Training device, estimation device, and program

Abstract

This training device comprises: a first acquisition unit that acquires feed composition information indicating the composition of nutrients in feed; a second acquisition unit that acquires blood amino acid composition information indicating the composition of the blood amino acid of an animal that was given the feed; and a training unit that generates a trained model that has been trained with the corresponding relationship between the composition of nutrients in the feed indicated by the feed composition information and the composition of the blood amino acid indicated by the blood amino acid composition information.

Description

Learning device, estimation device and program

[0001] This application claims priority to U.S. Provisional Patent Application No. 63 / 547,991, filed November 10, 2023, the contents of which are incorporated herein by reference.

[0002] It has become clear that eating a high-fat or low-protein diet can worsen the blood amino acid profile, which in turn can lead to the development of various lifestyle-related diseases such as ectopic fat accumulation and dyslipidemia. In other words, if a worsened blood amino acid profile can be improved to a healthy one, the onset of lifestyle-related diseases can be suppressed.

[0003] In fact, feeding a high-fat diet, which is known to cause various lifestyle-related diseases, deteriorates the blood amino acid profile, leading to the formation of fatty liver and an increase in the amount of non-HDL (high density lipoprotein) cholesterol, known as bad cholesterol. However, it is known that altering the amino acid composition of the high-fat diet and disrupting the blood amino acid profile can cure some lifestyle-related diseases (see, for example, Non-Patent Document 1). A low-arginine, high-fat diet (HFD (High Fat Diet) ΔArg) cures fatty liver. A low-methionine, high-fat diet (HFD ΔMet) and a low-lysine, high-fat diet (HFD ΔLys) suppress the rise in LDL (low density lipoprotein) cholesterol.

[0004] Hiroki Nishi, Yuki Goda, Ryosuke Okino, Ruri Iwai, Reona Maezawa, Koichi Ito, Shin-Ichiro Takahashi, Daisuke Yamanaka, and Fumihiko Hakuno, "Metabolic Effects of Short-Term High-Fat Intake Vary Depending on Dietary Amino Acid Composition," Current Developments in Nutrition 8 (2024) 103768

[0005] In rats fed a low-arginine diet, not only did blood arginine concentrations decrease, but methionine, histidine, and glutamine concentrations also increased. This shows that changing the concentration of one amino acid in the diet affects the blood concentrations of other amino acids, and that blood amino acid concentrations are complexly regulated. For example, it is easy to imagine that proposing a diet to improve a blood amino acid profile that has been deteriorated by a high-fat diet would be extremely complex and difficult.

[0006] An object of the present invention is to provide a learning device, an estimation device, and a program that can propose a nutrient composition in feed that will improve a deteriorated blood amino acid composition.

[0007] (1) One embodiment of the present invention is a learning device that includes a first acquisition unit that acquires feed composition information that indicates the composition of nutrients in feed, a second acquisition unit that acquires blood amino acid composition information that indicates the composition of amino acids in the blood of an animal that has been fed the feed, and a learning unit that generates a trained model that has learned the correspondence between the composition of nutrients in the feed that is indicated by the feed composition information and the composition of amino acids in the blood that is indicated by the blood amino acid composition information.

[0008] (2) In one embodiment of the present invention, in the aforementioned learning device, the learning unit generates a trained model by Lasso regression on the feed composition information and the blood amino acid composition information.

[0009] (3) In one embodiment of the present invention, in the aforementioned learning device, the learning unit generates a trained model by support vector regression on the feed composition information and the blood amino acid composition information.

[0010] (4) In one embodiment of the present invention, in the learning device described above, the nutrients include a plurality of types of amino acids, and the feed composition information includes at least an amino acid composition.

[0011] (5) One embodiment of the present invention is an estimation device that includes a blood amino acid composition acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed with feed, and a feed composition acquisition unit that acquires feed composition information indicating the composition of nutrients in the feed, which is output by a trained model trained by the learning device described in (1) above, by providing the blood amino acid composition information to the trained model.

[0012] (6) One embodiment of the present invention is an estimation device that includes a feed composition acquisition unit that acquires feed composition information indicating the composition of nutrients in feed, and a blood amino acid composition acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal that has been fed the feed, which is output by a trained model trained by the learning device described in (1) above when the trained model is provided with the feed composition information.

[0013] (7) One embodiment of the present invention is an estimation device that includes an estimation unit that estimates the nutrient composition in feed to improve the disease based on a trained model trained by the learning device described in (1) above, disease information indicating the correspondence between the composition of amino acids in the blood and the state of the disease, and information indicating the composition of amino acids in the blood to improve the disease.

[0014] (8) One embodiment of the present invention is an estimation device described in any one of (5) to (7) above, in which the nutrients include multiple types of amino acids and the feed composition information includes at least an amino acid composition.

[0015] (9) One embodiment of the present invention is a program for causing a computer to perform the following steps: acquire feed composition information indicating the composition of nutrients in feed; acquire blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed the feed; and generate a trained model that has learned the correspondence between the composition of nutrients in the feed indicated by the feed composition information and the composition of amino acids in the blood indicated by the blood amino acid composition information.

