Device and method for constructing biological age measurement index suitable for specific population, and constructed biological age measurement device and biological age measurement method

Abstract

The present invention provides a device and a method capable of constructing a model that has high prediction accuracy and is suitable for a specific group from a small number of DNA methylation data items obtained from the specific group while reducing the number of samples required for model construction. A biological age measurement device 1 according to the present invention measures the biological age on the basis of the DNA methylation level. By using, as an explanatory variable matrix (n×p matrix (p≤q)), first explanatory variable information based on DNA methylation information in CpG sites at q points included in DNA of each sample of a first population composed of n samples, and using, as an objective variable vector (n-dimensional), first objective variable information based on biological age information of the n samples, the biological age measurement device 1 performs transfer learning on the basis of a biological age measurement model estimated by a second population different from the first population, and outputs the learning result as a trained biological age measurement model.

Description

Apparatus and method for constructing a biological age measurement index adapted to a specific population, and constructed biological age measurement apparatus and biological age measurement method

[0001] The present invention relates to an apparatus and method for providing a mathematical algorithm or the like capable of constructing a biological age model or biological age index with high predictive accuracy adapted to a specific population from a small amount of DNA methylation data obtained from the specific population.

[0002] There are approximately 30 million CpG sites (also known as CpG regions, DNA sequences in which cytosine (C) is followed by consecutive guanine (G)) in human DNA. A method is known for predicting biological age (age based on the state of the body's cells or tissues, which reflects bodily function or the progression of aging, and the risk of age-related diseases) based on the DNA methylation level of some of these CpG sites. This method is generally called the epigenetic clock.

[0003] Non-Patent Document 1 discloses a method for estimating the chronological age of a healthy individual based on DNA methylation information. Non-Patent Document 2 discloses a method for estimating a human biological age by incorporating information on biomarkers of diseases that increase the risk of death in addition to DNA methylation information.

[0004] Steve Horvath, "DNA methylation age of human tissues and cell types", Genom Biology 2015; Morgan E Levine, Ake T Lu, Austin Quach, Brian H Chen, Themistocles L Assimes, Stefania Bandinelli, Lifang Hou, Andrea A Baccarelli, James D Stewart, Yun Li, Eric A Whitsel, James G Wilson, Alex P Reiner, Abraham Aviv, Kurt Lohman, Yongmei Liu, Luigi Ferrucci, Steve Horvath, "An epigenetic biomarker of aging for lifespan and healthspan", Aging (Albany NY) 2018; Albert T Higgins-Chen, Kyra L Thrush, Yunzhang Wang, Christopher J Minteer, Pei-Lun Kuo, Meng Wang, Peter Niimi, Gabriel Sturm, Jue Lin, Ann Zenobia Moore, Stefania Bandinelli, Christiaan H Vinkers, Eric Vermetten, Bart P F Rutten, Elbert Geuze, Cynthia Okhuijsen-Pfeifer, Marte Z van der Horst, Stefanie Schreiter, Stefan Gutwinski, Jurjen J Luykx, Martin Picard, Luigi Ferrucci, Eileen M Crimmins, Marco P Boks, Sara Hagg, Tina T Hu-SeligerMorgan E Levine, “A computational solution for supporting reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking” Nat Aging 2022,

[0005] As disclosed in Non-Patent Documents 1 and 2, methods for estimating human biological age from DNA methylation information are known, but the population used for development contains almost no specific races (e.g., Japanese), and there is a problem that the prediction accuracy of the above methods decreases when applied to specific races. Therefore, in order to construct a method specialized for a specific race, it has been desired to collect new DNA methylation data for the relevant race and develop an index that matches the genetic characteristics of the specific race.

[0006] However, the prediction accuracy of the models disclosed in Non-Patent Document 1 and Non-Patent Document 2 is highly dependent on the use of a sufficient number of samples for learning (in fact, existing technologies such as Non-Patent Document 2 use thousands of training data sets to build a model). Therefore, in order to build a highly accurate model, it is necessary to collect data sets from at least several thousand people, and preferably from more than 10,000 people, which is not easy from the viewpoint of cost, which is an issue.

[0007] In order to solve the above problems, the present invention aims to provide an apparatus and method that can construct a model with high prediction accuracy from a small amount of DNA methylation data while reducing the number of samples required for model construction.

[0008] In order to achieve the above object, the biological age measuring device and the biological age measuring method are characterized by the following invention-specific features.

[0009] The biological age measurement device measures biological age based on DNA methylation levels, and performs transfer learning based on a biological age measurement model estimated using a second population different from the first population, using first explanatory variable information based on DNA methylation information at q CpG sites contained in the DNA of each sample in a first population consisting of n samples as an explanatory variable matrix (n × p matrix (p≦q)), and first objective variable information based on the biological age information of the n samples as an objective variable vector (n dimensions), and outputs the learning result as a trained biological age measurement model.

[0010] In any one of the above biological age measurement devices, it is preferable that, in the transfer learning, a regression coefficient vector β̂ (p-dimensional) having an argument β (p-dimensional) that satisfies Expression 1 is calculated as the first regression coefficient information.

[0011]

[0012] Here, y is n-dimensional first dependent variable information, X is n×p first explanatory variable information, λ, α, and ρ are adjustment parameters, and β~ is p-dimensional second regression coefficient information in the biological age measurement model estimated by the second population.

[0013] In any one of the above biological age measurement devices, the biological age measurement model estimated using the second population is a biological age model in which second explanatory variable information based on DNA methylation information at q CpG sites contained in DNA of each sample of the second population consisting of n' samples is an explanatory variable matrix (n' x p matrix (p≦q)), and second objective variable information based on the biological age information of the n' samples is an objective variable vector (n' dimension), and the explanatory variable vector for each sample of the second explanatory variable information has as its elements coordinate values ​​in a p-dimensional space whose axes are first principal component to p-th principal component calculated after principal component analysis of the DNA methylation information at q CpG sites contained in DNA of each sample of the second population, and the explanatory variable vector for each sample of the first explanatory variable information preferably has as its elements coordinate values ​​in a p-dimensional space whose axes are the first principal component to p-th principal component.

[0014] The biological age measurement method measures biological age based on DNA methylation levels, and includes the steps of: defining first explanatory variable information based on DNA methylation information at q CpG sites contained in the DNA of each sample in a first population consisting of n samples as an explanatory variable matrix (n × p matrix (p≦q)); defining first objective variable information based on the biological age information of the n samples as an objective variable vector (n dimensions); performing transfer learning based on a biological age measurement model estimated using a second population different from the first population; and outputting the learning result as a trained biological age measurement model.

[0015] According to this biological age measurement device and biological age measurement method, DNA methylation information of a first population is subjected to principal component analysis, followed by transfer learning. Therefore, feature quantities of DNA methylation information can be reliably acquired from a small amount of training data. Furthermore, a model of the first population is created based on a model previously determined for the second population, using the feature quantities as explanatory variables. Therefore, a model of the first population can be constructed with high accuracy from information with a smaller number of samples than conventional methods, while reducing the amount of calculations.

