A biological age evaluation method and system based on whole-life cycle DNA methylation and application thereof

Abstract

The application discloses a biological age evaluation method and system based on whole life cycle DNA methylation and application, and belongs to the technical field of bioinformatics. First, the obtained original DNA methylation data is subjected to quality control and standardization processing; then, differential methylation sites significantly related to calendar age are screened as features; next, taking calendar age as a target variable, a regression model is constructed by adopting a LightGBM gradient boosting framework to obtain a methylation clock model; subsequently, the model is used to predict the DNA methylation age of an individual, and an epigenetic age acceleration value is calculated based on the deviation of the DNA methylation age from the calendar age; finally, the age acceleration value is subjected to correlation analysis with health or physiological indexes in different life stages, so as to evaluate the biological aging state of the individual and predict related health risks. The application covers the whole life cycle, is suitable for Chinese population, and can provide an effective tool for clinical disease risk prediction, health management and anti-aging intervention effect evaluation.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics technology and relates to a biological age assessment method, system and application based on DNA methylation throughout the entire life cycle. Background Technology

[0002] DNA methylation levels exhibit a regular change with age, and based on this, "epigenetic clocks" can be used to predict biological age. Existing technologies mainly fall into two categories: first-generation clocks (such as Horvath and Hannum) construct models by screening CpG sites strongly correlated with calendar age, achieving high predictive accuracy in the adult population; second-generation clocks (such as PhenoAge) further incorporate clinical phenotypes as training labels, enhancing their association with disease risk. However, existing technologies still have the following technical limitations: First, the age distribution of the training data used in model construction is concentrated between 20 and 80 years old, leading to significantly increased prediction errors in children, adolescents, and those over 90 years old, failing to meet the application needs of the entire life cycle; second, existing clocks are mostly based on data from European and American populations, resulting in prediction bias when directly applied to the Chinese population, affecting their accuracy in the local population; third, existing schemes lack standardized, phased application methods, failing to establish quantitative correlations between the same biological age indicator and specific health indicators at different life stages, limiting their operability in practical health assessments. Therefore, it is necessary to construct an epigenetic clock method based on the full life-cycle methylation data of the Chinese population, which can stably predict at all ages and has the function of phased health assessment. Summary of the Invention

[0003] In view of this, in order to address the problems that existing DNA methylation clock technologies are not suitable for the Chinese population and do not achieve full life-cycle coverage and phased assessment applications, this invention provides a biological age assessment method, system, and application based on full life-cycle DNA methylation. This assessment method utilizes large cohort data of the Chinese population aged 3 to 118 years, constructs a high-precision methylation clock through a machine learning process that integrates innovative feature screening strategies, and uses its derived age acceleration indicators to establish a health risk assessment system for different life stages.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A biological age assessment method based on DNA methylation throughout the entire life cycle includes the following steps:

[0006] S1. Methylation data standardization preprocessing: The acquired raw DNA methylation data is subjected to quality control and standardization processing to obtain high-quality CpG sites;

[0007] S2, Initial screening of aging-related CpG sites: Based on the high-quality CpG site data after standardization in step S1, differentially methylated sites that are significantly associated with calendar age are screened through epigenome-wide association analysis (EWAS).

[0008] S3. Methylation Clock Construction and Optimization: Using machine learning algorithms, the methylation level of differential CpG sites selected in step S2 is used as a feature, and calendar age is used as the target variable. A regression model is constructed using the LightGBM gradient boosting framework to obtain the final methylation clock model for predicting biological age. The BorutaShap algorithm is used to select features from the pre-trained methylation clock model to select a core set of minimum CpG site features that contribute the most to age prediction.

[0009] S4. Epigenetic Acceleration Calculation and Definition: The methylation clock model constructed in step S3 is used to predict the DNA methylation age of an individual, and the epigenetic age acceleration value is calculated based on the deviation between the DNA methylation age and the calendar age.

