliva, buccal swab, nasal swab, urine, fecal material, hair, breast tissue, ovarian tissue, uterine tissue, cervical tissue, prostate tissue, testicular tissue, brain tissue, neuronal cell, astrocyte, liver tissue, kidney, thyroid tissue, stomach tissue, intestine tissue,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO pancreatic tissue, vascular tissue, skin, lung tissue, bone tissue, cartilage, ligament, tendon, fat cells, muscle cells, neurons, astrocytes, cultured cells with different passage number, cancer / tumor cells, cancer / tumor tissue, normal cells, normal tissue, any tissue(s) or cell(s) with a nucleus containing genetic material, or genetic material in the form of DNA of a known or unknown subject.

[0168] The tumor or cancer cells may be derived from blood, lymph node, liver, brain, esophagus, trachea, stomach, intestine, pancreas, throat, tongue, bone, ovary, uterus, cervix, peritoneum, prostate, testes, breast, kidney, lung, or skin. The biological sample with tumor or cancer cells may be determined to have an older predicted age of at least about 30% or 40% more than the biological sample without tumor or cancer cells.

[0169] In an embodiment, a majority of the CpGs in the set of age-associated epigenetic markers of the subject may determine an older age for a biological sample with tumor than biological sample of the same type without tumor. Similarly, pre-cancerous lesions may show an older biological age than a normal tissue type without such a lesion.

[0170] In an embodiment, one or more probes (e.g., chemical probes) may differentially react with an unmodified cytosine and 5-methyl-modified cytosine. The probe may be chosen from a set comprising a sodium bisulfite, sodium metabisulfite, and / or bisulfite salts.

[0171] In another embodiment, the outcome of reacting the isolated genomic DNA or nuclear DNA with one or more probes may be the deamination of unmodified cytosine to uracil and unaltered 5-methylcytosine. Characterizing the outcome of probing (or reacting) the isolated genomic DNA or nuclear DNA with one or more probe(s) or analyzing the methylation levels may involve DNA amplification and nucleic acid sequence determination and detecting for the presence of either cytosine or thymine at the C position of the CpG within the set of age- associated epigenetic markers. Further, DNA amplification may be followed by phage RNA polymerase transcription, RNase cleavage and matrix-assisted laser desorption ionization time- of-flight mass spectrometry (MALDI-TOF MS) of RNase cleavage products.

[0172] The nucleic acid sequence determination may involve one or more of the following procedures: nucleic acid fragmentation, restriction enzyme digestion, nucleic acid hybridization, primer extension, pyrosequencing, single nucleotide extension, single nucleotide extension with biotin-labelled ddNTP, single nucleotide extension with 2,4-dinitrophenol (DNP)-labelled ddNTP, radioactive isotope labeling, nonradioactive label incorporation, fluorescent label incorporation, biotin incorporation, antigen-antibody complex formation, antibody detection,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO colorimetric detection, fluorescence detection, detection with fluorescent dye-labelled antibody, detection with labeled avidin or streptavidin, bead analysis or detection method, signal amplification, polymerase chain reaction, DNA amplification with thermostable DNA polymerase, phi-29 DNA polymerase DNA amplification, RNA production, in vitro transcription, phage RNA polymerase transcription, T7 RNA polymerase transcription, SP6 RNA polymerase transcription, T3 RNA polymerase transcription, RNAse digestion, RNase A digestion, DNA cloning, bacterial transformation, gel electrophoresis, mass spectroscopy, MALDI-TOF mass spectroscopy, microarray analysis, fluorescence scanner analysis, automated digital image capture, automated digital image analysis, ratiometric analysis, and Infinium® HumanMethylation450 BeadChip analysis.

[0173] In another embodiment, the proportion of unmodified cytosine initially present at each CpG in the set of age-associated epigenetic markers of the subject may be a fraction or percent of the thymine in that CpG in the set of age-associated epigenetic markers of the subject.

[0174] The ratio of 5-methylcytosine to unmodified cytosine initially at each CpG in the set of age-associated epigenetic markers of the subject may be the ratio of cytosine to thymine in that CpG after exposure to one or more probe and following analysis of products of nucleic acid amplification.

[0175] In yet another embodiment, the ratio of unmodified cytosine to 5-methylcytosine initially present at each CpG in the set of age-associated epigenetic markers of the subject may be the ratio of thymine to cytosine in that CpG after exposure to one or more probe and following analysis of products of nucleic acid amplification.

[0176] In an embodiment, one or more restriction enzyme probe which recognizes a specific DNA sequence, is affected by a CpG, within or adjacent to the restriction enzyme recognition or cleavage site, and differentially cleaves the DNA based on the presence or absence of a methyl group at C-5 position of cytosine of the dinucleotide may be selected. Examples of such restriction enzymes include but are not limited to AatII, Acc65I, AccI, AciI, AclI, AfeI, AgeI, AhdI, AleI, ApaI, ApaLI, AscI, AsiSI, AvaI, AvaII, BaeI, BanI, BbrPI, BbvCI, BceAI, BcgI, BcoDI, BfuAI, BfuCI, BglI, BmgBI, BsaAI, BsaBI, BsaHI, BsaI, BseYI, BsiEI, BsiWI, BslI, BsmAI, BsmBI, BsmFI, BspDI, BsrBI, BsrFI, BssHII, BssKI, BstAPI, BstBI, BstUI, BstZ17I, Cac8I, ClaI, DpnI, DraIII, DrdI, EaeI, EagI, EarI, EciI, Eco53kI, EcoRI, EcoRV, FauI, Fnu4HI, FokI, FseI, FspI, HaeII, HaeIII, HgaI, HhaI, HincII, HinfI, HinP1I, HpaI, HpaII, Hpy166II, Hpy188III, Hpy99I, HpyAV, HpyCH4IV, HpyCH4V, KasI, MboI, MluI, MmeI, MspAlI, MwoI, NaeI, NarI, NciI, NgoMIV, NheI, NlaIV, NotI, NruI, Nt. BbvCI, Nt. BsmAI, Nt. CviPII,UC Ref: BK2023-047-3 BFF Ref: BERK-518WO PaeR71, PhoI, PleI, PluTI, PmeI, PmlI, PshAI, PspOMI, PspXI, PvuI, RsaI, RsrII, SacII, SalI, Sau3AI, Sau96I, ScrFI, SfaNI, SfiI, SfoI, SgrAI, SmaI, SnaBI, StyD4I, TfiI, TliI, TseI, TspMI, XhoI, XmaI, and ZraI.

[0177] In an embodiment, the outcome of reacting the isolated genomic DNA or nuclear DNA with one or more restriction enzyme probe may be the production of a double-stranded DNA break at a restriction enzyme cleavage site when cytosine at a CpG is not modified or no double- stranded DNA break at a restriction enzyme cleavage site when cytosine at a CpG is modified at its C-5 position with a methyl group.

[0178] In another embodiment, analyzing the methylation levels may comprise DNA amplification and analysis of the set of age-associated epigenetic markers for specific DNA end(s) or fragment(s) due to cleavage by the restriction enzyme(s) and for intact restriction enzyme cleavage site associated at a particular age-associated epigenetic marker within the set.

[0179] In another embodiment, the proportion of 5-methylcytosine initially present at each age- associated epigenetic marker may be a fraction or percent of the age-associated epigenetic marker with an intact restriction enzyme cleavage site due to resistance to cleavage by the restriction enzyme.

[0180] In yet another embodiment, the proportion of unmodified cytosine initially present at each age-associated epigenetic marker may be a fraction or percent of the age-associated epigenetic marker cleaved by the restriction enzyme.

[0181] In another embodiment, the ratio of 5-methylcytosine to unmodified cytosine initially present for each CpG in the set of age-associated epigenetic markers of the subject may be the ratio of number or concentration of intact restriction enzyme cleavage sites to the number or concentration of double-stranded DNA breaks produced by the restriction enzyme for each CpG in the set of age-associated epigenetic markers of the subject.

[0182] In another embodiment, the ratio of unmodified cytosine to 5-methylcytosine initially present for each CpG in the set of age-associated epigenetic markers of the subject may be the ratio of number or concentration of double-stranded DNA breaks produced by the restriction enzyme to number or concentration of intact restriction enzyme cleavage sites resistant to cleavage by the restriction enzyme due to presence of 5-methylcytosine for each CpG in the set of age-associated epigenetic markers of the subject.

[0183] In one embodiment, detecting the methylation levels comprises isolating genomic DNA or nuclear DNA from the sample. Additionally, the step involves probing the isolated genomicUC Ref: BK2023-047-3 BFF Ref: BERK-518WO DNA or nuclear DNA with one or more probes which differentially reacts with unmodified and 5-methylmodified cytosine and amplifying the DNA. The step also involves digesting the amplified DNA with one or more restriction enzyme that recognizes a restriction enzyme site that contains a CpG but fails to digest the restriction enzyme site mutated to TpG dinucleotide from a CpG. Further, the step involves determining the proportion of 5-methylcytosine or unmodified cytosine initially present for each CpG in the set of age-associated epigenetic markers of the subject based on the fraction or percentage of restriction enzyme sites sensitive or resistant to digestion.

[0184] Alternatively, the methylation levels may involve determining the ratio of 5- methylcytosine to unmodified cytosine initially present for each CpG in the set of age-associated epigenetic markers of the subject based on the ratio of number or concentration of sensitive restriction enzyme sites to number or concentration of resistant restriction enzyme sites to digestion. The methylation levels may also involve determining the ratio of unmodified cytosine to 5-methylcytosine initially present for each CpG in the set of age-associated epigenetic markers of the subject based on the ratio of number or concentration of resistant restriction enzyme sites to number or concentration of sensitive restriction enzyme sites to digestion.

[0185] In another embodiment, detecting the methylation levels of the set of age-associated epigenetic markers in the epigenome of the subject described in the section entitled “Age- Associated Epigenetic Markers” may comprise isolating genomic DNA or nuclear DNA and fragmenting the genomic DNA or nuclear DNA. Additionally, the step involves exposing the fragmented DNA to a 5-methylcytosine-binding protein and separating 5-methylcytosine- binding protein-bound DNA fragments from 5-methylcytosine-binding protein-free DNA fragments. The step further involves determining for each CpG in the set of age-associated epigenetic markers in the epigenome of the subject described in the section entitled “Age- Associated Epigenetic Markers”, the proportion of 5-methylcytosine-containing DNA fragments or unmodified cytosine-containing DNA fragments by determining the fraction or percent of 5- methylcytosine-binding protein bound or free DNA fragments, respectively, for that CpG in the set of age-associated epigenetic markers.