[0016] Fig. 1 is a block diagram showing the functional configuration of a biological age measurement device according to an embodiment, and Fig. 2 is a flowchart showing a control process of a biological age measurement method performed by the biological age measurement device according to an embodiment.

[0017] (Configuration) A biological age measurement device 1 according to one embodiment will be described in the following description and drawings. In this embodiment, an information analysis process will be described as an example of the predetermined process. As shown in FIG. 1 , the biological age measurement device 1 of this embodiment includes a transfer learning unit 11, a storage unit 12, a control unit 13, an input unit 14, and an output unit 15.

[0018] Each of the transfer learning unit 11, the storage unit 12, and the control unit 13 is configured with a processing unit such as a CPU or a memory, an I / O interface, and a storage device such as a ROM, a RAM, an HDD, etc. The processing unit is configured with one or more CPUs that read necessary software and data from the memory and execute specified arithmetic processing on the data in accordance with the software, and, as necessary, communication equipment, a storage device (the memory), etc.

[0019] The storage unit 12 is configured to store and hold the results of the arithmetic processing performed by the arithmetic processing device.

[0020] The arithmetic processing device is composed of an information processing unit (CPU) that reads software and data from a specified area of ​​the memory that constitutes the storage device as needed, and then performs specified arithmetic processing on the data in accordance with the software, and, as needed, communication equipment, storage device (the memory), etc.

[0021] The input unit 14 is configured with operation buttons, a microphone, etc., and allows the user to perform contact-based operation or non-contact operation by speech. The output unit 15 is configured with a display device, an audio output device (speaker), etc., and displays the biological age according to the biological age measurement model or outputs audio content. The input unit 14 and the output unit 15 may be configured with a touch panel display.

[0022] (Function) Next, the biological age measurement method in the biological age measurement device 1 according to one embodiment of the present invention will be described in the following description and drawings. Figure 2 is a flowchart showing the control process of the biological age measurement method performed by the biological age measurement device according to one embodiment of the present invention.

[0023] (Generation of DNA Methylation Information) First, DNA methylation information for each sample of the first population is transmitted or uploaded to the biological age measurement device 1 via the input unit 14 (FIG. 2 / STEP 1).

[0024] In this embodiment, "DNA methylation" refers to the methylation of DNA. DNA methylation refers to the chemical modification of DNA molecules, a chemical reaction in which a methyl group is added to the carbon atom at the 5-position of a cytosine base in DNA. DNA methylation, which is passed down from cell to cell, is known to appear at CpG sites in genomic DNA. It is known that there are approximately more than 30 million CpG sites in the human genome, and in this embodiment, DNA methylation information can be obtained by measuring DNA methylation at multiple CpG sites.

[0025] DNA methylation at specific CpG sites in any sample may be measured by any method, including DNA methylation-specific PCR, whole-genome bisulfite sequencing, and enzymatic methylation sequencing.

[0026] By the above method, the DNA methylation information of an arbitrary sample is expressed by the DNA methylation degree at at least one (q sites) CpG site. That is, the DNA methylation information of a sample with an identifier i is expressed as a q-dimensional DNA methylation information vector x i The DNA methylation information vector x i Each element of the DNA methylation information vector x indicates the degree of DNA methylation at q specific CpG sites. i Each element of may be expressed in a normalized manner across multiple samples.

[0027] The DNA methylation information vector x for each sample obtained by the above method i is input to the biological age measuring device 1 via the input unit 14 and stored in the memory unit 12.

[0028] (Generation of biological age information) Following STEP 1, biological age information for each sample in the first population or variables associated therewith (clinical biomarkers), etc. are transmitted or uploaded to the biological age measurement device 1 via the input unit 14 (Figure 2 / STEP 2).

[0029] In this embodiment, "biological age" refers to a numerical value calculated using a method such as that described in U.S. Patent Application Publication No. 2020 / 0347461, which predicts differences in all-cause mortality, cause-specific mortality, physical function, cognitive performance measures, and risk of facial aging among individuals of the same age. Biological age is a function expressed by variables including clinical biomarkers of a sample. Examples of such variables include blood albumin concentration, blood creatinine concentration, blood glucose concentration, blood C-reactive protein concentration, blood lymphocyte concentration, mean corpuscular volume, erythrocyte volume particle size distribution width, blood alkaline phosphatase concentration, and white blood cell count per unit volume in blood. These variables can be obtained by blood tests, etc.

[0030] In addition to these variables, variables such as the age, sex, height, and weight of the specimen may be added when calculating the biological age.

[0031] In this embodiment, the biological age is calculated using the formula described in U.S. Patent Application Publication No. 2020 / 0347461, and the biological age of each sample is input as a target variable into the biological age measurement device 1 via the input unit 14 and stored in the memory unit 12.

[0032] (Transfer Learning) After STEP 2, the transfer learning unit 11 transfers the DNA methylation information vector x of n samples via the control unit 13. i or the dimension-reduced DNA methylation information vector x' iThe first explanatory variable information, whose elements are the explanatory variable matrix (n × p matrix (p≦q)), and the first objective variable information based on the biological age information of n samples is used as the objective variable vector (n-dimensional), transfer learning is performed based on the biological age measurement model estimated by a second population different from the first population, and the learning result is output as a trained biological age measurement model (FIG. 2 / STEP 3). Here, the first explanatory variable information is an n × p matrix, where p≦q, which is a q-dimensional DNA methylation information vector x i The dimension reduction method is not particularly limited, but it is preferable to reduce the dimension by principal component analysis. That is, the DNA methylation information vector x i is subjected to principal component analysis, and the dimension-reduced DNA methylation information vector x', which has the first principal component to the pth principal component as elements, is obtained. i The first explanatory variable information may be constructed by the following. In this case, in the biological age model for the second population described below, the explanatory variables of the second population data are processed using the parameters used when the dimension was reduced in the first population. Also, as described below, dimension reduction may be performed in the second population, and the DNA methylation information vector of the first population may be processed using the parameters and then used as the explanatory variables. Note that, in addition to principal component analysis, latent semantic analysis, linear discriminant analysis, etc. may also be used as the dimension reduction method.

[0033] The biological age model created in a predetermined second population used in transfer learning is a DNA methylation information vector x of each sample in the second population. i The biological age model is based on the DNA methylation information vector x obtained in the second population. i Each element is information calculated based on the same CpG sites as those used in the first population. Note that a portion of the samples in the second population may be the same as or different from a portion of the samples in the first population. If the second population targets all of humanity, a portion of that sample group (e.g., Japanese people) may be treated as the first population.

[0034] Next, a method for creating a biological age model for the second population will be described. The biological age model created for the second population is a DNA methylation information vector x i or the dimension-reduced DNA methylation information vector x' i The model is calculated by using the second explanatory variable information having elements of n' as an explanatory variable matrix (n' × p matrix (p≦q)), and the second objective variable information based on the biological age information of n' samples as an objective variable vector (n-dimensional).