[0010] S5. Clock Validation and Stage-Specific Association: The prediction accuracy of the methylation clock model constructed in step S3 was validated on an independent test set. The epigenetic age acceleration value obtained in step S4 was then correlated with specific health or physiological indicators at different life stages to establish a health risk assessment system for different life stages. To verify the inventiveness of this invention, the performance of the model was compared with the Horvath clock, Hannum clock, and PhenoAge clock in the general population. The results showed that the MAE (4.71 years) of the model was significantly better than that of the Horvath1 clock (MAE=6.38 years), Horvath2 clock (MAE=4.89 years), Hannum clock (MAE=5.06 years), and PhenoAge clock (MAE=6.79 years).

[0011] Furthermore, step S1 specifically includes:

[0012] S11. Obtain peripheral blood samples from a large cohort of 5,000 to 6,000 Chinese participants aged 3 to 118 years who meet the inclusion criteria, and calibrate the participants' calendar ages to ensure that the age used as the model label is highly accurate.

[0013] S12. Using the Illumina Infinium Methylation EPIC v2.0 BeadChip chip, perform full epigenomic DNA methylation detection on the peripheral blood samples obtained in step S1. Standardize the raw detection data through quality control to ensure data quality for subsequent analyses. Specific quality control measures for the raw detection data include: background correction and dye bias correction using the ENmix R package; quantile normalization and probe design bias correction; excluding samples and probes with a failure rate >5%; excluding probes located on sex chromosomes, non-CpG probes, and cross-reactive probes; imputing missing values ​​using the k-nearest neighbor method; correcting batch effects caused by different detection times using the ComBat algorithm; estimating the proportions of six leukocyte subtypes (neutrophils, monocytes, B cells, natural killer cells, CD4+ T cells, and CD8+ T cells) based on the methylation data; and retaining approximately 900,000 high-quality CpG sites after quality control. The values ​​are used for subsequent analysis and are part of the quality control data. The values ​​exhibit the expected bimodal distribution.

[0014] Furthermore, step S2 identifies differentially methylated sites through epigenome-wide association analysis (EWAS), specifically: based on the quality control methylation obtained in step S12. The values ​​were used as the dependent variable, with the calendar age precisely calibrated in step S11 as the independent variable. The sex and cell type ratio estimated in step S12 were corrected in the linear regression model, and the false discovery rate was controlled by the Bergemini-Hockberg method. CpG sites with FDR < 0.01 and an absolute value of age effect greater than 0.002 were defined as age-related DMPs. Finally, 3237 differentially methylated sites that were significantly associated with age were identified.

[0015] Furthermore, the training process of the methylation clock model in step S3 is as follows: the overall dataset is randomly divided into a training set and an independent test set in a ratio of 7:3, and hyperparameters are tuned on the training set by using 5-fold cross-validation combined with a Bayesian optimization strategy to minimize the mean absolute error (MAE); the model after hyperparameter tuning is evaluated on the independent test set by using the Pearson correlation coefficient (r), the coefficient of determination (R²), the root mean square error (RMSE), and the MAE.

[0016] Furthermore, in step S3, to further improve the interpretability and potential clinical translation convenience of the model, the BorutaShap algorithm was used to select features for the pre-trained model. This BorutaShap algorithm creates shadow features for comparison, iteratively eliminating redundant CpG sites whose importance is no higher than random noise, and finally selecting a core set of minimal features that contributes most to age prediction, namely 26 CpG sites. The performance of models built using all 3237 CpG sites and those built using the 26 sites selected by BorutaShap was compared on the test set. The results showed… The full-site model has a MAE of 4.43 years, while the BorutaShap selected model has a MAE of 4.92 years. This indicates that the 26 sites selected by BorutaShap can achieve prediction accuracy comparable to the full feature set with the fewest features, and therefore it is selected as the preferred feature set. Using this feature set, the final LightGBM model (MAE = 4.71 years) is retrained on the entire dataset using a 5-fold cross-validation strategy. This ensures that the "DNA methylation age" obtained for each individual is an out-of-sample prediction, thus obtaining an unbiased biological age estimator applicable to the entire life cycle.

