Epigenetic clock calculation model constructed based on DNA methylation heterogeneity index and construction method

Abstract

The invention provides an epigenetic clock calculation model constructed based on DNA methylation heterogeneity indexes and a construction method, and belongs to the technical field of bioinformatics. An epigenetic clock calculation model is constructed based on DNA methylation heterogeneity indexes in excrement, and the DNA methylation heterogeneity indexes comprise a methylation entropy index, an epigenetic polymorphism index, an inconsistent reading pair proportion index and an inconsistent reading pair quantitative proportion index. The problems that in the prior art, a wild animal or large animal age estimation method is difficult in material obtaining, difficult in CpG site detection, complex in age estimation method and the like are solved. The epigenetic clock calculation model provided by the invention is higher in accuracy, simple, convenient and fast.

Description

Technical Field

[0001] The present invention belongs to the technical field of bioinformatics, and in particular relates to an epigenetic clock calculation model and a construction method. Background Art

[0002] Age, as one of the important parameters of a population, has important indicative significance for the birth rate, mortality rate and other related life history characteristics of an animal population. Age can usually be divided into actual age and biological age. Actual age refers to the sum of time experienced from the birth of an individual to the present, while biological age is the age inferred based on the anatomical and physiological development status of the individual, reflecting the actual state of the body's tissue structure and physiological function. The matching degree between biological age and actual age can reveal the relationship and mechanism between individuals and the environment, which is of great significance in population monitoring, development trend prediction, life history research, and protection and management decisions. Therefore, methods to accurately determine the age of species are particularly important. However, there is no age record for animals in the wild. Traditionally, the age of wild animals is estimated mainly from the aspects of appearance, anatomical indicators, physiological indicators and omics indicators, but the materials are often limited. These methods usually rely on animal tissues, organs or blood. Such materials are difficult to collect and are restricted by many factors. They can only roughly estimate the age of animals (juvenile, sub-adult, adult and old). Therefore, traditional methods have obvious shortcomings in calculating wild animals or large animals.

[0003] At present, the epigenetic clock is considered to be the most promising molecular method for estimating biological age. By establishing a model of the DNA methylation level of biological tissues and the age of the organism, the age of the animal can be predicted. Epigenetic clocks have been applied to humans and experimental animals (such as mice) successively, and the materials obtained include blood, muscle, saliva, oral swabs and semen. However, the application of traditional epigenetic clocks to some large animals is limited in two aspects. First, the sequencing data of traditional epigenetic clocks mainly comes from microarrays, and microarrays can only detect about 3% of CpG sites (taking human genome research as an example), and microarrays can usually only measure the methylation level of each CpG independently, and cannot perform a continuous extraction of DNA methylation level mode; second, at this stage, the materials required for epigenetic clocks are difficult to obtain in some large animals (including wild animals). Therefore, the above method is not suitable for completing the age estimation of some large animals (including wild animals). In order to estimate the age of large animals (including wild animals) more quickly, conveniently and accurately, it is necessary to construct a new age estimation method. Summary of the invention

[0004] The present invention solves the problems in the prior art of age estimation methods for wild animals or large animals, such as difficulty in obtaining materials, difficulty in detecting CpG sites, and relatively complicated age estimation methods.

[0005] A method for constructing an epigenetic clock computational model, wherein the method uses feces as material and obtains DNA methylation heterogeneity indicators from the feces to construct a machine learning model for calculating age.

[0006] Preferably, the DNA methylation heterogeneity index comprises methylation entropy, apparent polymorphism, inconsistent read pair ratio or quantitative ratio of inconsistent read pairs.

[0007] Preferably, the machine learning model is an elastic regression model, a support vector machine regression model or a random forest regression model.

[0008] Preferably, the method is:

[0009] Step 1: Extract genomic DNA from feces. The nuclear CT value of the genomic DNA is no more than 27, and the bacterial CT value of the genomic DNA is greater than 7.

[0010] Step 2: subjecting the genomic DNA to sulfite conversion treatment to obtain bisulfite-modified genomic DNA;

[0011] Step 3: construct a library of bisulfite-modified genomic DNA, and then perform whole-genome DNA methylation sequencing and sequencing data processing to obtain a DNA methylation heterogeneity information report, and then use the dplyr package to detect and delete outliers according to the 3sigma principle. The DNA methylation heterogeneity information report includes a methylation entropy index, an apparent polymorphism index, an inconsistent read pair ratio index, or an inconsistent read pair quantitative ratio index;

[0012] Step 4: Analyze the sites in the four DNA methylation heterogeneity indicators after removing outliers according to the Spearman correlation, screen out the target heterogeneity sites, and model the target heterogeneity sites and age through the randomForest package, e1071 package and glmnet package.