[0186] Alternatively, the methylation levels may involve determining for each CpG in the set of age-associated epigenetic markers in the epigenome of the subject described in the section entitled “Age-Associated Epigenetic Markers”, the ratio of 5-methylcytosine-containing DNA fragments to unmodified cytosine-containing DNA fragments by determining the ratio ofUC Ref: BK2023-047-3 BFF Ref: BERK-518WO number or concentration of 5-methylcytosine-binding protein-bound DNA fragments to the number or concentration of 5-methylcytosine-binding protein-free DNA fragments.

[0187] The methylation levels step may also involve determining for each CpG in the set of age- associated epigenetic markers in the epigenome of the subject described in the section entitled “Age-Associated Epigenetic Markers”, the ratio of unmodified cytosine-containing DNA fragments to 5-methylcytosine-containing DNA fragments by determining the ratio of number or concentration of 5-methylcytosine-binding protein-free DNA fragments to the number or concentration of 5-methylcytosine-binding protein-bound DNA fragments for that CpG in the set of age-associated epigenetic markers.

[0188] In one embodiment, the 5-methylcytosine-binding protein may be an antibody for 5- methylcytosine, MeCP2, MBD2, MBD2 / MBD3Ll complex, core MBD domain of MBD2, or poly-MBD protein, a naturally occurring 5-methylcytosine binding protein, genetically engineered 5-methylcytosine binding protein, or derivative or fragment thereof.

[0189] In one embodiment, separating 5-methylcytosine-binding protein-bound DNA fragments from 5-methylcytosine-binding protein-free DNA fragments may include immunoprecipitation, immunocapture, solid phase chromatography, liquid chromatography, and / or gel electrophoresis.

[0190] In one embodiment, the invention provides methods for diagnosing the presence of tumor in a subject. The method comprises obtaining biological sample suspected to contain tumor and a second biological sample of the same type but known not to contain tumor. Additionally, the method comprises determining the age of each biological sample by the method of the invention.

[0191] The method further comprises comparing the ages determined for the two samples, such that a biological sample with tumor will have an older determined age than biological sample without tumor.

[0192] In an embodiment, the invention provides a forensic diagnosis of human actual age from a tissue from a human by determining age of a subject based on the epigenome of the subject by the method of the invention.

[0193] In another embodiment, the invention provides methods for health assessment of a subject by determining age of a subject based on the epigenome of the subject by the method of the invention.

[0194] In yet another embodiment, the invention provides methods for screening whether an agent of interest can retard or accelerate aging process. The method comprises obtaining aUC Ref: BK2023-047-3 BFF Ref: BERK-518WO biological sample from a living organism, and optionally, culturing cells, tissue, or organ derived from a living organism and determining age of the organism, following the method of the invention, using an organism appropriate set of age-associated epigenetic markers such that the age-associated epigenetic markers for a human subject may need to be substituted with age- associated epigenetic markers for the organism being examined. Additionally, the method comprises exposing the living organism or cultured living cells, tissue, or organ from the living organism to an agent of interest in a single dose, multiple doses, or continuous doses and obtaining a biological sample from the living organism or the cultured living cells, tissue, or organ. The method further comprises determining age of the organism from the biological sample using an organism appropriate set of age-associated epigenetic markers such that the age-associated epigenetic markers for a human subject may need to be substituted with age- associated epigenetic markers for the organism being examined. The method also comprises performing the same steps on another individual from the same organism or a duplicate cultured living cells, tissue, or organ from the same individual or organism but not treated with any agent of interest or treated with a placebo and comparing, for biological sample of the same determined age, the determined age of the agent-of-interest-treated organism / individual or cultured cells, tissue, or organ with the determined age of the untreated or placebo-treated organism / individual or cultured cells, tissue, or organ, such that a lower value or range of values for the agent-of-interest-treated organism / individual or cultured cells, tissue, or organ indicates that an agent of interest can retard an aging process whereas a higher value or range of values indicates an agent of interest can accelerate aging process. The agent of interest may be an antioxidant, reducing agent, DNA damaging agent, vitamin, dietary supplement, food, food additive, food coloring, salt, vegetable, vegetable extract, fruit, fruit extract, flower, flower extract, fragrance, seed, seed extract, herb, herb extract, plant extract, fiber, fat, fatty acid, oil, sugar, artificial sweetener, probiotics, alcohol, wine, fungus, mold, cream, lotion, powder, makeup, sun blocker, gas, pollutant, smoke, environmental pollutant, paint, solvent, organic solvent, plastic, plasticizers, bisphenol, phenolic compounds, tobacco, inhalant, drug, biologic, hormone, endocrine disruptor, environmental estrogen, hormone antagonist, hormone agonist, caffeine, phytoestrogen, metal, enzyme, chelator, yogurt, sulfur compound, physical barrier, electromagnetic barrier, and radiation barrier.

[0195] In an embodiment, the organism may be yeast, fruit fly, fish, worm, insect, zebra fish, nematode, plant, or mammal. Mammal includes, but is not limited to, human, murine, simian, feline, canine, equine, bovine, porcine, ovine, caprine, rabbit, mammalian farm animal, mammalian sport animal, and mammalian pet.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0196] In one embodiment, the invention provides methods for identifying type of tissue for a biological sample from a subject with a known chronological age. The method comprises ascertaining the chronological age of a subject and determining the age of the subject from the biological sample by the method of the invention. Additionally, the method comprises comparing to a reference standard relating the determined age for various types of tissue to chronological age and determining which value closely matches the age in the reference standard for various types of tissue. Further, the method comprises assigning the type of tissue for the biological sample based on the closest match.

[0197] The invention further provides methods for determining age of a subject based on age- associated epigenetic modification affecting gene expression. The method comprises obtaining a biological sample of the subject and determining the expression of one or more gene(s) associated with a set of age-associated epigenetic markers described herein. Additionally, the method comprises comparing the expression of one or more gene(s) associated with a set of age- associated epigenetic markers described herein whose expression changes with age with the expression of the same gene(s) from a control. The method further comprises obtaining a value or range of values for the determined age of the subject. Comparing the expression of one or more gene(s) associated with a set of age-associated epigenetic markers described herein whose expression changes with age with the expression of the same gene(s) from an age-correlated reference population may comprise any statistical method, multivariate regression method, linear regression analysis, tabular method, or graphical method used to determine the age of a subject based on expression of gene(s) associated with age-associated epigenetic marker(s) whose expression changes with age. In one embodiment, the statistical method may be a multivariate regression algorithm or linear regression algorithm.

[0198] In another embodiment, one or more gene(s) associated with age-associated epigenetic marker(s) whose expression changes with age may comprise one or more of the genes listed in a Table herein.

[0199] In another embodiment, the gene expression may be a transcription or translation. In another embodiment, the transcription results in the production of RNA transcripts and translation results in the production of proteins. In accordance with the practice of the invention, the invention provides a method of screening a tissue sample from a subject in order to determine the age of the tissue sample based on the epigenome of the subject by the method the invention.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0200] In one embodiment, the tissue sample may be exposed to at least one test agent in a high- throughput screening assay. In another embodiment, said process may be used for any one of diagnosis and / or high-throughput screening. The invention also provides methods for determining age of a tissue or organ of a subject based on the epigenome of the tissue or organ of the subject. The method comprises obtaining a biological sample of a tissue or organ from the subject and detecting the methylation levels of a set of age-associated epigenetic markers in the epigenome of the subject described herein. The method further comprises comparing the methylation levels of a set of age-associated epigenetic markers of the subject with the methylation levels of the same markers from a control so as to obtain a value or a range of values for the predicted age of the tissue or organ. The methylation levels of the same markers from a control may be determined on a same or a different type of tissue or organ. The methylation levels of the same markers from a control may be determined on blood or fractionated blood.

[0201] In an embodiment, the methods of the invention provides for determining differential aging rates of tissues or organs of a subject. The method comprises obtaining biological samples from different tissue(s) or organ(s) from the subject and determining the age of the tissue or organ using the methods of the invention. The method further comprises comparing the ages where a difference in the determined ages indicates a difference in the aging rate of the tissue(s) or organ(s) of the subject. Systems and Computer Implemented Methods

[0202] Aspects of the invention additionally include systems configured to perform the methods described herein, including determining the age (for example biological age) of a subject based on the epigenome of the subject, as well as including the method comprising use of a statistical method to compare the methylation levels of the set of age-associated epigenetic markers of the subject with the methylation levels of the same set of age-associated epigenetic markers in a control, as well as including a method for ascertaining the health of a subject relative to the subject’s chronological age.

[0203] In some instances the systems include one or more computers for complete automation or partial automation of the methods described herein. In an exemplary embodiment, the determining methylation levels of a set of age-associated epigenetic markers is automated and conducted by a computer. In an exemplary embodiment, the comparing the methylation levels of the set of age-associated epigenetic markers of the subject with the methylation levels of the same set of age-associated epigenetic markers in a control is automated and conducted by aUC Ref: BK2023-047-3 BFF Ref: BERK-518WO computer. In some embodiments, systems include a computer having a computer readable storage medium with a computer program stored thereon.

[0204] In embodiments, the system includes an input module, a processing module and an output module. The subject systems may include both hardware and software components, where the hardware components may take the form of one or more platforms, e.g., in the form of servers, such that the functional elements, i.e., those elements of the system that carry out specific tasks (such as managing input and output of information, processing information, etc.) of the system may be carried out by the execution of software applications on and across the one or more computer platforms represented of the system.

[0205] Systems may include a display and operator input device. Operator input devices may, for example, be a keyboard, mouse, or the like. The processing module includes a processor which has access to a memory having instructions stored thereon for performing the steps of the subject methods. The processing module may include an operating system, a graphical user interface (GUI) controller, a system memory, memory storage devices, and input-output controllers, cache memory, a data backup unit, and many other devices. The processor may be a commercially available processor or it may be one of other processors that are or will become available. The processor executes the operating system and the operating system interfaces with firmware and hardware in a well-known manner, and facilitates the processor in coordinating and executing the functions of various computer programs that may be written in a variety of programming languages, such as Java, Perl, C++, Python, other high-level or low-level languages, as well as combinations thereof, as is known in the art. The operating system, typically in cooperation with the processor, coordinates and executes functions of the other components of the computer. The operating system also provides scheduling, input-output control, file and data management, memory management, and communication control and related services, all in accordance with known techniques. The processor may be any suitable analog or digital system. In some embodiments, the processor includes analog electronics which provide feedback control, such as for example negative feedback control.