[0017] Furthermore, the 26 CpG sites in step S3 are cg19283806, cg07553761, cg27099280, cg03890691, cg26614073, cg22551157, cg02378183, cg01124297, cg24866418, cg07582229, cg13206721, cg07023764, cg 26638716, cg13984040, cg19076536, cg18902238, cg14558074, cg24455300, cg00481951, c g10351253, cg12773402, cg01885725, cg16604658, cg07927379, cg16867657, cg22454769.

[0018] Furthermore, step S4 employs "DNA methylation age deviation (DNAmAgeDev)" as the core indicator of accelerated aging. This indicator is obtained by linearly regressing each individual's DNA methylation age against their chronological age, and the residual is taken as the deviation between the model's predicted value and the actual value. Based on the predictive performance (MAE) of the final model, this invention further classifies individuals into aging phenotypes: individuals with DNAmAgeDev > MAE are defined as "rapid aging individuals," individuals with DNAmAgeDev < MAE are defined as "slow aging individuals," and the rest are "normal aging individuals." This quantitative and classification system provides a clear basis for subsequent risk stratification and precise intervention.

[0019] Furthermore, step S5 involves correlation analysis between the epigenetic age acceleration value and specific health or physiological indicators at different life stages. Specifically, for individuals aged 3-18, the epigenetic age acceleration value is correlated with growth and development indicators (height, weight, age-sex Z-score of BMI) and related indicators of sex hormones; for adults aged 18-90, the epigenetic age acceleration value is correlated with cardiovascular metabolic risk indicators (blood pressure, blood lipids, blood glucose), liver enzymes (ALT, AST), and related indicators of systemic inflammation (hs-CRP); for centenarians aged 90 and above, the epigenetic age acceleration value is correlated with frailty index (FI) and cognitive function (MMSE) indicators that reflect health resilience and functional maintenance.

[0020] A biological age assessment system based on DNA methylation throughout the entire life cycle includes:

[0021] The data acquisition module is configured to acquire methylation level data of individual biological samples at the aforementioned 26 CpG sites.

[0022] The processing and computation module is configured to run a trained machine learning model, taking methylation level data as input, to calculate and output an individual's DNA methylation age.

[0023] The application output module is configured to further calculate epigenetic age acceleration indicators based on DNA methylation age and generate a health risk assessment report.

[0024] An application of a biological age assessment method based on DNA methylation throughout the entire life cycle, which is used to prepare reagent kits or detection devices for assessing an individual's aging status or predicting the risk of age-related diseases; or for developing software platforms or information systems for individual health management, evaluation of the effectiveness of aging interventions, or actuarial calculations for insurance risks.

[0025] An in vitro diagnostic kit for biological age assessment based on life-cycle DNA methylation includes reagents for specifically detecting the methylation levels of the aforementioned 26 CpG sites, and data analysis components or access instructions for executing the aforementioned biological age assessment system program based on life-cycle DNA methylation.

[0026] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described biological age assessment method based on DNA methylation throughout the life cycle.

[0027] The beneficial effects of this invention are as follows:

[0028] 1. The biological age assessment method based on DNA methylation throughout the entire life cycle disclosed in this invention makes a core methodological contribution by deeply integrating advanced machine learning frameworks with multi-stage feature optimization strategies. Firstly, in terms of modeling algorithms, this invention does not adopt the linear models commonly used in existing clocks (such as elastic networks), but innovatively introduces a gradient boosting decision tree framework, namely the LightGBM algorithm. This choice is based on its superior ability to handle high-dimensional, complex nonlinear relationships and interaction effects, enabling a more thorough exploration of the deep patterns between methylation levels and age. Secondly, in terms of feature processing, this invention designs a two-stage optimization process: first, preliminary screening is performed through whole-epigenome association analysis, and then the BorutaShap algorithm is innovatively applied for refinement. BorutaShap is a method that combines Boruta feature selection with SHAP value interpretability. By creating random "shadow features" as a benchmark, it can stably identify core CpG sites that truly contribute to the prediction after multiple iterations, while efficiently eliminating redundant and noisy features. This combined strategy not only ensured the prediction accuracy of the final model (R²=0.9387, MAE=4.7143), but also significantly improved the robustness and interpretability of the model and reduced the risk of overfitting.