[0013] Preferably, the sequencing data is processed by using Bismark v.2.4.4 and metheor v.0.1.8 to obtain a DNA methylation heterogeneity information report for each sample.

[0014] Preferably, the target heterogeneous site is a target heterogeneous site whose correlation with age is greater than or equal to 0.7 or less than or equal to -0.7, which is screened out based on the methylation entropy index.

[0015] Preferably, the target heterogeneity site is a target heterogeneity site selected based on the apparent polymorphism index and having an age correlation greater than or equal to 0.7 or less than or equal to -0.7.

[0016] Preferably, the target heterogeneous site is a target heterogeneous site whose correlation with age is greater than or equal to 0.6 or less than or equal to -0.6, which is screened out based on the inconsistent read pair ratio index.

[0017] Preferably, the target heterogeneous site is a target heterogeneous site with an age correlation greater than or equal to 0.7 or less than or equal to -0.7 selected based on a quantitative ratio index of inconsistent read pairs.

[0018] A computational model of an epigenetic clock, wherein the model is constructed using the above method.

[0019] Beneficial Effects

[0020] Compared with the prior art, the method of predicting animal age by using fecal DNA methylation heterogeneity index in the present invention has the following advantages:

[0021] First, the selected material is feces, which is non-damaging and easily available. This solves the problem that when estimating the age of large animals (including wild animals) by traditional methods, the required materials including blood, muscle, saliva, oral swabs and semen are not easy to obtain. Therefore, the method constructed by the present invention is more convenient and rapid.

[0022] Secondly, choosing whole genome methylation sequencing (WGBS) can obtain more methylation heterogeneity indicator information. The DNA methylation heterogeneity indicator information obtained for each sample includes: ME (methylation entropy), PM (epimorphism), FDRP (discordant read pair ratio) and qFDRP (quantitative ratio of inconsistent read pairs).

[0023] Finally, based on the fecal DNA methylation heterogeneity index, DNA methylation heterogeneity sites related to age were screened, and a model for predicting age was established in combination with machine learning. The model for predicting age established by the present invention has a higher correlation than the traditional detection method. Therefore, the model for predicting animal age through fecal DNA methylation heterogeneity index of the present invention is more accurate, which can provide reference and inspiration for establishing age identification of other animals. DETAILED DESCRIPTION

[0024] Example 1.

[0025] The present invention is further described below in conjunction with implementation examples, but the present invention is not limited to the following implementation examples.

[0026] Implementation condition 1: The nuclear CT value of the fecal DNA extracted by the present invention does not exceed 27, and the bacterial CT value is greater than 7;

[0027] Implementation condition 2: When the present invention uses stool DNA to build a library, the kit selected is xGen TM Methyl-SeqDNA Library Prep Kit;

[0028] Implementation condition three: the correlation between the screening age and the target sites of the four DNA methylation heterogeneity indicators in the present invention is limited to greater than or equal to 0.6, greater than or equal to 0.7, less than or equal to -0.6 or less than or equal to -0.7;

[0029] Implementation condition four: The present invention constructs calculation models of target sites and age of four DNA methylation heterogeneity indicators based on three models: elastic regression, support vector machine regression model and random forest regression, and determines the optimal parameters of each model and evaluates the effect through leave-one-out cross-validation.

[0030] The present invention is further described in detail below in conjunction with specific implementation modes.

[0031] 1. DNA extraction and DNA sulfite conversion treatment.

[0032] 1. Sample preparation.

[0033] The samples were feces of 33 captive Siberian tigers from the Siberian Tiger Park in Heilongjiang Province and Hengdaohezi. The samples were numbered and the birth date, sampling date, and gender of each individual were recorded, and the samples were stored in a -80℃ refrigerator.

[0034] 2. Extraction of sample DNA.

[0035] The genomic DNA was extracted from the above 33 stool samples according to the method established by our research group (patent authorization number: CN113186185A). The extracted DNA was labeled and stored at -20°C for future use.

[0036] 3. DNA sulfite conversion treatment.

[0037] The DNA extracted from 33 stool samples was subjected to sulfite conversion using the EZ DNA Methylation-Gold Kit (D5006) (ZYMO, USA). The conversion was completed according to the instructions, and the obtained bisulfite-modified genomic DNA was stored in a -20°C refrigerator.