[0206] The system memory may be any of a variety of known or future memory storage devices. Examples include any commonly available random access memory (RAM), magnetic medium such as a resident hard disk or tape, an optical medium such as a read and write compact disc, flash memory devices, or other memory storage device. The memory storage device may be any of a variety of known or future devices, including a compact disk drive, a tape drive, a removable hard disk drive, or a diskette drive. Such types of memory storageUC Ref: BK2023-047-3 BFF Ref: BERK-518WO devices typically read from, and / or write to, a program storage medium (not shown) such as, respectively, a compact disk, magnetic tape, removable hard disk, or floppy diskette. Any of these program storage media, or others now in use or that may later be developed, may be considered a computer program product. As will be appreciated, these program storage media typically store a computer software program and / or data. Computer software programs, also called computer control logic, typically are stored in system memory and / or the program storage device used in conjunction with the memory storage device.

[0207] In some embodiments, a computer program product is described including a computer usable medium having control logic (computer software program, including program code) stored therein. The control logic, when executed by the processor the computer, causes the processor to perform functions described herein. In other embodiments, some functions are implemented primarily in hardware using, for example, a hardware state machine. Implementation of the hardware state machine so as to perform the functions described herein will be apparent to those skilled in the relevant arts.

[0208] Memory may be any suitable device in which the processor can store and retrieve data, such as magnetic, optical, or solid-state storage devices (including magnetic or optical disks or tape or RAM, or any other suitable device, either fixed or portable). The processor may include a general-purpose digital microprocessor suitably programmed from a computer readable medium carrying necessary program code. Programming can be provided remotely to processor through a communication channel, or previously saved in a computer program product such as memory or some other portable or fixed computer readable storage medium using any of those devices in connection with memory. For example, a magnetic or optical disk may carry the programming, and can be read by a disk writer / reader. Systems of the invention also include programming, e.g., in the form of computer program products, algorithms for use in practicing the methods as described above. Programming according to the present invention can be recorded on computer readable media, e.g., any medium that can be read and accessed directly by a computer. Such media include, but are not limited to: magnetic storage media, such as floppy discs, hard disc storage medium, and magnetic tape; optical storage media such as CD- ROM; electrical storage media such as RAM and ROM; portable flash drive; and hybrids of these categories such as magnetic / optical storage media.

[0209] The processor may also have access to a communication channel to communicate with a user at a remote location. By remote location is meant the user is not directly in contact with the system and relays input information to an input manager from an external device, such as aUC Ref: BK2023-047-3 BFF Ref: BERK-518WO computer connected to a Wide Area Network (“WAN”), telephone network, satellite network, or any other suitable communication channel, including a mobile telephone (i.e., smartphone).

[0210] In one embodiment, the subject systems are configured to wirelessly communicate with a server device via the communication interface, e.g., using a common standard such as 802.11 or Bluetooth® RF protocol, or an IrDA infrared protocol. The server device may be another portable device, such as a smart phone, Personal Digital Assistant (PDA) or notebook computer; or a larger device such as a desktop computer, appliance, etc. In some embodiments, the server device has a display, such as a liquid crystal display (LCD), as well as an input device, such as buttons, a keyboard, mouse or touch-screen.

[0211] In some embodiments, the communication interface is configured to automatically or semi-automatically communicate data stored in the subject systems, e.g., in an optional data storage unit, with a network or server device using one or more of the communication protocols and / or mechanisms described above.

[0212] Output controllers may include controllers for any of a variety of known display devices for presenting information to a user, whether a human or a machine, whether local or remote. If one of the display devices provides visual information, this information typically may be logically and / or physically organized as an array of picture elements. A graphical user interface (GUI) controller may include any of a variety of known or future software programs for providing graphical input and output interfaces between the system and a user, and for processing user inputs. The functional elements of the computer may communicate with each other via system bus. Some of these communications may be accomplished in alternative embodiments using network or other types of remote communications. The output manager may also provide information generated by the processing module to a user at a remote location, e.g., over the Internet, phone or satellite network, in accordance with known techniques. The presentation of data by the output manager may be implemented in accordance with a variety of known techniques. As some examples, data may include SQL, HTML or XML documents, email or other files, or data in other forms. The data may include Internet URL addresses so that a user may retrieve additional SQL, HTML, XML, or other documents or data from remote sources. The one or more platforms present in the subject systems may be any type of known computer platform or a type to be developed in the future, although they typically will be of a class of computer commonly referred to as servers. However, they may also be a main-frame computer, a workstation, or other computer type. They may be connected via any known or future type of cabling or other communication system including wireless systems, eitherUC Ref: BK2023-047-3 BFF Ref: BERK-518WO networked or otherwise. They may be co-located or they may be physically separated. Various operating systems may be employed on any of the computer platforms, possibly depending on the type and / or make of computer platform chosen. Appropriate operating systems include Windows, iOS, Oracle Solaris, Linux, IBM, Unix, and others.

[0213] Aspects of the present disclosure further include non-transitory computer readable storage mediums having instructions for practicing the subject methods. Computer readable storage mediums may be employed on one or more computers for complete automation or partial automation of a system for practicing methods described herein. In certain embodiments, instructions in accordance with the method described herein can be coded onto a computer- readable medium in the form of “programming”, where the term "computer readable medium" as used herein refers to any non-transitory storage medium that participates in providing instructions and data to a computer for execution and processing. Examples of suitable non- transitory storage media include a floppy disk, hard disk, optical disk, magneto-optical disk, CD- ROM, CD-R, magnetic tape, non-volatile memory card, ROM, DVD-ROM, Blue-ray disk, solid state disk, and network attached storage (NAS), whether or not such devices are internal or external to the computer. A file containing information can be “stored” on computer readable medium, where “storing” means recording information such that it is accessible and retrievable at a later date by a computer. The computer-implemented method described herein can be executed using programming that can be written in one or more of any number of computer programming languages. Such languages include, for example, Python, Java, Java Script, C, C#, C++, Go, R, Swift, PHP, as well as many others.

[0214] The non-transitory computer readable storage medium may be employed on one or more computer systems having a display and operator input device. Operator input devices may, for example, be a keyboard, mouse, or the like. The processing module includes a processor which has access to a memory having instructions stored thereon for performing the steps of the subject methods. The processing module may include an operating system, a graphical user interface (GUI) controller, a system memory, memory storage devices, and input-output controllers, cache memory, a data backup unit, and many other devices. The processor may be a commercially available processor, or it may be one of other processors that are or will become available. The processor executes the operating system and the operating system interfaces with firmware and hardware in a well-known manner, and facilitates the processor in coordinating and executing the functions of various computer programs that may be written in a variety of programming languages, such as those mentioned above, other high level or low-levelUC Ref: BK2023-047-3 BFF Ref: BERK-518WO languages, as well as combinations thereof, as is known in the art. The operating system, typically in cooperation with the processor, coordinates and executes functions of the other components of the computer. The operating system also provides scheduling, input-output control, file and data management, memory management, and communication control and related services, all in accordance with known techniques. Kits

[0215] In some embodiments, the present invention provides kit(s) for assaying the age of a subject or a biological sample. In some embodiments, the age being assayed is the biological age of the subject or the biological sample. In some embodiments, the age being assayed is the chronological age of the subject or the biological sample. In certain embodiments, the kit(s) comprise a set of probes and reagents for measuring the methylation levels of a set of age- associated epigenetic markers in the epigenome of the subject described herein, and said set of probes are specific for measuring the methylation of all the CpGs identified in the set of age- associated epigenetic markers in the epigenome of the subject described herein. In an exemplary embodiment, the kit(s) comprise a set of probes and reagents for measuring the methylation levels of a set of age-associated epigenetic markers in the epigenome of the subject described herein, and said set of probes and reagents are specific for measuring the methylation of the entire set of age-associated epigenetic markers in the epigenome of the subject described herein. In an exemplary embodiment, the kit(s) comprise a set of probes and reagents for measuring the methylation levels of a set of age-associated epigenetic markers in the epigenome of the subject described in the section entitled “Age-Associated Epigenetic Markers”, and said set of probes and reagents are specific for measuring the methylation of the entire set of age-associated epigenetic markers in the epigenome of the subject described in the section entitled “Age- Associated Epigenetic Markers”. In an exemplary embodiment, the kit(s) exclude probes that are specific to CpGs outside of the set of age-associated epigenetic markers in the epigenome of the subject.

[0216] In some embodiments, the present invention provides kit(s) for assaying the age of a subject or a biological sample. In some embodiments, the age being assayed is the chronological age of the subject or the biological sample, and the kit(s) comprise a set of probes and reagents for measuring the methylation levels of CpGs, and said set of probes are specific for measuring the methylation of all the CpGs identified in Table 1 or Table 2, wherein said kit(s) exclude probes that are specific to other CpGs. In some embodiments, the age being assayed is the biological age of the subject or the biological sample, and the kit(s) compriseUC Ref: BK2023-047-3 BFF Ref: BERK-518WO probes and reagents for measuring the methylation levels of CpGs, and said set of probes are specific for measuring the methylation of all the CpGs identified in Table 1 or Table 2, wherein said kit(s) exclude probes for measuring the methylation of other CpGs. In some embodiments, the kit(s) comprise probes and reagents for measuring the methylation levels of CpGs, and said set of probes are specific for measuring the methylation of all the following CpGs cg00868523, cg01767885, cg01869186, cg01900413, cg02449103, cg02247160, cg04153991, cg04495670, cg05557449, cg06362282, cg06494091, cg06918474, cg07165260, cg08499046, cg12615535, cg12717533, cg12779520, cg13399816, cg13966628, cg14993491, cg15556672, cg16443148, cg17096412, cg17216583, cg19355182, cg20261915, cg20283716, cg23417875, cg23801965, cg23983449, cg24027320, cg24112454, cg25007283, cg27621745, cg27642618, cg14811319, cg00479463, cg01681847, cg05207943, cg06899313, cg07710266, cg12518535, cg14400541, cg14657277, cg20037507, cg20445245, cg24201362, cg25734842, cg27219748, and cg27317439, wherein the kit(s) exclude probes for measuring the methylation of CpGs that are not listed in Table 1 and Table 2. The kit(s) may comprise a standard or control for identifying the age of the subject or biological sample from the measured methylation levels.