[0029] 2. The biological age assessment method based on DNA methylation throughout the entire life cycle disclosed in this invention demonstrates a systematic design in terms of the completeness and applicability of the technical solution. The entire scheme is trained and validated based on a specific dataset covering a Chinese population aged 3 to 118 years, ensuring the model's adaptability to local geographical genetic background and epigenetic patterns throughout life. In the feature selection process, this invention particularly emphasizes the consistency of associations of CpG sites across different age groups, tending to select methylation markers that maintain robust associations with key physiological state evolution throughout life, thereby mechanistically supporting the model's full-cycle effectiveness. Based on this model, this invention further defines a standardized "epigenetic age acceleration" residual index and provides empirical application examples of it in age-group health assessment, demonstrating a methodological closed loop from accurate prediction to risk assessment.

[0030] In summary, the fundamental advantage of this invention lies in its cutting-edge and rigorous methodology. By innovatively combining the LightGBM algorithm with BorutaShap feature selection and implementing it within an integrated, full-cycle validation framework, this invention not only yields a high-performance prediction tool but, more importantly, establishes a reproducible, interpretable, and biologically sound modeling methodology. This provides an effective technical solution to address the performance bottlenecks and insufficient generalization capabilities caused by the simplistic model selection and feature screening strategies in existing technologies.

[0031] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0032] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0033] Figure 1 This is a flowchart illustrating the establishment and verification process of the DNA methylation clock in this invention.

[0034] Figure 2 This is the preprocessing procedure for DNA methylation data in this invention;

[0035] Figure 3 This is a correlation diagram between the predicted age and the actual age obtained by the methylation clock model of this invention through 5-fold cross-validation;

[0036] Figure 4 This is a correlation diagram between the predicted age and the actual age of other mainstream methylation clock models in this invention;

[0037] Figure 5 This is a comparison chart showing the correlation between the DNA methylation clock of this invention and other mainstream methylation clocks and cell death.

[0038] Figure 6 This is a comparative graph showing the association between the DNA methylation clock of this invention and other mainstream methylation clocks with physiological and biochemical measurements and functional phenotypes at different age groups. Detailed Implementation

[0039] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0040] like Figure 1 As shown, the present invention provides a biological age assessment method based on DNA methylation throughout the entire life cycle, comprising the following steps:

[0041] S1, such as Figure 2 The DNA methylation data preprocessing workflow shown is the integration and standardization of a full-life-cycle multidimensional dataset, based on a core dataset containing 5636 Chinese participants aged 3 to 118 years. The calendrical age of all participants was calibrated through precise calculation of birth and enrollment dates and multiple verifications (checking against ID card information and Chinese zodiac). Whole blood DNA methylation was detected using the Illumina Infinium MethylationEPIC v2.0 chip, with quality control and preprocessing performed using the ENmix package, including removal of low-quality probes and samples, background and dye bias correction, quantile normalization, and batch effect correction using the ComBat algorithm, ultimately yielding a standardized dataset containing 889074 CpG sites. Value matrix. Simultaneously, the Houseman method was used to estimate the proportions of six leukocyte subtypes: neutrophils, monocytes, B cells, natural killer cells, CD4+ T cells, and CD8+ T cells, as covariates for subsequent analyses.