[0038] 2. Obtain sample methylation heterogeneity indicator information.

[0039] Using xGen TMMethyl-Seq DNA Library Prep Kit was used to construct libraries for bisulfite-modified genomic DNA. After the library test was qualified (the library concentration was preliminarily quantified using Qubit, and the total amount was greater than or equal to 200 ng), high-throughput sequencing was performed on the Illumina platform. The sequencing was commissioned to Novogene, and the sequencing method was whole-genome DNA methylation sequencing (WGBS).

[0040] The sequenced data were aligned and deduplicated using Bismark v.2.4.4 to obtain the bam file of each sample. On this basis, the bam file was processed using the corresponding modules of each indicator in metheor v.0.1.8 to obtain the DNA methylation heterogeneity information report of each sample, which included four indicators: methylation entropy, apparent polymorphism, proportion of inconsistent read pairs, and quantitative proportion of inconsistent read pairs.

[0041] Methylation entropy is an information-theoretic metric used to quantify the complexity or heterogeneity of the methylation status of a single genomic site. It evaluates heterogeneity by calculating the entropy of different methylation haplotypes in a single site.

[0042] Epigenetic polymorphism is an indicator used to measure DNA methylation heterogeneity in a specific genomic region, proposed by Landan et al. The essence of epigenetic polymorphism is to quantify the degree of uneven distribution of methylation status from a statistical perspective, which is similar to methylation entropy, another methylation heterogeneity indicator, but their relationship is similar to the relationship between entropy and Gini index in a decision tree: methylation entropy focuses on the randomness of the distribution of methylation status; epigenetic polymorphism focuses on the uneven distribution of methylation status. Therefore, epigenetic polymorphism is a more intuitive method to measure DNA methylation heterogeneity, which can reflect the complexity of methylation in a specific region of the genome.

[0043] The proportion of discordant read pairs is a metric for quantifying the epigenetic diversity of a single CpG site, proposed by Scherer et al. (2020). It assesses the heterogeneity of a site by comparing whether the methylation status of sequencing read pairs at the same CpG site is consistent.

[0044] The quantitative discordant read ratio is an extension of the discordant read ratio. It quantitatively measures the degree of difference between two sequencing reads by calculating the normalized Hamming distance, rather than simply judging whether they are consistent or inconsistent.

[0045] 3. Data processing and screening of age-related DNA methylation heterogeneity indicators.

[0046] 1. Data processing.

[0047] The dplyr package (R language) was used to integrate the DNA methylation heterogeneity reports of the 33 samples according to the same methylation heterogeneity indicators (methylation entropy, apparent polymorphism, proportion of inconsistent read pairs, and quantitative proportion of inconsistent read pairs). Then, outliers were detected according to the 3 sigma principle, and outliers were set as null values. Rows with 20% null values ​​in the entire data set were deleted.

[0048] 2. Screening of target heterogeneity sites related to age.

[0049] Based on Spearman correlation analysis, the correlation between sites and age in the four DNA methylation heterogeneity indicators was screened, and the correlation restrictions were as follows: in the methylation entropy indicator, Spearman correlation analysis was used to screen out target heterogeneity sites with a correlation with age greater than or equal to 0.7 or less than or equal to -0.7; in the epigenetic polymorphism indicator, Spearman correlation analysis was used to screen out target heterogeneity sites with a correlation with age greater than or equal to 0.7 or less than or equal to -0.7; in the inconsistent read pair ratio indicator, Spearman correlation analysis was used to screen out target heterogeneity sites with a correlation with age greater than or equal to 0.6 or less than or equal to -0.6; in the quantitative ratio indicator of inconsistent read pairs, Spearman correlation analysis was used to screen out target heterogeneity sites with a correlation with age greater than or equal to 0.7 or less than or equal to -0.7.

[0050] 4. Model, evaluate and predict the target sites and ages in the four DNA methylation heterogeneity indicators respectively.

[0051] Based on the glmnet package, e1071 package and randomForest package, machine learning models of age and target sites in the four indicators were constructed respectively. The machine learning models were elastic regression model, support vector machine regression model and random forest regression model. The leave-one-out cross-validation method was used to evaluate each model, and the evaluation index of each model (R 2 , MSE, RSME) and the optimal parameters of each model, and the correlation between the predicted age and the actual age in three different machine learning models was used to select the optimal model. 2 It represents correlation, MAE represents the mean of absolute errors (hereinafter referred to as errors), and RMSE represents the root mean square error. The evaluation criteria are that the higher the correlation, the smaller the error, and the higher the accuracy of the model.