[0035] The biological age model of the second population is calculated by performing any regression analysis. Here, the regression analysis method is not particularly limited, but it is preferable to calculate it by the Elastic Net method. However, without being limited to this, the biological age model of the second population may be calculated by various regression methods (Lasso method, SCAD method, MCP method, Bridge method, other regression methods with penalties, or regression analysis methods involving a variable selection step). Here, the obtained regression coefficient vector β ～→ (β0, ..., β p-1 ) T ∈R P (R P means a p-dimensional real space) is stored in the storage unit 12 in advance.

[0036] The biological age model for the second population is the dimension-reduced DNA methylation information vector x' i In this case, the explanatory variable vector for each sample of the second explanatory variable information has as its elements coordinate values ​​in a p-dimensional space whose axes are the first principal component to the p-th principal component calculated by performing principal component analysis on DNA methylation information at q CpG sites contained in the DNA of each sample of the second population.

[0037] In this case, the explanatory variable vector for each sample of the first explanatory variable information has as its elements the coordinate values ​​of a p-dimensional space whose axes are the first principal component to the p-th principal component, which are calculated based on the eigenvectors obtained by performing principal component analysis on the second population.

[0038] The number of dimensions to be reduced can be selected arbitrarily, but it is preferable to select principal components until the cumulative contribution rate of the principal components (cumulative contribution rate when accumulating principal components in descending order of their contribution rate) is 50-90% or more. This reduces the number of dimensions while preserving the information of the feature quantities, and the dimension-reduced DNA methylation information x' i is created.

[0039] The transfer learning method is not particularly limited, but it is preferable to use the Transfer Elastic Net method, which is a transfer learning method of a penalized estimation method. Other methods that can be used include parameter transfer learning methods such as the Transfer Lasso method, and transfer learning methods based on a method of correcting covariate shift.

[0040] Transfer learning allows for the calculation of an accurate model using a biological age model of a second population, even with a small number of samples from the first population.

[0041] Specifically, a method for creating a biological model in the first population using the Transfer Elastic Net method will be explained. The regression coefficient vector β is set as follows: ～ Let β be the initial estimate (the estimate obtained in a previous population (the regression coefficient of the biological age model created in the second population)), and β be the regression coefficient vector obtained by the Transfer Elastic Net method. The estimate by the Transfer Lasso is determined by the above-mentioned Equation 1. In Equation 1, y is the objective variable (biological age), and y is expressed as y = (y, ..., y n ) T ∈R n and x i is the explanatory variable vector (DNA methylation information vector for each sample or dimension-reduced DNA methylation information vector), and x is x i = (x i,0 ,...,x i,p-1 ) T ∈R n X is the first explanatory variable matrix, and X is expressed as X = (x1, ..., x n ) T ∈Rn×p The explanatory variables may be standardized, in which case the explanatory variable vector x i may be converted.

[0042]

[0043] β is a variable, and β is expressed as β = (β0, ..., β p-1 ) T ∈R p Also, the regression coefficient vector β ＾ is a regression coefficient vector constituting the biological age model in the first population, and is calculated by calculating the variable β.

[0044] The regression coefficient vector β of Equation 1 ＾ is β for each j j When the elements of the variable β except for j That is, the latter part of the equation (1) is determined by repeating the process of finding β j The derivative after partial differentiation is β j By repeatedly solving for ＾ Each element of the formula is calculated. j When partially differentiated with respect to, the following relational expression (3) is obtained.

[0045]

[0046] Here, the subscript (-j) indicates a negative identification number j, which represents a matrix or vector with the j-th column or element removed. -j is β ｊ It is a vector excluding X and is a p-1 dimensional vector. -j is X j is a matrix obtained by removing p and is an n×(p−1) dimensional matrix.

[0047] The minimum value of the formula 3 is determined by the following equation: ｊ By solving for β ｊ When the equation in formula 3 is 0, β ｊWhen we solve for this, we obtain the following equation:

[0048]

[0049] Here, τ is a function that satisfies the following equation 5.

[0050]

[0051] By repeating the calculation of Equation 4 for j = 0, ..., p-1, the regression coefficient vector β ＾ Each element of the above is calculated, and a biological age model for the first population is created (Figure 2 / END).

[0052] The above method allows for the creation of an accurate biological age model within a first population from limited data.

[0053] Next, examples of the present invention will be described together with comparative examples.

[0054] Example 1 In Example 1, a biological age model is created by performing transfer learning using Japanese people (N=143) as the first population and a multi-ethnic dataset (N=4282) as the second population, as shown in Table 1 below (cited from Non-Patent Document 3). Note that unqualified samples are deleted in advance from the second population.

[0055]

[0056] A total of 78,464 CpG sites described in Non-Patent Document 3 were used as the CpG sites for the above samples. As a method for creating biological ages, as described above, principal component analysis was performed on the second population, and an explanatory variable matrix for the first population was created based on the results of the principal component analysis. The biological age model was then created by using the biological age model cited in Non-Patent Document 3 as a pre-trained model in transfer learning and performing transfer learning based on the above-described Equation 1. Note that the results of the principal component analysis (eigenvectors, eigenvalues), etc., were calculated by citing the results based on Non-Patent Document 3, calculating biological ages using the method disclosed in Non-Patent Document 3, and using the biological ages as the target variables during learning.

[0057] The final model formula is as follows: where the k-th principal component is represented as PCk (k is any integer) (i.e., the first principal component is "PC1").