[0042] S2, initial screening of aging-related methylation features; epigenome-wide association analysis (EWAS) was conducted, with each CpG site... The values ​​were used as the dependent variable, with calendar age as the independent variable. Sex and cell type ratios were adjusted to identify differentially methylated sites significantly associated with age. The Bergemini-Hockberg method was used to control the false discovery rate. CpG sites with an FDR < 0.01 and an absolute age effect greater than 0.002 were defined as age-related differentially methylated sites (DMPs), resulting in 3237 sites as the initial candidate feature library.

[0043] S3. Construction and Optimization of a Lifetime Methylation Clock Based on Machine Learning: Using the methylation levels of 3237 sites obtained in step S2 as features and calendar age as the prediction target, a regression model was constructed using the LightGBM gradient boosting framework. The dataset was randomly divided into a training set and an independent test set in a 7:3 ratio. On the training set, hyperparameter tuning was performed using 5-fold cross-validation combined with Bayesian optimization, aiming to minimize the mean absolute error (MAE). To further improve model interpretability and reduce feature dimensionality, the BorutaShap algorithm was used for feature selection of the pre-trained model. By creating shadow features and comparing their importance with the original features, redundant sites were iteratively eliminated, ultimately selecting 26 core CpG sites. The performance of models built using all 3237 loci and those using 26 loci selected by BorutaShap was compared. The results showed that the MAE of the model using all loci was 4.43 years, while the MAE of the model using the BorutaShap selection was 4.92 years. This indicates that the 26 loci selected by BorutaShap can achieve prediction accuracy comparable to the full feature set with the fewest features, and therefore should be considered the preferred feature set. Using this feature set, such as... Figure 3 The final LightGBM model was retrained on the entire dataset using a 5-fold cross-validation strategy to ensure that the DNA methylation age (DNAmAge) of each individual was an out-of-sample prediction.

[0044] S4. Calculation and Quantitative Definition of Epigenetic Age Acceleration Indicator: A linear regression of each individual's DNAmAge against their calendar age is performed. The regression residual is taken as the DNA methylation age bias (DNAmAgeDev), with positive values ​​indicating accelerated aging and negative values ​​indicating decelerated aging. Individuals are classified based on the model MAE (4.71 years): DNAmAgeDev > 4.71 indicates rapid aging, DNAmAgeDev < -4.71 indicates slow aging, and the rest are considered normal aging individuals.

[0045] S5. System validation and phased application correlation analysis of clock performance; to validate the model's predictive accuracy on independent test sets, and to perform correlation analysis between DNAmAgeDev and specific health or physiological indicators at different life stages.

[0046] like Figure 4The correlation diagram of predicted age and actual age for other mainstream methylation clock models is shown. The results show that the MAE of the model in this invention (4.71 years) is significantly better than that of Horvath1 clock (MAE=6.38 years), Horvath2 clock (MAE=4.89 years), Hannum clock (MAE=5.06 years) and PhenoAge clock (MAE=6.79 years).

[0047] The technical solution and its effects of the present invention will be described in detail below through specific embodiments.

[0048] Example 1: Application in adults aged 45 and above

[0049] This study selected a sample of 4,647 adults aged 45 and above to analyze the association between DNAmAgeDev and all-cause mortality risk. Mortality risk was assessed using a Cox proportional hazards model.

[0050] like Figure 5 As shown, in individuals aged 45 and older, each year of increase in DNAmAgeDev was associated with a 5% increase in the risk of all-cause mortality (HR=1.05, 95% CI: 1.04–1.06, P<0.001). On the same dataset, the HR associated with age acceleration and all-cause mortality for other mainstream methylation clocks was less than 1.045, while the HR value of the clock in this invention was higher, indicating superior predictive ability.

[0051] Example 2: Application in minors aged 3-18

[0052] This study selected 589 samples of minors aged 3-18 years to analyze the association between DNAmAgeDev and growth and development indicators and sex hormones. Growth and development indicators included height Z-score (zHeight), weight Z-score (zWeight), and BMI Z-score (zBMI), all calculated according to the Chinese Children's Growth Reference Standard; sex hormone indicators included estradiol (E2) and testosterone (T).