[0052] 5. Results

[0053] 1. Results of modeling based on methylation entropy indicators.

[0054] A total of 9 target heterogeneity sites with correlations with age greater than or equal to 0.7 or less than or equal to -0.7 were screened out through Spearman correlation analysis. In order to further verify the accuracy of the estimated age, the leave-one-out cross-validation method was combined with three machine learning models: elastic regression model, support vector machine regression model, and random forest regression model to predict age for the 9 points. The results are shown in Table 1. The results show that the support vector machine regression model and the random forest regression model can be applied to the fecal DNA methylation entropy index to predict age.

[0055] When the elastic regression model is used to calculate age, in the training set, the correlation between the calculated age and the actual age can reach 0.7815, with an error of 18.5 months, and in the test set, the correlation between the predicted age and the actual age can reach 0.6988, with an error of only 20.70 months; when the support vector machine regression model is selected as the analysis model, in the training set, the correlation between the predicted age and the actual age can reach 0.9130, with an error of only 10.29 months, and in the test set, the correlation between the predicted age and the actual age can reach 0.8938, with an error of only 9.00 months; and when the random forest regression model is selected as the analysis model, in the training set, the correlation between the predicted age and the actual age can reach 0.9229, with an error of only 12.41 months, and in the test set, the correlation between the predicted age and the actual age can reach 0.9118, with an error of only 8.34 months. Since the random forest regression model has the highest correlation and the smallest error, the random forest regression model is a superior model in predicting age calculation based on fecal DNA methylation entropy.

[0056] Table 1

[0057]

[0058] 2. Modeling results of epigenetic polymorphism indicators.

[0059] A total of 22 target heterogeneous sites with correlations with age greater than or equal to 0.7 or less than or equal to -0.7 were screened out through Spearman correlation analysis. In order to further verify the accuracy of the estimated age, the leave-one-out cross-validation method was combined with three machine learning models: elastic regression model, support vector machine regression model, and random forest regression model to predict age for the 22 points. The results are shown in Table 2. The results show that the support vector machine regression model and the random forest regression model can be applied to the prediction of age by fecal DNA epigenetic polymorphism indicators.

[0060] When the elastic regression model was used to calculate age, in the training set, the correlation between the calculated age and the actual age reached 0.8748, with an error of 13.90 months, and in the test set, the correlation between the predicted age and the actual age reached 0.9532, with an error of only 15.32 months; when the support vector machine regression model was selected as the analysis model, in the training set, the correlation between the predicted age and the actual age reached 0.9593, with an error of only 8.01 months, and in the test set, the correlation between the predicted age and the actual age reached 0.9163, with an error of only 12.20 months; and when the random forest regression model was selected as the analysis model, in the training set, the correlation between the predicted age and the actual age reached 0.9553, with an error of only 9.87 months, and in the test set, the correlation between the predicted age and the actual age reached 0.9435, with an error of only 9.33 months. Since the random forest regression model had the highest correlation and the smallest error, the random forest regression model was a superior model in predicting age calculation based on fecal DNA methylation epigenetic polymorphism.

[0061] Table 2

[0062]

[0063] 3. Modeling results of inconsistent read ratio indicators.

[0064] A total of 51 target heterogeneous sites with correlations with age greater than or equal to 0.6 or less than or equal to -0.6 were screened out through Spearman correlation analysis. In order to further verify the accuracy of the estimated age, the leave-one-out cross-validation method was combined with three machine learning models: elastic regression model, support vector machine regression model, and random forest regression model to predict age for the 51 sites. The results are shown in Table 3. The results show that the random forest regression model can be applied to the fecal DNA methylation inconsistent read ratio index to predict age.

[0065] When the elastic regression model is used to calculate age, in the training set, the correlation between the calculated age and the actual age can reach 0.9962, with an error of 2.68 months, and in the test set, the correlation between the predicted age and the actual age can reach 0.8938, with an error of only 19.68 months; when the support vector machine regression model is selected as the analysis model, in the training set, the correlation between the predicted age and the actual age can reach 0.8984, with an error of only 11.93 months, and in the test set, the correlation between the predicted age and the actual age can reach 0.9733, with an error of only 13.07 months; and when the random forest regression model is selected as the analysis model, in the training set, the correlation between the predicted age and the actual age can reach 0.9620, with an error of only 8.36 months, and in the test set, the correlation between the predicted age and the actual age can reach 0.9709, with an error of only 12.34 months. Since the random forest regression model has the highest correlation and the smallest error, the random forest regression model is a superior model in predicting age calculation based on the proportion of inconsistent read pairs of fecal DNA methylation.