[0058] y = 1.0684937560723364 + 3.865131787197643e-07 × PC1 + 0.0003093000532258508 × PC2 + -0.034711923293021855 × PC3 + 0.034754635067519635 × PC4 + -0.013743287066026016 × PC5 + 0.07321044909585893 × PC6 + -0.16320136554867656 × PC7 + 0.10780070423154997 × PC8 + -0.002729867598070007 × PC9 + -0.00010621455139944036 × PC10 + 0.06291343514053144 × PC11 + 0.09540923558190754 × PC12 + 0.017386948372940164 × PC13 + 0.014292068112307528 × PC14 + 0.008055542946309826 × PC15 + -0.00457524118449414 × PC17 + 0.044726511205009464 × PC18 + 0.01838813742594696 × PC20 + -0.1108159548686523 × PC21 + 0.0637210750499973 × PC22 + -0.2241636882345891 × PC23 + -0.0018763430964875195 × PC24 + 0.03544506216197382 × PC26 + -0.2067540509014271 × PC27 + 0.04407987179060844 × PC29 + 0.0026862947073898274 × PC30 + -0.1045223496761178 × PC31 + 0.06756517648202467 × PC32 + 0.02547892504524941 × PC33 + -0.14379454237123793 × PC34 + 0.15007347007017005 × PC36 + 0.16472989700498838 × PC37 + 0.1461327022053658 × PC38 + 0.013014437938349235 × PC39 + 0.18531931731077048 × PC40 + 0.024848850389579992 × PC41 + -0.08334571782068884 × PC42 + 0.09325267561275595 × PC43 + 0.08200857308239107 × PC44 + -0.1343605007497738 × PC46 + 0.1383350579905576 × PC47 + -0.04101947768603644 × PC48 + -0.12494967920625522 × PC49 + 0.006521749211561273 × PC50 + -0.0996730864432895 × PC51 + -0.2432716795174693 × PC52 + -0.11617871317529194 × PC53 + -0.019184389297169802 × PC54 + -0.07771603427919238 × PC55 + -0.1402881559884475 × PC60 + -0.09774728186275583 × PC61 + -0.06622473699963534 × PC62 + 0.15447599210679552 × PC63 + -0.10046448501316184 × PC65 + 0.11278926380214811 × PC66 + 0.13502774089466194 × PC68 + -0.14475026640221572 × PC70 + 0.052545122206327406 × PC72 + -0.12540833871303686 × PC73 + 0.0025788087910837277 × PC74 + 0.053863363838963806 × PC75 + 0.10939324686645796 × PC76 + -0.09137957549549387 × PC78 + 0.0032004360212760387 × PC79 + 0.11030504798336076 × PC80 + -0.08119806918317159 × PC81 + -0.1384690975111963 × PC82 + -0.10650931475151605 × PC83 + 0.07071898461729759 × PC84 + 0.034954148034633536 × PC91 + -0.1274641058830218 × PC92 + -0.043444186493263356 × PC94 + -0.08755763638700281 × PC96 + 0.08459769586324213 × PC98 + 0.041100459971020825 × PC99 + -0.01686693635107783 × PC103 + 0.017306154562532917 × PC104 + -0.11655195002870508 × PC109 + 0.006350336300432371 × PC110 + 0.08190091010526081 × PC111 + 0.10540816887655918 × PC113 + 0.1126630218840087 × PC114 + -0.013145646697197165 × PC115 + 0.0563583796271912 × PC117 + -0.09579443394314933 × PC120 + -0.011854978275113998 × PC121 + 0.11477981113954372 × PC128 + -0.15566524508707802 × PC138 + -0.15240134869563493 × PC139 + 0.011059674966686342 × PC141 + -0.055712421356374275 × PC142 + 0.09307395650313306 × PC144 + 0.05857889162415979 × PC157 + -0.10152801555293484 × PC161 + 0.050834459795783306 × PC162 + 0.07575476533604135 × PC171 + -0.11799229127616641 × PC179 + -0.12708493923628986 × PC182 + -0.04249738267627002 × PC185 + -0.06263101818964804 × PC187 + 0.07864505891834456 × PC199 + 0.029549620857190807 × PC213 + -0.05692037619747273 × PC229 + -0.07324026869440954 × PC242 + 0.023230788875735035 × PC256 + 0.06330250825855949 × PC258 + -0.023999462698742285 × PC270 + -0.0186100732424434 × PC276 + 0.06518782350670763 × PC291 + 0.05151925274340331 × PC292 + 0.0030102667208926435 × PC316 + -0.03678240975952578 × PC321 + -0.10032130620292898 × ​​PC356 + 0.018012012388189873 × PC371 + 0.06199463674031713 × PC373 + 0.09686890368066704 × PC419 + 0.0976140009997348 × PC422 + 0.09101165041272279 × PC459 + -0.013609873701760648 × PC483 + 0.08048434275771621 × PC564. .

[0059] Using the above biological age model, the data of 50 Japanese people not included in the training data were used as evaluation data, and the root mean square error (RMSE) and mean absolute error (MSE) between the calculated biological age and chronological age were calculated as evaluation indices. Similarly, as a comparative example, using similar evaluation data, biological age was calculated by the methods described in Non-Patent Documents 1 and 3, and the root mean square error and mean absolute error between biological age and chronological age were calculated as evaluation indices. The results are shown in Table 2.

[0060]

[0061] As shown in Table 2, the biological age model calculated by the method of this embodiment has a smaller root mean square error and a smaller absolute mean error than the biological age models calculated by the methods of Non-Patent Document 1 and Non-Patent Document 3. As is clear from this, a model with high prediction accuracy was constructed from a small amount of DNA methylation data, while reducing the number of samples required for model construction.

[0062] Example 2 Example 2 is a biological age model using Japanese people (N=143, DNA methylation array data quantified using a bead array provided by Illumina, Inc.) as the first population and a multi-ethnic dataset (N=4505) as the second population, as shown in Table 3 below (citing Non-Patent Document 3).

[0063]

[0064] A total of 78,464 CpG sites described in Non-Patent Document 3 were used as the CpG sites for the above samples. As a method for creating biological ages, as described above, principal component analysis was performed on the second population, and an explanatory variable matrix for the first population was created based on the results of the principal component analysis. The biological age model was then created by using the biological age model cited in Non-Patent Document 3 as a pre-trained model in transfer learning and performing transfer learning based on the above-mentioned Equation 1. Note that the results of the principal component analysis (eigenvectors, eigenvalues), etc., were based on Non-Patent Document 3, and biological ages were calculated using the method disclosed in Non-Patent Document 3, and the biological ages were then used as the objective variables during learning.

[0065] The final model formula is as follows: where the k-th principal component is represented as PCk (k is any integer) (i.e., the first principal component is "PC1").