[0053] like Figure 6 Top-left results: Linear regression analysis showed that DNAmAgeDev was not significantly associated with zHeight, zWeight, or zBMI, indicating that the age acceleration index of this invention is decoupled from programmed growth and development signals. DNAmAgeDev showed a significant negative correlation with testosterone but no significant association with estradiol. On the same dataset, the age acceleration of the Horvath2 clock showed a weaker association with testosterone, demonstrating the superiority of the clock in identifying adolescent sex hormone abnormalities.

[0054] Example 3: Application in adults aged 18-65

[0055] This embodiment selected 729 adult samples aged 18-65 years to analyze the association between DNAmAgeDev and cardiovascular metabolic risk indicators, including systolic blood pressure (SBP), diastolic blood pressure (DBP), fasting blood glucose (FBG), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C).

[0056] like Figure 6 The results in the upper right corner show that, after adjusting for confounding factors such as age, sex, and cell type, DNAmAgeDev was significantly positively correlated with SBP in the 18-65 age group, with an effect size of 0.2842 (95% CI: 0.0805–0.4878; P = 0.006). On the same dataset, the clock model of this invention outperformed the Horvath1 clock, Horvath2 clock, and PhenoAge clock in these associations.

[0057] Example 4: Application in adults aged 65-90

[0058] This embodiment selected 2332 adult samples aged 65-90 years to analyze the association between DNAmAgeDev and plasma creatinine, high-sensitivity C-reactive protein (hs-CRP), basic activities of daily living (BADL), and instrumental activities of daily living (IADL).

[0059] like Figure 6 The results in the lower left corner show that, after adjusting for confounding factors such as age, sex, and cell type, DNAmAgeDev was significantly positively correlated with plasma creatinine (effect value: 0.6255; 95% CI: 0.3673–0.8836; P < 0.001) and hs-CRP (effect value: 0.0931; 95% CI: 0.0237–0.1624; P = 0.009) in the 65–90 age group. On the same dataset, the clock model of this invention outperforms most other mainstream methylation clock models in these associations.

[0060] Example 5: Application in centenarians aged 90-118 years

[0061] This study selected 1,986 long-lived elderly individuals aged 90-118 years to analyze the association between DNAmAgeDev and the Frailty Index (FI) and Mental Health Scale (MMSE). FI is calculated based on 37 health deficits, with higher values ​​indicating more severe frailty; MMSE is a cognitive function score, with lower values ​​indicating more severe cognitive impairment.

[0062] like Figure 6The results in the lower right corner show that, after adjusting for confounding factors such as age, sex, and cell type, DNAmAgeDev was significantly positively correlated with FI (effect value 0.0024, 95% CI: 0.0008–0.0040; P = 0.004) and significantly negatively correlated with MMSE (effect value -0.0992, 95% CI: -0.1908–-0.0075; P = 0.034). This indicates that the greater the acceleration of aging, the greater the degree of frailty and the worse the cognitive function. On the same dataset, the association between the acceleration of aging and FI was not significant for other mainstream methylation clocks, while the clock of this invention showed a significant and stronger effect, demonstrating its advantage in assessing the health resilience of the elderly.

[0063] Comparative Test Case: Verification of Feature Optimization Effect

[0064] To verify the optimization effect of BorutaShap feature selection, this comparative test case compares the prediction performance of models constructed using different feature sets on the same dataset. The following two feature sets were used: (1) all 3237 age-related CpG loci; (2) 26 preferred loci selected by BorutaShap. All models were trained using the LightGBM algorithm with the optimal hyperparameters, and MAE and R² were calculated on the test set.

[0065] Results: For the all-site model: MAE = 4.43 years, R² = 0.9445; For the BorutaShap optimized model: MAE = 4.92 years, R² = 0.9351. It is evident that the BorutaShap-selected 26-site model has comparable accuracy to the all-site model, but with a 99.2% reduction in the number of features, significantly lowering model complexity. This demonstrates that the BorutaShap algorithm used in this invention can effectively identify core features, greatly improving model interpretability and clinical translation potential while maintaining prediction accuracy.