[0066] Table 3

[0067]

[0068] 4. Modeling results of quantitative ratio indicators of inconsistent read pairs.

[0069] A total of 30 target heterogeneity sites with correlations with age greater than or equal to 0.7 or less than or equal to -0.7 were screened out through Spearman correlation analysis. In order to further verify the accuracy of the estimated age, the leave-one-out cross-validation method was combined with three machine learning models: elastic regression model, support vector machine regression model, and random forest regression model to predict age for the 30 points. The results are shown in Table 4. The results show that the support vector machine regression model and the random forest regression model can be applied to the quantitative index of the proportion of inconsistent read pairs of fecal DNA methylation to predict age.

[0070] When the elastic regression model was used to calculate age, in the training set, the correlation between the calculated age and the actual age reached 0.8106, with an error of 19.63 months, and in the test set, the correlation between the predicted age and the actual age reached 0.9821, with an error of only 8.67 months; when the support vector machine regression model was selected as the analysis model, in the training set, the correlation between the predicted age and the actual age reached 0.9051, with an error of only 11.03 months, and in the test set, the correlation between the predicted age and the actual age reached 0.9740, with an error of only 18.16 months; and when the random forest regression model was selected as the analysis model, in the training set, the correlation between the predicted age and the actual age reached 0.9614, with an error of only 9.41 months, and in the test set, the correlation between the predicted age and the actual age reached 0.9656, with an error of only 12.83 months. Since the random forest regression model had the highest correlation and the smallest error, the random forest regression model was a superior model in predicting age calculation based on the proportion of fecal DNA methylation inconsistent reads.

[0071] Table 4

[0072]

Claims

1. A method for constructing an epigenetic clock computational model, characterized in that: The method uses feces as material and obtains DNA methylation heterogeneity indicators from the feces to construct a machine learning model for calculating age.

2. The method according to claim 1, characterized in that The DNA methylation heterogeneity index includes methylation entropy, apparent polymorphism, inconsistent read pair ratio or quantitative ratio of inconsistent read pairs.

3. The method according to claim 1, characterized in that The machine learning model is an elastic regression model, a support vector machine regression model or a random forest regression model.

4. The method according to claim 1, characterized in that The method is: Step 1: Extract genomic DNA from feces. The nuclear CT value of the genomic DNA is no more than 27, and the bacterial CT value of the genomic DNA is greater than 7. Step 2: subjecting the genomic DNA to sulfite conversion treatment to obtain bisulfite-modified genomic DNA; Step 3: construct a library of bisulfite-modified genomic DNA, and then perform whole-genome DNA methylation sequencing and sequencing data processing to obtain a DNA methylation heterogeneity information report, and then use the dplyr package to detect and delete outliers according to the 3sigma principle. The DNA methylation heterogeneity information report includes a methylation entropy index, an apparent polymorphism index, an inconsistent read pair ratio index, or an inconsistent read pair quantitative ratio index; Step 4: Analyze the sites in the four DNA methylation heterogeneity indicators after removing outliers according to the Spearman correlation, screen out the target heterogeneity sites, and model the target heterogeneity sites and age through the randomForest package, e1071 package and glmnet package.

5. The method according to claim 4, characterized in that The sequencing data were processed using Bismark v.2.4.4 and metheor v.0.1.8 to obtain a DNA methylation heterogeneity information report for each sample.

6. The method according to claim 4, characterized in that The target heterogeneity site is a target heterogeneity site whose correlation with age is greater than or equal to 0.7 or less than or equal to -0.7, which is screened out based on the methylation entropy index.

7. The method according to claim 4, characterized in that The target heterogeneity site is a target heterogeneity site that is screened out based on the apparent polymorphism index and has an age correlation greater than or equal to 0.7 or less than or equal to -0.

7.

8. The method according to claim 4, characterized in that The target heterogeneous site is a target heterogeneous site whose correlation with age is greater than or equal to 0.6 or less than or equal to -0.6, which is screened out based on the inconsistent read ratio index.

9. The method according to claim 4, characterized in that The target heterogeneity sites are target heterogeneity sites with an age correlation greater than or equal to 0.7 or less than or equal to -0.7, which are screened based on the quantitative ratio index of the inconsistent read pairs.

10. A computational model of an epigenetic clock, characterized in that: The model is constructed by the method according to any one of claims 1 to 9.

Images