[0217] In some embodiments, the kit(s) include a carrier, package, or container that may be compartmentalized to receive one or more containers, such as vials, tubes, and the like. In some embodiments, the kit(s) optionally include an identifying description or label or instructions relating to its use. In some embodiments, the kit(s) include information prescribed by a governmental agency that regulates the manufacture, use, or sale of compounds and compositions as contemplated herein. In some embodiments, gene expression, epigenetics and / or proteomics are assayed by other than DNAme array methods, nevertheless, the data is amendable to the ML-free determination of chronological age and / or measurements (not predictions) of biological age of sample donors.

[0218] The phrase "package" means any vessel containing probes and / or reagents presented herein. In an exemplary embodiment, the package can be a box or wrapping. Packaging materials for use in packaging pharmaceutical products are well known to those of skill in the art. Examples of pharmaceutical packaging materials include, but are not limited to, blister packs, bottles, tubes, inhalers, pumps, bags, vials, containers, syringes, bottles, and any packaging material suitable for a selected formulation and intended mode of administration and treatment.

[0219] The kit(s) can also contain items that are not contained within the package but are attached to the outside of the package, for example, pipettes. The kit(s) can include compoundsUC Ref: BK2023-047-3 BFF Ref: BERK-518WO in the solid phase or in a liquid phase (such as buffers provided) in a package. The kit(s) also can include buffers for preparing solutions for conducting the methods, and pipettes for transferring liquids from one container to another.

[0220] The invention provides for kit(s) for determining age of a subject based on epigenetic modification of subject's genetic material comprising the set of age-associated epigenetic markers described herein as described in the methods of the invention, supra.

[0221] The following examples are intended to illustrate but not to limit the invention. EXAMPLES EXAMPLE 1

[0222] Code availability: The code used to process data and generate the models and Figures can be found at https: / / github.com / SkinnerCM / conboy-laboratory / tree / main / dnam-clocks, as well as at https: / / github.com / jeblanchard / DNAme-clocks-fail-tests-and-development-of-noise- barometer?search=1 DATASETS SMALL (27K) MODEL DATASETS

[0223] GSE41037: WB, Schizophrenia and health patients, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE41037

[0224] GSE20067: WB, type 1 diabetes with nephropathy and type 1 diabetes without neproprathy, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE20067

[0225] GSE20236: WB, healthy females, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE20236

[0226] GSE19711: WB, postmenopausal women with ovarian cancer and healthy controls, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE19711 MEDIUM (450K) MODEL DATASETS

[0227] GSE40279: WB, no labeled diseased samples, which can interpret to be an average population, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE40279 OUT-OF-SAMPLE (450K) DATASETS

[0228] GSE57285: WB, BRCA1 set, cancer, healthy mutants, and healthy wild types, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE57285UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0229] GSE53840: WB, HIV set, all HIV+ males, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE53840

[0230] GSE42861: PBL, Rheumatoid arthritis set, rheumatoid arthritis and healthy controls, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE42861

[0231] GSE32148: Peripheral blood; Crohns’ disease, ulcerative colitis, and normal controls; https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE32148

[0232] GSE52588: WB, Down syndrome and healthy relatives as controls, https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE52588

[0233] GSE106648: PBL, Multiple sclerosis with healthy controls https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE106648

[0234] GSE87640: WB, Irritable bowel disease with healthy controls https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE87640

[0235] GSE72774: WB, Parkinsons disease with healthy controls https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE72774

[0236] GSE100825: WB, Werner Syndrome with healthy controls https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE100825 BETA VALUES

[0237] Beta values for the set of age-associated epigenetic markers with 460 CpGs are available in the ‘460 Cyt beta value’ tab in the Supplementary Data Excel file in the Supplementary Materials for Mei, et al. (2023) Fail-tests of dna methylation clocks, and development of a noise barometer for measuring epigenetic pressure of aging and disease. Aging (Albany NY) 15(17), 8552. This data can be accessed at: https: / / www.ncbi.nlm.nih.gov / pmc / articles / PMC10522373 / or https: / / www.aging- us.com / article / 205046 / text. ELASTIC NET REGRESSION SMALL AND MEDIUM MODELS

[0238] To construct the small and medium EN models, a random 80 / 20 split was made to get a training and testing set, respectively. To create the models, the ElasticNetCV class from the sklearn.linear model library was used. After standardizing the array data using the sklearn preprocessing function StandardScaler, ElasticNetCV’s built-in cross-validation functionality was used to find optimal values for the regression parameters alpha and l1 ratio (used in Scikit-UC Ref: BK2023-047-3 BFF Ref: BERK-518WO learn’s EN regression) for each set of data. These parameters were also used to fit the randomized versions of these clocks. The predictive ability of these models was tested on the test dataset. After the models were trained and tested, their performance was also tested on out- of-sample datasets. The same data preprocessing procedure on out-of-sample datasets was followed. ELASTIC NET REGRESSION

[0239] To construct the small and medium EN models, we first did a random 80 / 20 split to get a training and testing set respectively. To create our models we used the ElasticNetCV class from the sklearn.linear model library. After standardizing the array data using the sklearn preprocessing function StandardScaler, we used ElasticNetCV’s built-in cross-validation functionality to find optimal values for the regression parameters alpha and l1 ratio (used in Scikit-learn’s EN regression) for each set of data. We used these parameters to also fit the randomized versions of these clocks. We then tested the predictive ability of these models on the test dataset. After the models were trained and tested, we also tested their performance on out- of-sample datasets. We followed the same data preprocessing procedure on out- of-sample datasets. DNAme EN PREDICTOR OF US POPULATION

[0240] We built a model that was able to predict the growth of US population, using the same dataset, GSE40279, as well as the same training techniques as for in the standard age predictor, but instead of correlating with known ages of the donors of blood samples, this EN model predicts the number of people living in US at the time of the donors’ birth. Although we’re able to accurately predict the US population from the birth year of a sample based on the sample’s methylation data, there of course is no direct connection between one’s methylation data and the US population from their birth year. In other words, the methylation profile of an individual does not determine the US population at their birth. MODELS 0 AND 1

[0241] The data were imported from the GEO database using the GEOparse package in Python. For GSE40279, kernel density estimation plots for each GSM were compared to identify potential outliers. None were found and all samples were used. The data were split 80 / 20 into training and test sets respectively. The training set was scaled using StandardScaler in the Scikit-learn library. ElasticNetCV, with 10-fold cross validation (cv=10), was used to train an elastic net regression model. The l1 ratio was set to 0.5 and the alpha hyperparameter wasUC Ref: BK2023-047-3 BFF Ref: BERK-518WO learned from a set of ten using cross validation (n alphas = 10). To check model accuracy the test data were transformed using the training set StandardScaler fit prior to predictions. UMAP AND TSNE

[0242] In certain instances, a conventional T-distributed Stochastic Neighborhood Embedding (tSNE) unsupervised Machine Learning algorithm and Uniform Manifold Approximation and Projection (UMAP) method was employed to reduce the dimension of 482421 CpGs (GSE4027) in each individual and compare the closeness between individuals in terms of CpGs beta value. Proximity in low-dimensional UMAP space identifies groups of individuals that correspond to their health condition or disease type. t-Distributed Stochastic Neighbor Embedding (t-SNE) is used for data exploration and visualization of associations between DNAme array values and chronological age.

[0243] In certain instances, we employed conventional unsupervised Machine Learning algorithm and Uniform Manifold Approximation and Projection (UMAP) method to reduce the dimension of 482421 CpGs (GSE4027) in each individual and compare the closeness between individuals in terms of CpGs beta value. Proximity in low-dimensional UMAP space identifies groups of individuals that correspond to their health condition or disease type. RETRAINING AN ELASTIC NET MODEL AFTER REMOVING NON-SIGNIFICANT CPGS

[0244] In certain instances, the significant CpGs of the elastic net model were identified by their non-zero weights. A modified dataset was made by removing these significant CpGs and a new model was trained and tested with this modified dataset. First, an Elastic Net model was trained using the ElasticNetCV function from the “sklearn” python library with trained alpha values, a L1 value of 0.5, a tolerance of 0.001, and a maximum iterations of 5000, on the GSE40279 dataset used by the DNAme Clock developed by Hannum et. al. and found that this model relies on 849 Clock CpG’s. Non-clock CpG’s were randomly removed from the dataset and retrained Elastic Net with the same parameters on datasets with 1−10% of non-clock CpG’s removed in 1% intervals and on datasets with 10−100% of non-clock CpG’s removed in 10% intervals. These novel datasets are created by independently removing, by selecting random integers with the “random” python library, the desired percent of CpG’s from the original list of non-clock CpG’s. These newly trained Elastic Net models are then simply analyzed with the python libraries “Sklearn”, “matplotlib”, and “pandas” to create FIG 4 and FIG 9.

[0245] In certain instances, the significant CpGs of the elastic net model were identified by their non-zero weights. A modified dataset was made by removing these significant CpGs and a newUC Ref: BK2023-047-3 BFF Ref: BERK-518WO model was trained and tested with this modified dataset. We first trained an Elastic Net model, using the ElasticNetCV function from the “sklearn” python library with a trained alpha value, a L1 value of 0.5, a tolerance of 0.001, and a maximum iteration of 5000, on the GSE40279 dataset used by the DNAme Clock developed by Hannum et al. and found that this model relies on 849 Clock CpGs. We then randomly removed non-clock cytosine’s from our dataset and retrained Elastic Net with the same parameters on datasets with 1-10% of non-clock cytosine’s removed in 1% intervals and on datasets with 10-100% of non-clock cytosine’s removed in 10% intervals. These novel datasets are created by independently removing, by selecting random integers with the “random” python library, the desired percent of cytosine’s from the original list of non-clock cytosine’s. These newly trained Elastic Net models are then simply analyzed with the python libraries “Sklearn”, “matplotlib”, and “pandas” to create FIG 16 and FIG 17. GENE ANNOTATIONS FOR CPGS, GENE FUNCTIONS AND FAMILIES

[0246] The gene / genome location of the CpGs, and their annotations were taken from the Illumina HumanMethylation450 manifest. The gene functions and families were determined from the GeneCards and Entrez databases (https: / / www.genecards.org / , https: / / www.ncbi.nlm.nih.gov / Web / Search / entrezfs.html) OBSERVATION OF INDIVIDUAL CPG CORRELATION WITH AGE