[0066] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A biological age assessment method based on DNA methylation throughout the entire life cycle, characterized in that, Includes the following steps: S1. Methylation Data Standardization Preprocessing: The acquired raw DNA methylation data undergoes quality control, removing low-quality probes and samples. ENmix is ​​used for background correction, dye bias correction, and quantile normalization. The ComBat algorithm is then used to correct for batch effects, resulting in high-quality, standardized CpG sites. Value matrix; S2. Initial screening of aging-related CpG sites: Based on the data after standardization in step S1, differentially methylated sites significantly associated with calendar age were screened using epigenome association analysis (EWAS) and linear regression models. The screening criteria were: false detection rate (FDR) < 0.01 and absolute value of the regression coefficient of age effect > 0.

002. S3. Methylation Clock Construction and Optimization: Using machine learning algorithms, the methylation levels of differentially expressed CpG sites selected in step S2 are used as features, and calendar age is used as the target variable. The LightGBM gradient boosting framework is employed, combined with 5-fold cross-validation and Bayesian optimization for hyperparameter tuning, to construct a regression model and obtain the methylation clock model. Furthermore, the BorutaShap algorithm is used to select features from the pre-trained model, identifying a set of core CpG site features that contribute most to age prediction. This feature set is then used to retrain the optimized model. To further improve the model's interpretability and reduce feature dimensionality, the BorutaShap algorithm is used again for feature selection from the pre-trained model. The BorutaShap algorithm creates shadow features and compares their importance with the original features, iterating through multiple rounds to remove redundant CpG sites whose importance is no higher than random noise, ultimately selecting a set of core features that contribute most to age prediction, containing 26 CpG sites. This feature set is then used for 5-fold cross-validation across the entire dataset. The strategy involves retraining the final LightGBM model to ensure that the DNA methylation age of each individual is an out-of-sample prediction, thus obtaining a biological age estimator applicable to the entire life cycle; the 26 CpG sites are cg19283806, cg07553761, cg27099280, cg03890691, cg26614073, cg22551157, cg02378183, cg01124297, cg24866418, and cg075822.

29. cg13206721, cg07023764, cg26638716, cg13984040, cg19076536, cg18902238, cg14558074, cg2445530 0, cg00481951, cg10351253, cg12773402, cg01885725, cg16604658, cg07927379, cg16867657, cg22454769; S4. Calculation and definition of accelerated epigenetic age: The methylation clock model constructed in step S3 is used to predict the DNA methylation age DNAmAge of an individual. The calendar age is linearly regressed by DNAmAge, and the regression residual is used as the accelerated epigenetic age value DNAmAgeDev. Positive values ​​indicate accelerated aging, and negative values ​​indicate decelerated aging. S5. Clock Validation and Phased Association Application: The methylation clock model constructed in step S3 is validated on an independent test set. The Pearson correlation coefficient r, coefficient of determination R², root mean square error RMSE, and mean absolute error MAE are calculated. The epigenetic age acceleration value obtained in step S4 is correlated with health or physiological indicators at different life stages to assess an individual's biological aging status, predict the risk of age-related diseases, and evaluate the effectiveness of anti-aging interventions.