[0066] y = 63.111746667196826 + 0.26922107840806214 × PC1 + -0.7298858820830854 × PC2 + -0.03552784016699162 × PC3 + 1.5718822306988958 × PC4 + 2.9010901103367868 × PC5 + 0.1817512039031038 × PC6 + -1.0679817439682435 × PC7 + -4.080478634196635 × PC8 + -0.7829358847827792 × PC9 + 0.8913534875353344 × PC11 + 5.436171963348199 × PC12 + -2.2508837464769957 × PC13 + 0.33919280247663447 × PC14 + -0.19107104121633633 × PC15 + -0.0769574735125609 × PC16 + -1.816747680792368 × PC17 + -0.6856702844170304 × PC20 + 0.2126690744412973 × PC21 + 3.1902962368390866 × PC22 + 3.5686286591879655 × PC23 + 1.4617646308827583 × PC24 + -0.7579865465853979 × PC25 + -0.20364362459016028 × PC26 + -2.758276028543011 × PC28 + -2.282920963180542 × PC29 + -0.11388125265590952 × PC30 + 1.2817606570201963 × PC31 + -3.0479714801670905 × PC33 + -1.3687177971388154 × PC36 + 1.2293506494667703 × PC37 + -2.403186936548198 × PC38 + 2.6371930815047646 × PC39 + -2.868124119728694 × PC40 + -1.0253680223200898 × PC41 + -3.2375198401299605 × PC42 + 0.3252167095858183 × PC43 + -2.1936427430650007 × PC44 + 2.7479523144487352 × PC45 + -0.15693586942047544 × PC47 + 1.4594214706773618 × PC50 + 3.0744432950493272 × PC51 + 3.0819744142623944 × PC52 + -1.412844233205062 × PC53 + -1.6344526574649427 × PC55 + -0.08200017651764531 × PC56 + 0.5802002492976321 × PC57 + 0.7599152894543612 × PC59 + 0.6949551059349163 × PC60 + -0.12979774646171865 × PC61 + -1.5677402627569927 × PC62 + 1.60830959197858 × PC63 + 0.6291846919357673 × PC64 + 0.7074785462476649 × PC65 + -2.4975634144251617 × PC66 + -2.7191234567992577 × PC69 + 1.1592237814028794 × PC71 + 0.8183577745685418 × PC72 + -1.2417885169272775 × PC75 + -0.7193486554154812 × PC77 + 0.9739303439176458 × PC78 + -0.7525164202090525 × PC79 + 1.6571112340570195 × PC84 + -2.0331089974155434 × PC87 + 0.20211926060127922 × PC91 + 1.7731118423680208 × PC93 + 2.3542671573724614 × PC94 + 0.22558529503395447 × PC95 + -1.2312820231005412 × PC97 + -0.06473257142490603 × PC98 + 0.8699660048769089 × PC100 + -0.11932650569350255 × PC101 + 0.1142024328863405 × PC103 + -0.10604330089266317 × PC104 + 1.7148059584781103 × PC105 + -0.8702366499633138 × PC109 + -0.34168963690599236 × PC110 + 1.58946918571943 × PC111 + -0.5227407256422456 × PC117 + 0.7966087567411375 × PC118 + -1.4584716771102795 × PC119 + -1.8713094258744933 × PC120 + 1.293070127572819 × PC121 + -0.3298463215099348 × PC122 + 1.1082139934179795 × PC123 + -1.8188839794547118 × PC124 + -0.5528772589426211 × PC125 + 0.17004641625350647 × PC126 + -0.04682491046763926 × PC127 + 0.40203555634510085 × PC130 + 1.1667400562693104 × PC131 + -0.3028272282101481 × PC132 + 1.2600661784931941 × PC135 + -0.8424436292871652 × PC136 + -0.9514104793691754 × PC137 + 1.0070073535991206 × PC138 + 1.5803399792219035 × PC139 + -0.8806225940604838 × PC143 + -1.0798100729563846 × PC149 + -0.4897454540242623 × PC153 + -0.9556491951918603 × PC154 + 1.2865414917036169 × PC155 + 0.6845467213064929 × PC157 + 0.10916232362898734 × PC158 + -1.6392963064133177 × PC159 + 0.6407518984249425 × PC162 + -1.237144223453998 × PC164 + -0.2133288206490249 × PC165 + 1.1175314359713882 × PC167 + 1.1756522278005421 × PC168 + 1.2669893050205046 × PC171 + -1.202818730101287 × PC174 + 0.37634405425641493 × PC180 + -0.3822169177767356 × PC191 + -0.9130315623942084 × PC194 + -0.8487263784627334 × PC197 + 0.02923515814125916 × PC199 + 0.9821125666864426 × PC200 + 0.8859554424665247 × PC201 + 1.61268766792111 × PC202 + -0.8733296823971699 × PC214 + -0.8648683370745637 × PC217 + -1.3121805698312627 × PC219 + -0.6852256007979154 × PC225 + -0.5878323315394106 × PC228 + -0.1578803223821195 × PC231 + -0.2981044037355804 × PC232 + 1.3325581066892638 × PC234 + 0.18148887430638297 × PC235 + 0.5043632456452196 × PC238 + -0.4239266792452775 × PC241 + -0.28214152371004536 × PC244 + -0.9379243939902578 × PC248 + -0.16519899054703732 × PC249 + 1.0555323365853837 × PC252 + 1.398087355002323 × PC256 + 0.3566619377546422 × PC262 + -0.8357172933733062 × PC265 + -0.24421010249053637 × PC266 + -1.0030050448429588 × PC267 + 0.9739586622560846 × PC269 + -0.30217151704364753 × PC280 + -1.046733699500965 × PC283 + -1.022270702462819 × PC284 + -0.2729325292921159 × PC286 + 0.4163950444172007 × PC287 + 0.8880361564485226 × PC288 + -0.122659594510236 × PC290 + -1.1561206558002877 × PC291 + -0.8695276252699665 × PC292 + 1.4281890041685974 × PC293 + -0.8654607727213595 × PC299 + -0.11188800616446466 × PC300 + 0.8484328107547114 × PC303 + -0.0027489031215913538 × PC308 + -0.04265304454884609 × PC311 + 1.0740737208399862 × PC313 + 1.1933415407316503 × PC317 + -1.3498658292768617 × PC318 + 1.288819312244302 × PC319 + 1.16041874085255 × PC321 + -0.4525280997004218 × PC324 + 1.482054524405375 × PC325 + 1.033930805577118 × PC327 + -1.178056892495577 × PC329 + -1.3700379410346382 × PC330 + 0.314737055611534 × PC332 + 0.06824964715810404 × PC334 + 0.896417781614081 × PC337 + -0.6973537848303589 × PC340 + 0.48267149062680353 × PC344 + -0.3029351947302698 × PC354 + 0.937182391962428 × PC359 + 1.5046967217322407 × PC373 + -1.1344056565849328 × PC375 + -0.7241566746578377 × PC376 + 0.3747906974979381 × PC377 + -0.9673189596285033 × PC379 + -0.7984039075634631 × PC385 + -0.29270243793935063 × PC386 + 0.3328449839939359 × PC387 + -0.17229453481558166 × PC393 + -0.6386210335558072 × PC397 + 0.9488492868872823 × PC402 + 1.2907513436408493 × PC405 + 0.035471823576875124 × PC406 + 0.6352332647254619 × PC407 + -0.41990083780539694 × PC423 + -0.8535194758338498 × PC424 + -0.8181575375311504 × PC426 + -1.1913013546133615 × PC433 + 0.12397570638120187 × PC435 + -1.4598044305617974 × PC437 + -1.6548401283068643 × PC438 + -0.20012912589350212 × PC444 + -0.25254091226936876 × PC445 + -0.6740289643479296 × PC450 + -0.058407144157358544 × PC453 + -0.14979749322314426 × PC463 + -1.4208782354569889 × PC466 + -1.0933107669347384 × PC469 + -0.21785448309671504 × PC478 + -1.2337147155568178 × PC480 + 0.45904357371315896 × PC484 + 2.131463625328412 × PC489 + 0.08396096529562781 × PC490 + 0.11165261927333722 × PC493 + 0.859525577129201 × PC495 + -0.9883764191021158 × PC499 + -1.9965064221515307 × PC504 + -0.7805743877157412 × PC507 + 0.2471253438386581 × PC519 + 0.8520489735227632 × PC527 + -0.01905074969429247 × PC530 + 0.8668040680524957 × PC532 + -1.3065885486494409 × PC533 + -1.0019942470261565 × PC534 + 0.4832316276767328 × PC537 + 0.549289678176 × PC539 + -0.25546404831758607 × PC545 + 0.8484682169602652 × PC546 + -0.28108347055848654 × PC552 + 0.9781312164472251 × PC556 + 1.176425664183624 × PC564 + 0.046135245912087744 × PC569 + -0.8046378956526167 × PC577 + -1.3273465372948787 × PC584 + 0.7877534782125428 × PC587 + -0.01912988042100864 × PC590 + -0.9778469968108285 × PC594 + 1.0949724181085636 × PC597 + -0.5285400152604083 × PC602 + -0.5693612034079895 × PC603 + 0.8432872676137926 × PC604 + -1.1050298754453627 × PC611 + 1.5213761679619324 × PC613 + -0.5625995228791907 × PC614 + 0.6608201476501512 × PC615 + -0.7693341777876052 × PC619 + -1.4704308542462678 × PC620 + 0.05606750848621045 × PC623 + 0.4057505179513613 × PC627 + -0.5387047547701279 × PC630 + -0.8302235340386873 × PC631 + -1.4477115829516654 × PC633 + 0.7166143672230727 × PC635 + 0.13732585792555288 × PC638 + 0.8734434770612034 × PC643 + 0.34510444860326417 × PC644 + 1.0056971922791975 × PC648 + -0.7511178537030346 × PC653 + -0.8679091495284654 × PC655 + -0.18449329060454356 × PC657 + -0.30262017565404986 × PC658 + 0.3782711349648734 × PC666 + -1.2256245391241527 × PC679 + -1.4916391816497863 × PC681 + 0.8808408998844491 × PC682 + -0.6843749878591097 × PC688 + -0.46637334634657296 × PC691 + 0.044953872154085896 × PC692 + -1.1751521462812948 × PC695 + -0.6928414515706711 × PC699 + -0.22536803723053467 × PC708 + -0.5792113773387146 × PC715 + -0.3451398285706757 × PC716 + 0.9696111606701618 × PC724 + -0.5552949966982792 × PC734 + -0.7921012822603863 × PC735 + -0.1462198588140244 × PC736 + -0.1563127525107274 × PC738 + -0.7875323339681394 × PC743 + -1.128541617500316 × PC744 + 1.00812574381799 × PC746 + -0.5754164012165478 × PC747 + -0.7368907117411361 × PC758 + -0.009870937211172453 × PC764 + 0.09484706735758276 × PC772 + -0.8109840946814013 × PC786 + 0.49477256354747706 × PC787 + 0.8291285293095355 × PC788 + 0.7483576579925603 × PC793 + -0.7630253694600663 × PC796 + 0.2662889119207861 × PC825 + 0.1944294516040025 × PC827 + -0.43075148701330634 × PC833 + -0.13861827355455808 × PC839 + -0.7660097387256596 × PC840 + 2.1468399451117497 × PC841 + 0.22208827917740712 × PC852 + -1.1263017128107338 × PC853 + 0.5395759230934849 × PC855 + -0.055047496656576836 × PC858 + 0.4746009047483465 × PC861 + 0.727812333165279 × PC869 + 0.7332653664174068 × PC872 + -0.07162633164808167 × PC884 + -0.2564332079372071 × PC895 + -0.2949691281516769 × PC907 + -1.248585480073261 × PC911 + 0.6449657750793661 × PC917 + -0.40078713383206815 × PC920 + -1.1268334219529523 × PC926 + 1.0915070426198166 × PC931 + 1.083010800096405 × PC932 + -0.7737637266089399 × PC938 + -0.7051697221078365 × PC939 + -0.777528237145316 × PC943 + -0.7462359238640877 × PC945 + 0.8977568450659577 × PC946 + -0.36927142620273223 × PC951 + 1.288393531445071 × PC954 + 0.41338675479417264 × PC958 + -0.7160881290870678 × PC960 + -1.0850421752046047 × PC966 + -1.1276781631498667 × PC984 + 1.5560859108543843 × PC985 + 0.5489022602350867 × PC989 + 0.9522291234799042 × PC991 + 0.38657900764674646 × PC999 + -0.9596543064643399 × PC1002 + -0.3364838375541013 × PC1004 + -0.9261406506756724 × PC1014 + 0.004850850632490142 × PC1019 + -0.9368397253737891 × PC1020 + 0.4311856159692088 × PC1022 + -0.034978083432234346 × PC1034 + 0.5065515864783159 × PC1037 + 0.509998066227387 × PC1040 + -1.0036453860621364 × PC1052 + -0.43552978104678497 × PC1053 + -0.6587070684070995 × PC1061 + -0.9070439636075442 × PC1063 + 0.27111327067319313 × PC1066 + -1.2428096532923278 × PC1079 + 1.1802440752127754 × PC1089 + -1 .3681067389391608 × PC1092 + 0.5691294460235092 × PC1093 + 0.6888208236311402 × PC1094 + 0.5351525725826619 × PC1096 + 0.16361619194666238 × PC1107 + -0.10091217718636686 × PC1109 + -0.6895355885810754 × PC1115 + 0.6290851347197604 × PC1120 + 0.2277741805270195 × PC1128 + -0.7142105340181559 × PC1129 + -0.9538842578812784 × PC1136 + 0.5135941521351525 × PC1138 + -0.077334629268024 × PC1146 + 0.6833189400403291 × PC1152 + -0.4181232304677069 × PC1154 + -0.16611707400787593 × PC1158 + 0.16506408038531634 × PC1166 + 0.854583547659567 × PC1174 + 0.645500712975673 × PC1190 + -0.33192560439834967 × PC1193 + 0.726612528242891 × PC1194 + 1.2731839707752768 × PC1195 + 0.8179743522893723 × PC1231 + 1.1139439561360072 × PC1236 + 0.956838211745448 × PC1237 + -0.7903569425378941 × PC1260 + 0.811273620190096 × PC1264 + 0.9297125265837636 × PC1266 + 0.9294579406612353 × PC1284 + 1.1314199586837579 × PC1291 + -0.5402235065820549 × PC1296 + 0.9125973765595541 × PC1299 + -1.0914021272901262 × PC1300 + -0.009236006311060573 × PC1304 + 0.8657738242067076 × PC1323 + 0.5484989676860571 × PC1325 + 0.07170986532512955 × PC1331 + 1.1169106186522377 × PC1335 + 0.27185880195508233 × PC1343 + 0.25663158249348056 × PC1349 + 0.2804467611705637 × PC1354 + 1.0758980934701274 × PC1356 + 1.0060141266536669 × PC1366 + -0.08519174399164461 × PC1375 + 1.0222214868427453 × PC1379 + 1.5277269313296606 × PC1387 + 1.249079354649137 × PC1393 + 0.5476999380113162 × PC1395 + -0.4346055180275326 × PC1398 + -1.5291718602907594 × PC1402 + -0.6128556930478282 × PC1404 + 0.6181301686951538 × PC1410 + 0.8598743485048197 × PC1419 + 1.2436130044912739 × PC1420 + -0.8187540738744669 × PC1424 + 1.2179441164392506 × PC1430 + -0.1485972032708283 × PC1431 + 1.287128844629768 × PC1437 + 1.187503064941656 × PC1441 + 0.920455956335615 × PC1443 + 1.1729389010087081 × PC1447 + -0.03738299954888227 × PC1448 + 1.080823743939649 × PC1469 + 1.0007074895865304 × PC1502 + -0.9316727792756393 × PC1516 + 0.8116122084918351 × PC1520 + 0.38223736773472616 × PC1526 + -0.2739934176707099 × PC1532 + -1.4357604920237925 × PC1560 + 0.9176226465987137 × PC1569 + 0.21006937649497365 × PC1574 + -0.39535537488566347 × PC1576 + -0.027122877762613525 × PC1578 + 0.4096389601593757 × PC1580 + -1.3478524889078587 × PC1581 + 0.8211505762245885 × PC1595 + -0.2636028431613542 × PC1607 + -1.2737956452419414 × PC1617 + -0.9954039586809647 × PC1619 + -0.109528048296303 × PC1628 + -1.9173924826321924 × PC1630 + 0.9223347863883072 × PC1631 + -0.5354419173009779 × PC1635 + 1.5033745718758684 × PC1646 + 0.3579636811845547 × PC1660 + 0.9794688026594365 × PC1662 + -0.6919804694077056 × PC1666 + -1.4926400772576154 × PC1684 + 0.3789519072136572 × PC1704 + 0.12494078894696939 × PC1705 + -0.6059020926719798 × PC1706 + -0.06626792154768317 × PC1709 + -0.8532032621128705 × PC1711 + 0.9302416984311742 × PC1718 + 0.7657827743678347 × PC1725 + -0.8041066409938532 × PC1728 + -0.8643082188698398 × PC1731 + -0.8197834810797171 × PC1739 + 0.0449826587435612 × PC1753 + 0.07384707918954987 × PC1755 + 0.4736401077901558 × PC1767 + 0.23148537750525588 × PC1768 + 1.4187548919669435 × PC1772 + -0.20169346563536997 × PC1780 + 1.1069258983037082 × PC1791 + 0.3902158015952793 × PC1811 + -1.0352109178872406 × PC1814 + 1.1513737225037501 × PC1829 + 0.9229891601486028 × PC1849 + -0.2888775963806889 × PC1865 + -0.9483321446271571 × PC1875 + -1.0438720429709318 × PC1883 + 0.8056834077923207 × PC1902 + -1.0095138885558557 × PC1905 + -0.23438112133941427 × PC1907 + 0.525380323112532 × PC1912 + -0.10869695697369229 × PC1942 + 0.9640216956111622 × PC1946 + -0.8404033108634823 × PC1948 + -0.8717303186194654 × PC1951 + 0.4954097695448641 × PC1966 + -0.9216382618161479 × PC1969 + 0.4333387217184146 × PC1975 + 0.8541053302013787 × PC1978 + -0.7250962572302478 × PC1981 + 0.33416958469014624 × PC1985 + -0.06663059357955031 × PC1990 + 1.393534688155795 × PC1994 + 0.9223060475672247 × PC1999 + 0.9684736251610293 × PC2002 + -1.1998418971065734 × PC2010 + 0.3527795603989891 × PC2019 + 0.059534478390552495 × PC2020 + -1.1479759898019002 × PC2024 + 0.9837800001406855 × PC2036 + -0.44486090162731146 × PC2040 + 1.085098737779112 × PC2042 + -2.153011287798299 × PC2050 + -0.7866929504727147 × PC2058 + -1.4009342905493338 × PC2063 + -1.1404828143094714 × PC2074 + -0.3387486608495367 × PC2080 + -0.9153070293472866 × PC2089 + -0.2743332475418231 × PC2102 + -0.3368350280922717 × PC2112 + 1.034698204763641 × PC2117 + 0.448542945963305 × PC2121 + -0.09689469615084463 × PC2128 + -1.226695613918115 × PC2146 + -0.2293857887601697 × PC2163 + 0.16508536093903348 × PC2178 + 0.4044252378821144 × PC2181 + 0.027680579696913954 × PC2183 + -1.3669801639253705 × PC2189 + 1.21440157710717 × PC2197 + -0.5627087931530748 × PC2199 + 0.2898935084421982 × PC2206 + 0.6480350550918204 × PC2211 + -1.0039347168923467 × PC2216 + -1.0884535927838033 × PC2247 + -0.5984977489328072 × PC2268 + -1.25655772795395 × PC2275 + -0.7555988385972258 × PC2277 + -1.5499887111213662 × PC2281 + 1.021634013230252 × PC2299 + -0.4473775217927492 × PC2308 + 1.825641605369938 × PC2311 + 0.8253696777815654 × PC2324 + 0.8942954853026011 × PC2325 + 0.07714910854789489 × PC2347 + -0.9449840823744458 × PC2352 + 0.8769381354327463 × PC2367 + -0.5143983627824752 × PC2383 + 1.064090237120295 × PC2384 + -0.4777215554864697 × PC2385 + -0.15372529606795687 × PC2395 + -0.12695387581618442 × PC2409 + -0.9083060293952133 × PC2418 + 0.7872794367191932 × PC2442 + 1.1613424169713065 × PC2449 + 0.7336569433503886 × PC2468 + -0.281640431190804 × PC2503 + -0.27008558494015444 × PC2505 + -1.4855428305666425 × PC2509 + -1.5997460791652578 × PC2525 + 0.7407938070293335 × PC2527 + -1.798666783220541 × PC2538 + 1.386124593927042 × PC2543 + -0.23268545892953454 × PC2544 + 0.9338161314564373 × PC2547 + 1.1148300311915262 × PC2567 + 0.9895715305891174 × PC2573 + -1.205308383115116 × PC2602 + 1.04793563166749 × PC2607 + 0.4103282736638466 × PC2621 + 0.31918621320557367 × PC2658 + 0.7420078326260372 × PC2666 + 1.0775006511175025 × PC2689 + -0.9105154377910384 × PC2692 + -0.6911297006906036 × PC2696 + -1.0998130926195295 × PC2701 + 0.7161610357949935 × PC2702 + -0.9632516431058665 × PC2713 + 1.577019288703735 × PC2714 + -1.1473177426464383 × PC2729 + -1.7410907465899623 × PC2732 + -0.9244095838953645 × PC2754 + -0.8064697297957332 × PC2757 + -0.9087663837857273 × PC2763 + 0.7284969513990861 × PC2766 + 1.1588477113459255 × PC2781 + 0.6787750064508451 × PC2786 + -1.4043642474903852 × PC2790 + 1.0136433163679233 × PC2807 + 0.9140097093583017 × PC2809 + 0.9910691344679617 × PC2819 + 0.777528745299505 × PC2832 + -0.9779803784430643 × PC2834 + 0.4819464864187081 × PC2847 + 1.2452240290682843 × PC2856 + -1.3984487303559097 × PC2866 + 0.3922897984402952 × PC2870 + -1.5248818066780159 × PC2890 + -2.1423970091661007 × PC2894 + 1.1414949543881834 × PC2900 + 0.026220295096397582 × PC2910 + -0.409197554580786 × PC2914 + 0.8987600960662196 × PC2915 + -0.07342779779216604 × PC2918 + -0.5396317041682942 × PC2931 + -0.38924398322408726 × PC2934 + 0.9192629059378943 × PC2937 + 1.2269944737246516 × PC2952 + 1.4346335575850688 × PC2973 + 0.763523917447537 × PC2975 + 0.9558069869168456 × PC2981 + -0.09093571042542274 × PC2986 + -0.7077248685590314 × PC2987 + 1.1147738196225725 × PC2990 + -0.7356841121750335 × PC2996 + -0.4808460417047035 × PC3009 + 1.0107771789961986 × PC3014 + -0.8855730537875612 × PC3015 + -0.0792560869231036 × PC3018 + 0.12207501450740138 × PC3025 + -0.07612853171631152 × PC3047 + -0.8659097219202504 × PC3059 + 1.126515870231891 × PC3060 + -0.35844885716181313 × PC3062 + -0.24310454357526137 × PC3066 + 0.9986486336925937 × PC3076 + 1.3532423677072543 × PC3107 + 0.9935982360948099 × PC3122 + 0.5126783716231297 × PC3142 + 1.0144166098042886 × PC3145 + 0.24568844784609514 × PC3171 + 0.7817898616028325 × PC3175 + 1.1976733861974849 × PC3179 + -0.823250494369886 × PC3185 + 0.2932359732911836 × PC3224 + -0.5836332915004158 × PC3241 + -1.1788110253562267 × PC3248 + 1.0497068560247718 × PC3249 + -1.5749789743127027 × PC3263 + 0.10986296627759312 × PC3296 + -0.9050658336675534 × PC3307 + 1.0183742456458227 × PC3312 + -1.617470677210034 × PC3351 + 0.38210253941813643 × PC3369 + -0.30143747969102663 × PC3376 + -1.5799213053655359 × PC3385 + 0.32960488908953656 × PC3386 + 1.2104809003078485 × PC3435 + -0.42413708532687855 × PC3444 + -1.8343198591413044 × PC3461 + 0.4531265069114384 × PC3483 + 1.6902428536786673 × PC3484 + -0.9534241804171446 × PC3489 + 0.7433453730673456 × PC3511 + -1.3339230823500168 × PC3513 + -0.24261025905235248 × PC3514 + -0.7237797906497998 × PC3534 + -0.7883646976268682 × PC3540 + 0.47535314587911603 × PC3554 + 0.8659507670383181 × PC3556 + -0.7362311561820232 × PC3613 + 1.1822889897097695 × PC3621 + 0.883529870809268 × PC3653 + -0.7337722550167723 × PC3662 + -0.11451656581951053 × PC3667 + 1.1281956448880923 × PC3692 + 1.4509361068184294 × PC3696 + -1.2953906415604857 × PC3699 + 0.8826339809178165 × PC3710 + 1.0487965004551434 × PC3721 + -1.1097150402853704 × PC3731 + 0.15155220073887737 × PC3766 + -1.585921817626694 0.780629504413845 -2.316493714082503×PC3877+ 0.882423887835842 × PC3883 + 0.2216111119894136 × PC3910 + 2.6205579059309776 × PC3929 + -1.2018058687436988 × PC3943 + 1.7274408308176914 × PC3952 + -1.6600405841389823 × PC3960 + 0.3963416610722133 × PC3961 + 0.7304732307810656 × PC3969 + -1.57084496903186 × PC3975 + 0.20166169993678915×PC4025+0.38422564494234407×PC4037。.