[0247] Datasets GSE53840 (HIV) and GSE42861 (Rheumatoid arthritis with healthy control) were employed. Diseased and healthy data from GSE42861 into were separated into sub datasets. The Pearson’s r of each CpG against age were found within their own dataset. A total of 20,898 CpGs out of 480,000 were observed to have a Pearson’s r < 0.05 in each of the three datasets. Each dataset was divided into five-year bins and the standard deviation of the 20,898 filtered CPGs was calculated. The average standard deviation vs age was plotted for each dataset using Python’s “matplotlib” module. MODIFIED FORWARD STEP-WISE SELECTION

[0248] Chronological age was regressed on the beta values for each CpG in GSE40279, and the CpGs were ranked in descending order by absolute value of their Pearson correlation with age. Individual CpGs were added to the model iteratively by rank. For each iteration the dataset was split 80 / 20 into training and validation sets respectively. An out-of-sample dataset (GSE42861) was used for the test set. The training data were scaled using StandardScaler and ordinary least squares regression was performed using linear model. LinearRegression in the Scikit-learn library using the default settings. The validation and test data sets were then transformed usingUC Ref: BK2023-047-3 BFF Ref: BERK-518WO the training set scaler. The Pearson correlation and median absolute error (sklearn.metrics.median absolute error) (MAE) for the validation and test set age predictions versus the actual sample ages were recorded. The minimum MAE achieved in the test data was used to select the optimum number of CpGs to include in the final FSW OLS model. QUANTIFICATION OF BIOLOGICAL AGE SELECTION OF THE TIGHTLY REGULATED SHOULD-BE AGE-INVARIANT CPGS

[0249] The healthy controls of four datasets (GSE40279, GSE42861, GSE52588, GSE72774) were pooled together into a new dataset by taking the intersection of their CpG sets. To identify CpGs which are tightly regulated through chronology and which should be age-invariant, CpGs with an individual correlation with age where 0.01 < r < 0.05 were filtered out. This yielded 6,282 CpGs out of 480,000. The mean beta values for each of the 6,282 CpGs were calculated and the absolute difference from the mean (ADM) was calculated for each sample, for each CpG (Equation 6,7). For each CpG, age was regressed on the ADMs of the samples and CpGs with r < 0.2 were discarded. This yielded a set of 50 CpGs. CALCULATION OF NOISE SCORE

[0250] The ADMs (surrogate for noise) of the 50 CpGs were summed for the young (25−27 year) samples, and a normalizing factor (ζ) was calculated by taking the average of these sums (Equation 8). A final noise score σ for all samples was calculated by summing the ADMs for each sample and then normalizing by ζ (Equation 9). GENERATING BIOLOGICAL AGE CURVE

[0251] A polynomial curve was then fit to the median value of the normalized sums of absolute differences from the mean (noise) for each chronological age, using np.polyfit() function The Y axis biological age numbers were calculated through the 8th degree polynomial fitted for the median value of normalized sum for each age in arthritis healthy control dataset and in the combined healthy datasets for all studied diseases. The noise score was set to be Y and find root, or the X values, from the 8th degree polynomial equation.

[0252] Briefly, during data collection, 20,000−800,000+ optical outputs (each reflecting a CpG locus and / or non-specific probe binding) of the DNA hybridization arrays are assigned relative values between 0 (unmethylated), and 1 (methylated), called beta values, based on the ratio of the methylated signal to the sum of the methylated and unmethylated signals from the aggregate of a heterogeneous cell population. In preprocessing, the beta values are scaled. The data are then fit to a linear penalized regression model which return weights, e.g., how to variablyUC Ref: BK2023-047-3 BFF Ref: BERK-518WO multiply the data points, for a linear combination of a sparse set (0.2−0.5%) features that minimize the EN cost function. This is done on what may be a non-linear process. It is assumed without empirical or theoretical basis that the EN-selected features—even those with no correlation with chronological age, and no role in regulation of gene expression—all hold some biological relevance. The prediction error for chronological age is assumed to be indicative of biological age, which is unfounded, particularly, for the experiment where N=1; and might reflect experimental errors, or natural, healthy variation in the human methylome. There is little overlap of the selected CpGs of the various published clocks, and this further calls into question the biological relevance of the age predictors. Even if the actual experimental data and not its numerical adjustment was the predictor, using chronological age as the primary response variable contradicts healthy trajectory of human aging: methylome deviates widely between healthy individuals, or even for the same individual, based on circadian rhythms, sense of well- being and financial status. NOISE BAROMETER

[0253] In approach one, to identify CpGs which are on average age-invariant but are age- deregulated, we performed regression analysis and selected CpGs with an individual correlation with age of 0.02 < r < 0.05. This yielded 6,282 CpGs out of 480,000. The mean beta values for each of the 6,282 CpG was calculated and the absolute difference from the mean (ADM) was calculated (Equation 6, 7). Then for each cytosine, age was regressed on the ADMs and CpGs with r < 0.2 were discarded. This yielded a set of 50 CpGs. The ADMs (surrogate for noise) of the 50 CpGs were summed for the young samples, and a normalizing factor (ζ) was calculated by taking the average of these sums (Equation 8). A final noise score σ for all samples was calculated by summing the ADMs for each sample and then normalizing by ζ (Equation 9). Y axis biological age numbers were calculated through the polynomial fit for the median value of normalized sum for each age in the healthy control dataset. We set the noise score to be Y and find root, or the X values, from the polynomial equation.

[0254] In approach two, to identify CpGs which are on average age-invariant, we calculated the Means of beta values of all CpGs of the six independent 450K array datasets, in the healthy cohorts, and selected the CpGs that had nearly same (>0.1% difference) Means in the young (25-28 years) and the old (<67 years) groups. This selection was then cross-checked by having Pearson’s (r) 0.02-0.05 of beta values over age, e.g., excluding the zeroes of experimental errors; this yielded 460 CpGs out of 480,000 that had such near same Mean in all six studies with the 450K DNAme arrays, in healthy subjects. These 460 CpGs where then narrowedUC Ref: BK2023-047-3 BFF Ref: BERK-518WO down to those in all six datasets, healthy subjects, with the SD of 0.015 of the absolute and relative (set to 1) Means in the young cohort (highly regulated 8 and 5 CpGs) and those with the SD of 0.3 of the absolute and relative (set to 1) Means in the young cohort (least regulated 17 and 13 CpGs).

[0255] In approach three: Two filters were used: 1. Unequal variance, e.g., the heteroskedasticity test (White’s test) for each CpG, finding all with significant (p < 0.05) unequal variance between old and young (there were 372509 of such CpGs); 2. Old (67+ years of age) SD > Young (25-28 years of age) SD by at least 20%; after the first plus second filters, 143448 CpGs remained from the 480k CpGs, e.g., the typical number of probes of the 450K DNAme array. This approach was also compared with summing SD of all 480K CpGs, e.g., the entire 450K DNAme array data. When analyzing the 143448 CpGs in the six combined datasets, some samples of the DNAme array had missing values for a few CpGs in some datasets; to ensure data consistency, all data for such samples was removed from the analyses and the plots. Polynomial curves were fit to the values, using np.polyfit function. DIRECT TEST OF THE BATCH EFFECT

[0256] Polynomial curve was fitted to the chronological age v. Median of the 293 summed SDs (the 293 CpGs of the 460 group have at least 20% higher SD in the old than young cohort). Healthy control datasets of arthritis, Multiple Sclerosis and Parkinson’s were used for generating the Median curve and overlaying it with the scatter plot of each individual dataset. The three datasets were chosen because they have the most samples out of all other studied here datasets. EXEMPLARY METHOD

[0257] Given the 50 CpG candidates of Table 1, 2, or 3: 1) calculate the absolute deviation from the mean beta value of the population for a CpG; repeats this for each of the 50 candidates; 2) add all the 50 absolute deviations and regard this total as the sum of noise for this individual; 3) repeat steps 1 and 2 for healthy young subjects whose immune systems recently matured (e.g., ages 25-27), and take the average of the sum of noise in all the young healthy subjects; 4) use the average sum of deviation from the young to normalize the individual of interest; in other words, use the result from step 2 to divide by the result from step 3; 5) Among all the normalized sum of deviations for a population, find the median normalized value for each age;UC Ref: BK2023-047-3 BFF Ref: BERK-518WO 6) find the best degree of a polynomial through the medians to minimize the sum squared error; For example, for the experiments exemplified herein, the 5thdegree polynomial equation is y = (5.240675828340669 * x) + (-0.2188129249734282 * x^2) + (0.004474781331195266 * x^3) + (-4.483740648065997e-05 * x^4) + (1.7658492552100836e-07 * x^5) + (-48.49899824287247) and the 8thdegree polynomial equation is y = (318.0470861000581 * x) + (-25.24397997218023 * x^2) + (1.1274410950264393 * x^3) + (-0.03097071704329002 * x^4) + (0.0005356837380989208 * x^5) + (-5.697293442369076e-06 * x^6) + (3.407386789366876e-08 * x^7) + (-8.777464346524329e-11 * x^8) + (-1727.276932311053), wherein x is age in years and y is the normalized sum of deviation from mean beta value of the given CpG of a given subject; 7) the plot of such polynomial should have normalized sum of deviation on the y axis and chronologize age on the x axis; and 8) convert the normalized sum of noise value on the y axis to biological age by reversely calculating the age by plugging in the normalized sum of noise in the polynomial formula as Y and the root(X) derived from the formula will be the biological age. REFERENCES