2. The biological age assessment method as described in claim 1, characterized in that, Step S1 is as follows: S11. Obtain peripheral blood samples from 5,000 to 6,000 Chinese subjects aged 3 to 118 years who meet the inclusion criteria; calculate the calendar age accurate to a decimal year based on the subjects' birth date and enrollment date: using the birth date as the baseline, calculate the actual number of years from birth to enrollment, and use multiple validation methods to calibrate the calculated age, and determine the age with the smallest deviation as the final calendar age; S12. Use the Illumina Infinium Methylation EPIC v2.0 BeadChip chip to perform full epigenomic DNA methylation detection on the peripheral blood samples obtained in step S1, and perform quality control on the raw detection data to standardize it and ensure the data quality of subsequent analyses. Quality control of the raw test data specifically included: background correction and dye bias correction using the ENmix R package, quantile normalization, and probe design bias correction; excluding samples and probes with a failure rate >5%, excluding probes located on sex chromosomes, non-CpG probes, and cross-reactive probes; imputing missing values ​​using the k-nearest neighbor method; correcting for batch effects caused by different test times using the ComBat algorithm; estimating the proportions of six leukocyte subtypes—neutrophils, monocytes, B cells, natural killer cells, CD4+ T cells, and CD8+ T cells—based on methylation data; and retaining approximately 900,000 high-quality CpG sites after quality control. The values ​​are used for subsequent analysis; they are part of the quality control data. The values ​​exhibit the expected bimodal distribution.

3. The biological age assessment method as described in claim 2, characterized in that, Step S2, which involves screening for differentially methylated sites that are significantly associated with calendar age, specifically uses the quality control methylation sites obtained in step S12. The values ​​were used as the dependent variable, with the calendar age precisely calibrated in step S11 as the independent variable. Sex and the proportion of cell types estimated in step S12 were included as covariates in the linear regression model. The Bergemini-Hockberg method was used to control the false discovery rate. CpG sites with FDR < 0.01 and absolute value of the regression coefficient of the age effect > 0.002 were defined as age-related differential methylation sites. Finally, 3237 differential methylation sites that were significantly associated with age were identified.

4. The biological age assessment method as described in claim 3, characterized in that, The training process of the methylation clock model in step S3 is as follows: the overall dataset is randomly divided into a training set and an independent test set in a 7:3 ratio; on the training set, hyperparameters are tuned by using 5-fold cross-validation combined with a Bayesian optimization strategy to minimize the mean absolute error (MAE); the tuned model is then evaluated on the independent test set, and the evaluation metrics include the Pearson correlation coefficient r, the coefficient of determination R², the root mean square error (RMSE), and the MAE.

5. The biological age assessment method as described in claim 1, characterized in that, In step S4, DNAmAgeDev (DNA methylation age deviation) is used as the core indicator of accelerated aging. This indicator is calculated by linearly regressing an individual's DNA methylation age against their calendar age and taking the regression residual as the calculated value. Based on the sign of the residual, a positive value corresponds to accelerated aging, and a negative value corresponds to decelerated aging. Further, based on the mean absolute error (MAE) of the final model, aging phenotypes are classified: individuals with DNAmAgeDev > MAE are defined as rapidly aging individuals, individuals with DNAmAgeDev < -MAE are defined as slowly aging individuals, and the rest are normally aging individuals.

6. The biological age assessment method as described in claim 5, characterized in that, In step S5, the epigenetic age acceleration value is correlated with specific health or physiological indicators at different life stages. Specifically, for individuals aged 3-18, it is correlated with growth and development indicators and sex hormone-related indicators; for adults aged 18-90, it is correlated with cardiovascular metabolic risk indicators, liver enzymes, and systemic inflammation-related indicators; and for centenarians over 90 years old, it is correlated with frailty index and cognitive function indicators that reflect health resilience and functional maintenance. These correlations are used to assess the biological aging status and related health risks at different life stages.

7. A biological age assessment system based on DNA methylation throughout the entire life cycle, characterized in that, include: The data acquisition module is configured to acquire methylation level data of individual biological samples at the 26 CpG sites in claim 1. The processing and computation module is configured to run a trained machine learning model, taking methylation level data as input, to calculate and output an individual's DNA methylation age. The application output module is configured to further calculate epigenetic age acceleration indicators based on DNA methylation age and generate a health risk assessment report.

8. The biological age assessment method based on DNA methylation throughout the entire life cycle as described in any one of claims 1 to 6, characterized in that, It is used in the preparation of reagent kits or detection devices for assessing an individual's biological aging status or predicting the risk of age-related diseases.

Images