[0067] Using the above biological age model, the data of 50 Japanese people not included in the training data were used as evaluation data, and the root mean square error (RMSE) and mean absolute error (MSE) between the calculated biological age and chronological age were calculated as evaluation indices. Similarly, as a comparative example, using similar evaluation data, biological age was calculated by the methods described in Non-Patent Documents 2 and 3, and the root mean square error and mean absolute error between biological age and chronological age were calculated as evaluation indices. The results are shown in Table 4.

[0068]

[0069] As shown in Table 4, and similarly to Example 1, the biological age model calculated by the method of this embodiment has a smaller root mean square error and a smaller absolute mean error than the biological age models calculated by the methods of Non-Patent Documents 2 and 3. As is clear from this, a model with high prediction accuracy was constructed from a small amount of DNA methylation data, while reducing the number of samples required for model construction.

[0070] From the above results, it is clear that the biological age measurement device and biological age method according to the present embodiment and examples construct a model with higher prediction accuracy than existing technologies (Non-Patent Documents 1 to 3). As such, according to the present embodiment and examples, a model with high prediction accuracy can be constructed from a small amount of DNA methylation data while reducing the number of samples required for model construction.

[0071] The above describes one embodiment and example of the present invention, but the present invention is not limited to the above embodiment or example, and it is clear that the scope of the present invention can be modified and altered within a range that is obvious to those skilled in the art. It is also clear that the scope of the present invention is not limited to the above embodiment or example, but also includes modifications and alterations thereof.

[0072] 1. Biological age measuring device, 11. Transfer learning unit, 12. Memory unit, 13. Control unit, 14. Input unit, 15. Output unit.

Claims

1. A biological age measurement device that measures biological age based on DNA methylation levels, wherein first explanatory variable information based on DNA methylation information at q CpG sites contained in the DNA of each sample of a first population consisting of n samples is used as an explanatory variable matrix (n x p matrix (p ≤ q)), and first objective variable information based on the biological age information of the n samples is used as an objective variable vector (n dimensions), and transfer learning is performed based on a biological age measurement model estimated using a second population different from the first population, and the learning result is output as a trained biological age measurement model.

2. A biological age measurement device according to claim 1, wherein in the transfer learning, a regression coefficient vector β^ (p-dimensional) having an argument β (p-dimensional) that satisfies equation 1 is calculated as first regression coefficient information. Here, y is n-dimensional first dependent variable information, X is n×p first explanatory variable information, λ, α, and ρ are adjustment parameters, and β~ is p-dimensional second regression coefficient information in the biological age measurement model estimated by the second population.

3. The biological age measurement device according to claim 1, wherein the biological age measurement model estimated using the second population is a biological age model in which second explanatory variable information based on DNA methylation information at q CpG sites contained in DNA of each sample of the second population consisting of n' samples is an explanatory variable matrix (n' x p matrix (p≦q)), and second objective variable information based on the biological age information of the n' samples is an objective variable vector (n' dimension), wherein the explanatory variable vector for each sample of the second explanatory variable information has as its elements coordinate values ​​in a p-dimensional space whose axes are first to p-th principal components calculated after principal component analysis of the DNA methylation information at q CpG sites contained in DNA of each sample of the second population, and the explanatory variable vector for each sample of the first explanatory variable information has as its elements coordinate values ​​in a p-dimensional space whose axes are the first to p-th principal components.

4. A biological age measurement method for measuring biological age based on DNA methylation levels, comprising the steps of: defining first explanatory variable information based on DNA methylation information at q CpG sites contained in the DNA of each sample of a first population consisting of n samples as an explanatory variable matrix (n × p matrix (p≦q)); defining first objective variable information based on the biological age information of the n samples as an objective variable vector (n dimensions); performing transfer learning based on a biological age measurement model estimated using a second population different from the first population; and outputting the learning result as a trained biological age measurement model.

Images