[0258] The following references are herein incorporated by reference in their entirety with the exception that, should the scope and meaning of a term conflict with a definition explicitly set forth herein, the definition explicitly set forth herein controls: Ashraf & Ip (1998) Transcriptional control: Repression by local chromatin modification. Curr Biol.8: 683–686 DOI: 10.1016 / S0960-9822(98)70435-X Lai, et al. (2019) Multi-level remodeling of transcriptional landscapes in aging and longevity. BMB Reports.52: 86–108 DOI: 10.5483 / BMBRep.2019.52.1.296 Simon, et al. (2015) Evolution of the immune system in humans from infancy to old age. Proc Roy Soc B: Biol Sci.282: 20143085 DOI: 10.1098 / rspb.2014.3085 Gibney & Nolan (2010) Epigenetics and gene expression. Heredity 105: 4–13 DOI: 10.1038 / hdy.2010.54 Rhodes, et al. (1998) Prevalence of male pattern hair loss in 18–49 year old men. Derm Surg. 24: 1330–1332 DOI: 10.1111 / j.1524-4725.1998.tb00009.x Shankar, et al. (2009) Male androgenetic alopecia: Population-based study in 1,005 subjects. Int J Trichol.1: 131 DOI: 10.4103 / 0974-7753.58556 Galkin, et al. (2021) Deepmage: a methylation aging clock developed with deep learning. Aging Dis.12: 1252 DOI: 10.14336 / AD.2020.1202 Lapierre & Ritambhara (2022) A pan-tissue dna-methylation epigenetic clock based on deep learning. NPJ Aging Mech Dis.8: 1–15 DOI: 10.1038 / s41514-022-00085-yUC Ref: BK2023-047-3 BFF Ref: BERK-518WO Zou & Hastie (2005) Regularization and variable selection via the elastic net. J Roy stat Soc.67: 301–320 DOI: 10.1111 / j.1467-9868.2005.00503.x Du, et al. (2010) Comparison of Beta-value and M-value methods for quantifying methylation levels by microarray analysis. BMC Bioinformatics.11: 587 DOI: 10.1186 / 1471-2105-11-587 Kantlehner, et al. (2011) A high-throughput dna methylation analysis of a single cell. Nucl Acids Res.39: 44 DOI: 10.1093 / nar / gkq1357 Lake, et al. (2018) Integrative single-cell analysis of transcriptional and epigenetic states in the human adult brain. Nat Biotechnol.36: 70–80 DOI: 10.1038 / nbt.4038 Cheung, et al. (2018) Single-cell chromatin modification profiling reveals increased epigenetic variations with aging. Cell.173: 1385–1397 DOI: 10.1016 / j.cell.2018.03.079 Kim, et al. (2015) Single-cell transcriptome analysis reveals dynamic changes in lncrna expression during reprogramming. Cell Stem Cell.16: 88–101 DOI: 10.1016 / j.stem.2014.11.005 Li, et al. (2011) Epigenetic regulation of caloric restriction in aging. BMC Medicine.9: 1–12 DOI: 10.1186 / 1741-7015-9-98 Zhou, et al. (2018) SeSAMe: reducing artifactual detection of dna methylation by infinium beadchips in genomic deletions. Nucl Acids Res.46: 123 DOI: 10.1093 / nar / gky691 Hop, et al. (2020) Cross-reactive probes on illumina dna methylation arrays: a large study on als shows that a cautionary approach is warranted in interpreting epigenome-wide association studies. NAR Genom Bioinform.2: 105 DOI: 10.1093 / nargab / lqaa105 Quach, et al. (2017) Epigenetic clock analysis of diet and exercise and education and lifestyle factors. Aging.9: 419 DOI: 10.18632 / aging.101168 Mann, et al. (2018) Heterogeneous responses of hematopoietic stem cells to inflammatory stimuli are altered with age. Cell Rep.25: 2992–3005 DOI: 10.1016 / j.celrep.2018.11.056 Kirschner, et al. (2017) Proliferation drives aging-related functional decline in a subpopulation of the hematopoietic stem cell compartment. Cell Rep.19: 1503–1511 DOI: 10.1016 / j.celrep.2017.04.074 Bell, et al. (2019) Dna methylation aging clocks: challenges and recommendations. Gen Biol. 20: 1–24 DOI: 10.1186 / s13059-019-1824-y Jones, et al. (2015) Dna methylation and healthy human aging. Aging Cell.14: 924–932 DOI: 10.1111 / acel.12349 Unnikrishnan, et al. (2019) The role of dna methylation in epigenetics of aging. Pharm Therap. 195: 172–185 DOI: 10.1016 / j.pharmthera.2018.11.001 Bergsma & Rogaeva (2020) Dna methylation clocks and their predictive capacity for aging phenotypes and healthspan. Neuro Ins.15: 2.63311E+15 DOI: 10.1177 / 2633105520942221 Mehdipour, et al. (2020) Rejuvenation of three germ layers tissues by exchanging old blood plasma with saline-albumin. Aging.12: 8790 DOI: 10.18632 / aging.103418 Kanherkar, et al. (2014) Epigenetics across the human lifespan. Front Cell Develop Biol.2: 49 DOI: 10.3389 / fcell.2014.00049 The dynamic epigenome (2013) Nat Struct Mole Biol.20: 58 DOI: 10.1038 / nsmb.2534 Horvath, S. (2013) Dna methylation age of human tissues and cell types. Gen Biol.14: 1–20 DOI: 10.1186 / gb-2013-1410-r115UC Ref: BK2023-047-3 BFF Ref: BERK-518WO Hannum, et al. (2013) Genome-wide methylation profiles reveal quantitative views of human aging rates. Molec Cell.49: 359–367 DOI: 10.1016 / j.molcel.2012.10.016 Dugue, et al. (2018) Dna methylation-based biological aging and cancer risk and survival: Pooled analysis of seven prospective studies. Int J Cancer.142: 1611–1619 DOI: 10.1002 / ijc.31189 Karouzakis, et al. (2011) Epigenetic deregulation in rheumatoid arthritis. Adv Exp Med Biol. 711: 137-149 DOI: 10.1007 / 978-1-4419-8216-2_10 Nemtsova, et al. (2019) Epigenetic changes in the pathogenesis of rheumatoid arthritis. Front Genet.10: 570 DOI: 10.3389 / fgene.2019.00570 Webb, et al. (2007) Haematology of down syndrome. Archiv Dis Child Neonat Ed.92: 503–507 DOI: 10.1136 / adc.2006.104638 Oshima, et al. (2017) Werner syndrome: Clinical features pathogenesis and potential therapeutic interventions. Ageing Res Rev.33: 105–114 DOI: 10.1016 / j.arr.2016.03.002 Guastafierro, et al. (2017) Genome-wide dna methylation analysis in blood cells from patients with werner syndrome. Clinic Epigen.9: 1–10 DOI: 10.1186 / s13148-017-0389-4 McInnes, et al. (2018) Umap: Uniform manifold approximation and projection for dimension reduction. arXiv:1802.034261-63 DOI: 10.48550 / arXiv.1802.03426 Georgountzou, et al. (2017) Postnatal innate immune development: from birth to adulthood. Front Immunol: 957 DOI: 10.3389 / fimmu.2017.00957 Simon, et al. (2015) Evolution of the immune system in humans from infancy to old age. Proc Roy Soc B: Biol Sci.282: 20143085 DOI: 10.1098 / rspb.2014.3085 Eling, et al. (2019) Challenges in measuring and understanding biological noise. Nat Rev Genet. 20: 536–548 DOI: 10.1038 / s41576-019-0130-6 Wilhelm-Benartzi, et al. (2013) Review of processing and analysis methods for dna methylation array data. Brit J Canc.109: 1394–1402 DOI: 10.1038 / bjc.2013.496 Wai-Shin, et al. (2016) Profiling genome-wide DNA methylation. Epigenet Chrom.9: 1–16 DOI: 10.1186 / s13072-016-0075-3 Sugden, et al. (2020) Patterns of reliability: assessing the reproducibility and integrity of dna methylation measurement. Patterns.1: 100014 DOI: 10.1016 / j.patter.2020.100014 Belsky, et al. (2022) DunedinPACE, a DNA methylation biomarker of the pace of aging. Elife; 11:e73420. Levine, et al. (2018) An epigenetic biomarker of aging for lifespan and healthspan. Aging (Albany NY).; 10:573–91. Lu, et al. (2019) DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY).; 11:303–27. Du, et al. (2010) Comparison of Beta-value and M-value methods for quantifying methylation levels by microarray analysis. BMC Bioinformatics.; 11:587. Salameh, et al. (2020) DNA Methylation Biomarkers in Aging and Age-Related Diseases. Front Genet.; 11:171. Krupa, et al. (2009) Male androgenetic alopecia: population-based study in 1,005 subjects. Int J Trichology.; 1:131–3.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO Drewelies, et al. (2022) Using blood test parameters to define biological age among older adults: association with morbidity and mortality independent of chronological age validated in two separate birth cohorts. Geroscience.; 44:2685–99 Janssen, et al. (2005) Relation between trends in late middle age mortality and trends in old age mortality--is there evidence for mortality selection? J Epidemiol Community Health.; 59:775– 81. Reichert, et al. (2012) The age distribution of mortality due to influenza: pandemic and peri- pandemic. BMC Med.; 10:162. French National Institute for Demographic Studies (INED). Age and risk of mortality.2018. from https: / / www.ined.fr / en / everything_about_population / graphs-maps / interpreted-graphs / age- risk-mortality / Becht, et al. (2018) Dimensionality reduction for visualizing single-cell data using UMAP. Nat Biotechnol. https: / / doi.org / 10.1038 / nbt.4314 PMID:30531897 Kumar, et al. (2018) Human T Cell Development, Localization, and Function throughout Life. Immunity.; 48:202–13. Eling, et al. (2019) Challenges in measuring and understanding biological noise. Nat Rev Genet.; 20:536–48. Yong, et al. (2016) Profiling genome-wide DNA methylation. Epigenetics Chromatin.; 9:26. Lu, et al. (2023) Universal DNA methylation age across mammalian tissues. Nat Aging. https: / / doi.org / 10.1038 / s43587-023-00462-6 PMID:37563227 Horvath, S., Raj, K. (2018) Dna methylation-based biomarkers and the epigenetic clock theory of ageing. Nature reviews genetics 19(6), 371–384. Bozack, et al. (2023) Dna methylation age at birth and childhood: performance of epigenetic clocks and characteristics associated with epi-genetic age acceleration in the project viva cohort. Clinical Epigenetics 15(1), 62. He, et al. (2021) The use of dna methylation clock in aging research. Experimental Biology and Medicine 246(4), 436–446. Stenvinkel, et al. (2007) Impact of inflammation on epigenetic dna methylation–a novel risk factor for cardiovascular disease? Journal of internal medicine 261(5), 488–499 (2007). Kangaspeska, et al. (2008) Transient cyclical methylation of promoter dna. Nature 452(7183), 112–11. Benbrahim-Tallaa, et al. (2005) Molecular events associated with arsenic-induced malignant transformation of human prostatic epithelial cells: aberrant genomic dna methylation and k-ras oncogene activation. Toxicology and applied pharmacology 206(3), 288–298. Bose, et al. (2012) Inherited effects of low-dose exposure to methylmercury in neural stem cells. Toxicological Sciences 130(2), 383–390. Christiansen, et al. (2021) Novel dna methylation signatures of tobacco smoking with trans- ethnic effects. Clinical epigenetics 13, 182. Massart, et al. (2014) The genome-wide landscape of dna methylation and hydroxymethylation in response to sleep deprivation impacts on synaptic plasticity genes. Translational psychiatry 4(1), 347–347.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO Cedernaes, et al. (2015) Acute sleep loss induces tissue-specific epigenetic and transcriptional alterations to circadian clock genes in men. The Journal of Clinical Endocrinology Metabolism 100(9), 1255–1261. Bhatti, et al. (2015) Nightshift work and genome-wide dna methylation. Chronobiology International 32(1), 103–112. Bind, et al. (2014) Effects of temperature and relative humidity on dna methylation. Epidemiology 25(4), 561–569. McDade, et al. (2019) Genome-wide analysis of dna methylation in relation to socioeconomic status during development and early adulthood. American journal of physical anthropology 169(1), 3–11. Mei, et al. (2023) Fail-tests of dna methylation clocks, and development of a noise barometer for measuring epigenetic pressure of aging and disease. Aging (Albany NY) 15(17), 8552. Santos-Moreno, et al. (2021) Inflammaging as a link between autoimmunity and cardiovascular disease: the case of rheumatoid arthritis. RMD open 7(1), 001470. Rezus, et al. (2019) The link between inflammaging and degenerative joint diseases. International journal of molecular sciences 20(3), 614. Calabrese, et al. (2018) Aging and parkinson’s disease: Inflammaging, neuroinflammation and biological remodeling as key factors in pathogenesis. Free Radical Biology and Medicine 115, 80–91. Russo, et al. (2022) Age-related midbrain inflammation and senecence in parkinson’s disease. Frontiers in Aging Neuroscience 14, 917797. Perdaens, et al. (2022) Molecular mechanisms of immunosenescene and inflammaging: relevance to the immunopathogenesis and treatment of multiple sclerosis. Frontiers in neurology 12, 811518. Deleidi, et al. (2015) Immune aging, dysmetabolism, and inflammation in neurological diseases. Frontiers in neuroscience 9, 140325. Simon, et al. (2021) Monocyte mitochondrial dysfunction, inflammaging, and inflammatory pyroptosis in major depression. Progress in Neuro-Psychopharmacology and Biological Psychiatry 111, 110391. Emmert-Streib, et al. (2019) High-dimensional lasso-based computational regression models: regularization, shrinkage, and selection. Machine Learning and Knowledge Extraction 1(1), 359–383. Sawilowsky, et al. (2009) New effect size rules of thumb. Journal of modern applied statistical methods 8, 597–599.

[0259] All scientific and technical terms used in this application have meanings commonly used in the art unless otherwise specified.

[0260] As provided herein, specific CpGs are referred to the Illumina probe name as provided in Illumina’s “HumanMethylation450 v1.2 Manifest File”, which is available at https: / / webdata.illumina.com / downloads / productfiles / humanmethylation450 / humanmethylation 450_15017482_v1-2.csv and is herein incorporated by reference in its entirety, and as set forth in Table 1 and Table 1A. For example, “cg00868523” refers to the CpG located onUC Ref: BK2023-047-3 BFF Ref: BERK-518WO chromosome 19 (Accession No. NC_000019.9) at map location 54387036 of the reverse (or bottom) strand. As used herein, the “level” of methylation for a given CpG in a sample is the measured fraction of the given CpG that are methylated compared to the total amount of the given CpG in a test sample. In some embodiments, the “level” of methylation of a given CpG is the beta-value obtained using Illumina’s microarray-based Infinium methylation assay platform.

[0261] Notwithstanding the appended claims, the disclosure is also defined by the following clauses:

[0262] 1. A method of determining the biological age of a subject or a biological sample obtained from the subject, which comprises a) obtaining a biological sample of the subject; b) detecting the methylation levels of a set of age-associated epigenetic markers in the epigenome of the subject, wherein the set of age-associated epigenetic markers in the epigenome of the subject is described herein; c) comparing the methylation levels of the set of age-associated epigenetic markers of the subject with the methylation levels of the same set of age-associated epigenetic markers in a control, thereby determining the biological age of a subject or a biological sample obtained from the subject.

[0263] 2. The method according to clause 1, wherein the set of age-associated epigenetic markers in the epigenome of the subject comprises or consists of those selected from the group consisting of FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28.

[0264] 3. The method according to clause 1, herein the obtaining a biological sample of the subject comprises isolating genomic DNA or nuclear DNA from the biological sample.

[0265] 4. The method according to any one of clauses 1 − 3, wherein the set of age-associated epigenetic markers consists of a number of CpGs, further comprising: − calculating the absolute deviation from the mean beta value of the measured methylation levels for each selected CpG to obtain a plurality of absolute deviations, − summing the plurality of absolute deviations to obtain a sum of deviation, and − comparing the sum of deviation to at least one standard to determine the age of the subject or the biological sample.

[0266] 5. The method according to clause 4, wherein the sum of deviation is a normalized sum of deviation.

[0267] 6. The method according to clause 5, wherein the normalized sum of deviation is obtained by dividing the sum of deviation by an average sum of deviation obtained from a control group.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0268] 7. The method according to clause 6, wherein the control group is a group of healthy subjects having a chronological age of 25–27 years (i.e., at least 25 years and less than 28 years).

[0269] 8. The method according to clause 4, wherein the at least one standard is at least one sum of deviation obtained from a control subject of a given age.

[0270] 9. The method according to clause 4, wherein the at least one standard is at least one average of a plurality of sums of deviation obtained from a control population of a given age.

[0271] 10. The method according to clause 4, wherein the at least one standard is at least one median of a plurality of sums of deviation obtained from a control population of a given age.

[0272] 11. The method according to any one of clauses 4 − 10, wherein the at least one standard is normalized for the given age.

[0273] 12. The method according to any one of clauses 1 − 11, wherein the method does not comprise a mathematical simulation.

[0274] 13. The method according to any one of clauses 1 − 12, wherein the method does not comprise data weighting of the methylation levels of the set of age-associated epigenetic markers.

[0275] 14. The method according to clause 4, wherein the set of age-associated epigenetic markers consists of one CpG.

[0276] 15. The method according to clause 14, wherein the one CpG is cg26570901.

[0277] 16. The method according to clause 4, wherein the set of age-associated epigenetic markers consists of CpGs with a correlation with age of 0.02 < r < 0.05 for a group of healthy subjects having a chronological age of 25–27 years and a group of healthy subjects having a chronological age of more than 67 years.

[0278] 17. A method for determining the biological age of a subject or a biological sample obtained from the subject, which comprises a) obtaining a biological sample of the subject; b) detecting the methylation levels of each of the CpGs in a set of age-associated epigenetic markers in the epigenome of the subject, wherein the set of age-associated epigenetic markers in the epigenome of the subject comprises or consists of those selected from the group consisting of FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; c) calculating the absolute deviation from the mean beta value of the measured methylation levels for each of the CpGs in the set of age-associated epigenetic markers in the epigenome of the subject to obtain a plurality of absolute deviations; d) summing the plurality of absolute deviations to obtain a sum ofUC Ref: BK2023-047-3 BFF Ref: BERK-518WO deviation; e) normalizing the sum of deviation; and f) correlating the normalized sum of deviation to a standard, thereby determining the biological age of the subject or the biological sample.

[0279] 18. The method according to clause 17, wherein the standard is (i) a best-fit polynomial curve of medians of normalized sums obtained from healthy control subjects of chronological ages ranging from 24 – 80 years from birth, (ii) a polynomial equation representing the best-fit polynomial curve, or (iii) a standard sum of deviation for a given biological age.

[0280] 19. The method according to clause 17, wherein the standard is (ii), and the polynomial equation is a 5th or a 6thor an 8thdegree polynomial equation derived with np.polyfit function.

[0281] 20. A method for ascertaining the health of a subject relative to the subject’s chronological age, which comprises: a) obtaining a biological sample of the subject; b) detecting the methylation levels of each of the CpGs in a set of age-associated epigenetic markers in the epigenome of the subject, wherein the set of age-associated epigenetic markers in the epigenome of the subject comprises or consists of those selected from the group consisting of FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; d) calculating the absolute deviation from the mean beta value of the measured methylation levels for each of the CpGs in the set of age-associated epigenetic markers in the epigenome of the subject to obtain a plurality of absolute deviations; e) summing the plurality of absolute deviations to obtain a sum of deviation; f) normalizing the sum of deviation; g) correlating the normalized sum of deviation to a standard, thereby determining the biological age; and h) ascertaining the health of a subject as being (1) poor where the determined biological age is significantly or statistically greater than the subject’s chronological age, (2) average where the determined biological age and chronological age of the subject are substantially similar, or (3) superior where the determined biological age is significantly or statistically less than the subject’s chronological age.

[0282] 21. The method according to clause 20, wherein the standard sum of deviation and given biological age is selected from one of the following tables: Standard Sums of Deviation Given Biological AgesUC Ref: BK2023-047-3 BFF Ref: BERK-518WO 0.70 32.9 0.75 47.2 .an ar ums o ev a on ven o og ca ges 040 276 . ationlevels of a set of age-associated epigenetic markers in the epigenome of the subject described herein, and said set of probes are specific for measuring the methylation of all the CpGs identified in the set of age-associated epigenetic markers in the epigenome of the subject described herein; and b) instructions for the use of the kit.

[0284] 23. The kit of clause 22, wherein the kit excludes a probe that is specific to a CpG outside of the set of age-associated epigenetic markers in the epigenome of the subject.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0285] Notwithstanding the appended claims, the disclosure is also defined by the following clauses:

[0286] 1. A method of determining the biological age, health relative to chronological age, of a subject or a biological sample obtained from the subject, which comprises: a) selecting a set of CpGs that are nearly the same in their methylation values from a group of healthy subjects having a chronological age of 25–27 years (i.e., at least 25 years and less than 28 years) and from a group of healthy subjects having a chronological age of more than 67 years, wherein the set of CpGs also has increased standard deviation in the group of healthy subjects having a chronological age of more than 67 years; b) detecting the methylation levels of the set of CpGs; c) comparing the methylation levels of the set of CpGs with the methylation levels of the same set of CpGs in a control, wherein the method does not require data weighting for fitting it into an equation, thereby determining the biological age, health relative to chronological age, of a subject or a biological sample obtained from the subject.

[0287] 2. The method according to clause 1, wherein the set of CpGs is a set of age-associated epigenetic markers described herein.

[0288] 3. The method according to clause 1, wherein the set of CpGs are different in their methylation values between young and old, as long as they have higher SD in the old than in the young.

[0289] 4. The method according to clause 1, wherein the set of CpGs consists of one CpG.

[0290] 5. The method according to clause 4, wherein the one CpG is cg26570901.

[0291] 6. The method according to clause 1, wherein the set of CpGs comprise or consist of those selected from the group consisting of FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28.

[0292] 7. The method according to any one of clauses 1 − 6, further comprising: − calculating the absolute deviation from the mean beta value of the measured methylation levels for each CpG of the set of CpGs to obtain a plurality of absolute deviations; − summing the plurality of absolute deviations to obtain a sum of deviation; and − comparing the sum of deviation to at least one standard to determine the biological age of the subject or the biological sample.

[0293] 8. The method according to clause 7, wherein the sum of deviation is a normalized sum of deviation.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0294] 9. The method according to clause 8, wherein the normalized sum of deviation is obtained by dividing the sum of deviation by an average sum of deviation obtained from the control.

[0295] 10. The method according to clause 9, wherein the control is a group of healthy subjects having a chronological age of 25–27 years (i.e., at least 25 years and less than 28 years).

[0296] 11. The method according to clause 7, wherein the at least one standard is at least one sum of deviation obtained from a control subject of a given age.

[0297] 12. The method according to clause 7, wherein the at least one standard is at least one average of a plurality of sums of deviation obtained from a control population of a given age.

[0298] 13. The method according to clause 7, wherein the at least one standard is at least one median of a plurality of sums of deviation obtained from a control population of a given age.

[0299] 14. The method according to any one of clauses 7 − 13, wherein the at least one standard is normalized for the given age.

[0300] 15. The method according to clause 7, wherein the set of age-associated epigenetic markers consists of CpGs with a correlation with age of 0.02 < r < 0.05 for a group of healthy subjects having a chronological age of 25–27 years and a group of healthy subjects having a chronological age of more than 67 years.

[0301] 16. A method for determining the biological age of a subject or a biological sample obtained from the subject, which comprises: a) obtaining a biological sample of the subject; b) detecting the methylation levels of each of the CpGs in a set of age-associated epigenetic markers in the epigenome of the subject, wherein the set of age-associated epigenetic markers in the epigenome of the subject comprises or consists of those selected from the group consisting of FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; c) calculating the absolute deviation from the mean beta value of the measured methylation levels for each of the CpGs in the set of age-associated epigenetic markers in the epigenome of the subject to obtain a plurality of absolute deviations; d) summing the plurality of absolute deviations to obtain a sum of deviation; e) normalizing the sum of deviation; and f) correlating the normalized sum of deviation to a standard, thereby determining the biological age of the subject or the biological sample.

[0302] 17. The method according to clause 16, wherein the standard is (i) a best-fit polynomial curve of medians of normalized sums obtained from healthy control subjects of chronologicalUC Ref: BK2023-047-3 BFF Ref: BERK-518WO ages ranging from 24 – 80 years from birth, (ii) a polynomial equation representing the best-fit polynomial curve, or (iii) a standard sum of deviation for a given biological age.

[0303] 18. The method according to clause 16, wherein the standard is (ii), and the polynomial equation is derived with np.polyfit function.

[0304] 19. The method according to clause 16, wherein the standard is (ii), and the polynomial equation is described herein.

[0305] 20. The method according to clause 16, wherein the standard is (ii), and the polynomial equation is a 5thdegree polynomial equation, or a 6thdegree polynomial equation, or an 8thdegree polynomial equation.

[0306] 21. The method according to clause 16, wherein the standard is (ii), and the polynomial equation 5thdegree polynomial equation, or a 6thdegree polynomial equation, or an 8thdegree polynomial equation derived with np.polyfit function.

[0307] As used herein, the terms “subject”, “patient”, and “individual” are used interchangeably to refer to humans and non-human animals. The terms “non-human animal” and “animal” refer to all non-human vertebrates, e.g., non-human mammals and non-mammals, such as non-human primates, horses, sheep, dogs, cows, pigs, chickens, and other veterinary subjects and test animals. In some embodiments, the subject is a mammal. In some embodiments, the subject is a human.

[0308] The use of the singular can include the plural unless specifically stated otherwise. As used in the specification and the appended claims, the singular forms “a”, “an”, and “the” can include plural referents unless the context clearly dictates otherwise.

[0309] As used herein, “and / or” means “and” or “or”. For example, “A and / or B” means “A, B, or both A and B” and “A, B, C, and / or D” means “A, B, C, D, or a combination thereof” and said “A, B, C, D, or a combination thereof” means any subset of A, B, C, and D, for example, a single member subset (e.g., A or B or C or D), a two-member subset (e.g., A and B; A and C; etc.), or a three-member subset (e.g., A, B, and C; or A, B, and D; etc.), or all four members (e.g., A, B, C, and D).

[0310] As used herein, the phrase “at least one” and “one or more of” are used interchangeably and, e.g., “one or more of A, B, and / or C” means “one or more of A”, “one or more of B”, “one or more of C”, “one or more of A and one or more of B”, “one or more of B and one or more of C”, “one or more of A and one or more of C” and “one or more of A, one or more of B, and one or more of C”.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO

[0311] The phrase “comprises or consists of A” is used as a tool to avoid excess page and translation fees and means that in some embodiments the given thing at issue: comprises A or consists of A. For example, the sentence “In some embodiments, the composition comprises or consists of A” is to be interpreted as if written as the following two separate sentences: “In some embodiments, the composition comprises A. In some embodiments, the composition consists of A.”

[0312] Similarly, a sentence reciting a string of alternates is to be interpreted as if a string of sentences were provided such that each given alternate was provided in a sentence by itself. For example, the sentence “In some embodiments, the composition comprises A, B, or C” is to be interpreted as if written as the following three separate sentences: “In some embodiments, the composition comprises A. In some embodiments, the composition comprises B. In some embodiments, the composition comprises C.” As another example, the sentence “In some embodiments, the composition comprises at least A, B, or C” is to be interpreted as if written as the following three separate sentences: “In some embodiments, the composition comprises at least A. In some embodiments, the composition comprises at least B. In some embodiments, the composition comprises at least C.”

[0313] To the extent necessary to understand or complete the disclosure of the present invention, all publications, patents, and patent applications mentioned herein are expressly incorporated by reference therein to the same extent as though each were individually so incorporated.

[0314] Having thus described exemplary embodiments of the present invention, it should be noted by those skilled in the art that the within disclosures are exemplary only and that various other alternatives, adaptations, and modifications may be made within the scope of the present invention. Accordingly, the present invention is not limited to the specific embodiments as illustrated herein, but is only limited by the following claims.

Claims

UC Ref: BK2023-047-3 BFF Ref: BERK-518WO What is claimed is:

1. A method of determining the biological age, health relative to chronological age, of a subject or a biological sample obtained from the subject, which comprises: a) selecting a set of CpGs that are nearly the same in their methylation values from a group of healthy subjects having a chronological age of 25–27 years (i.e., at least 25 years and less than 28 years) and from a group of healthy subjects having a chronological age of more than 67 years, wherein the set of CpGs also has increased standard deviation in the group of healthy subjects having a chronological age of more than 67 years; b) detecting the methylation levels of the set of CpGs; c) comparing the methylation levels of the set of CpGs with the methylation levels of the same set of CpGs in a control, wherein the method does not require data weighting for fitting it into an equation thereby determining the biological age, health relative to chronological age, of a subject or a biological sample obtained from the subject.

2. The method according to claim 1, wherein the set of CpGs are different in their methylation values between young and old, as long as they have higher SD in the old than in the young.

3. The method according to claim 1, wherein the set of CpGs consists of one CpG.

4. The method according to claim 3, wherein the one CpG is cg26570901.

5. The method according to claim 1, wherein the set of CpGs comprise or consist of those selected from the group consisting of FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28.

6. The method according to any one of claims 1 − 5, further comprising − calculating the absolute deviation from the mean beta value of the measured methylation levels for each CpG of the set of CpGs to obtain a plurality of absolute deviations, − summing the plurality of absolute deviations to obtain a sum of deviation, and − comparing the sum of deviation to at least one standard to determine the biological age of the subject or the biological sample.

7. The method according to claim 6, wherein the sum of deviation is a normalized sum of deviation.UC Ref: BK2023-047-3 BFF Ref: BERK-518WO 8. The method according to claim 7, wherein the normalized sum of deviation is obtained by dividing the sum of deviation by an average sum of deviation obtained from the control.

9. The method according to claim 8, wherein the control is a group of healthy subjects having a chronological age of 25–27 years (i.e., at least 25 years and less than 28 years).

10. The method according to claim 6, wherein the at least one standard is at least one sum of deviation obtained from a control subject of a given age.

11. The method according to claim 6, wherein the at least one standard is at least one average of a plurality of sums of deviation obtained from a control population of a given age.

12. The method according to claim 6, wherein the at least one standard is at least one median of a plurality of sums of deviation obtained from a control population of a given age.

13. The method according to any one of claims 6 − 12, wherein the at least one standard is normalized for the given age.

14. The method according to claim 6, wherein the set of age-associated epigenetic markers consists of CpGs with a correlation with age of 0.02 < r < 0.05 for a group of healthy subjects having a chronological age of 25–27 years and a group of healthy subjects having a chronological age of more than 67 years.

15. A method for determining the biological age of a subject or a biological sample obtained from the subject, which comprises a) obtaining a biological sample of the subject; b) detecting the methylation levels of each of the CpGs in a set of age-associated epigenetic markers in the epigenome of the subject, wherein the set of age-associated epigenetic markers in the epigenome of the subject comprises or consists of those selected from the group consisting of FIG 23, FIG 24, FIG 25, FIG 26, FIG 27, and FIG 28; c) calculating the absolute deviation from the mean beta value of the measured methylation levels for each of the CpGs in the set of age-associated epigenetic markers in the epigenome of the subject to obtain a plurality of absolute deviations; d) summing the plurality of absolute deviations to obtain a sum of deviation; e) normalizing the sum of deviation; andUC Ref: BK2023-047-3 BFF Ref: BERK-518WO f) correlating the normalized sum of deviation to a standard, thereby determining the biological age of the subject or the biological sample.

16. The method according to claim 15, wherein the standard is (i) a best-fit polynomial curve of medians of normalized sums obtained from healthy control subjects of chronological ages ranging from 24 – 80 years from birth, (ii) a polynomial equation representing the best-fit polynomial curve, or (iii) a standard sum of deviation for a given biological age.

17. The method according to claim 15, wherein the standard is (ii), and the polynomial equation is derived with np.polyfit function.

Images