[0016] (10) One embodiment of the present invention is a program for causing a computer to perform the following steps: obtain blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed with feed; provide the blood amino acid composition information to a trained model that has learned the correspondence between the composition of nutrients in feed and the composition of amino acids in the blood of the animal fed with the feed; and obtain feed composition information output by the trained model that indicates the composition of nutrients in the feed.

[0017] According to an embodiment of the present invention, it is possible to propose a nutrient composition in feed that improves a deteriorated blood amino acid composition.

[0018] 1 is a diagram showing an example of an analysis result of amino acids in the blood of an animal. FIG. 1 is a diagram showing an example of an analysis result of amino acids in the blood of an animal. FIG. 1 is a diagram for explaining an example of the operation of a learning device according to the present embodiment. FIG. 1 is a diagram showing an example of the configuration of a learning device according to the present embodiment. FIG. 2 is a flowchart showing an example of the operation of a learning device according to the present embodiment. FIG. 2 is an image diagram showing an example of a regression equation for multiple regression analysis. FIG. 3 is a diagram showing an example of the verification result of a regression equation created using Lasso regression. FIG. 4 is a diagram showing Comparative Example 1 for each amino acid. FIG. 5 is a diagram showing Comparative Example 2 for each amino acid. FIG. 6 is a diagram showing Comparative Example 3 for each amino acid. FIG. 7 is a diagram showing an example of the verification result of a regression equation created using support vector regression. FIG. 8 is a diagram showing Comparative Example 1 for each amino acid. FIG. 9 is a diagram showing Comparative Example 2 for each amino acid. FIG. 10 is a diagram showing Comparative Example 3 for each amino acid. FIG. 11 is a diagram showing an example of an estimation device according to the present embodiment. A flowchart showing an example of the operation of an estimation device according to the present embodiment. FIG. 12 is a diagram showing an example of the operation of an estimation device according to a modified example of an embodiment. FIG. 13 is a diagram showing an example of the operation of an estimation device according to a modified example of an embodiment.

[0019] Next, the learning device, estimation device, and program of this embodiment will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiment to which the present invention is applied is not limited to the following embodiment. In all drawings used to explain the embodiment, the same reference numerals are used for components having the same functions, and repeated explanations are omitted. Furthermore, "based on XX" in this application means "based on at least XX" and includes cases where the element is based on XX and other elements. Furthermore, "based on XX" is not limited to cases where XX is directly used, but also includes cases where the element is based on XX after calculation or processing. "XX" is any element (for example, any information).

[0020] (Embodiment) FIG. 1 is a diagram showing an example of the results of analyzing amino acids in the blood of an animal. Referring to FIG. 1, the effects of low protein nutrition on growth and lipid accumulation in growing animals are shown. The animal may be an animal classified as a mammal. The animals may or may not include humans. Here, rats are shown as an example of animals. The composition of multiple types of amino acids contained in the blood of rats in a normal physiological state and rats with accumulated fat was analyzed. The liver and muscle of rats with accumulated fat were analyzed.

[0021] The various amino acids are Asp (aspartic acid), Thr (threonine), Ser (serine), Gly (glycine), Ala (alanine), Val (valine), Met (methionine), Ile (isoleucine), Leu (leucine), Tyr (tyrosine), Phe (phenylalanine), Lys (lysine), His (histidine), and Arg (arginine). Asp (aspartic acid) is an acidic amino acid. Arg (arginine), His (histidine), and Lys (lysine) are basic amino acids.

[0022] In Figure 1, as shown in (1), when normal rats are normalized so that the values ​​of multiple amino acids are each 1 (in reality, each amino acid has a range from one digit to several hundred μM), as shown in (2), the profile of a rat with accumulated fat shows excess and deficient amino acids. Specifically, a profile of multiple amino acid composition in which Arg (arginine) is deficient (half that of normal rats) is obtained for the liver, and a profile of multiple amino acid composition in which Lys (lysine) is deficient is obtained for the muscle. The above results are not limited to rats; similar results can be obtained when analyzing blood amino acids in humans.

[0023] FIG. 2 shows an example of the results of analyzing amino acids in an animal's blood. To return a body to a normal physiological state from a state of accumulated fat, the amino acid composition of nutrients in the diet is derived from the composition profile of multiple types of amino acids contained in the blood amino acids in a normal physiological state. For example, the amino acid composition of nutrients in the diet is calculated backward from the composition profile of multiple types of amino acids contained in the blood amino acids in a normal physiological state. This allows the amino acid composition in the diet to be predicted, thereby enabling flexible control of the physiological state.

[0024] FIG. 3 is a diagram illustrating an example of the operation of a learning device according to this embodiment. The learning device according to this embodiment creates a learning model for predicting the amino acid composition in a diet that will induce a specific amino acid in the blood. Specifically, an appropriate amino acid composition for the amino acids in the diet is predicted from a composition profile of multiple types of amino acids contained in the blood shown in (1). The amino acid composition in the diet is predicted from a composition profile of multiple types of amino acids contained in the blood that are deficient in arginine shown in (2)-1. The amino acid composition in the diet is predicted from a composition profile of multiple types of amino acids contained in the blood that are deficient in lysine shown in (2)-2. The learning device that creates the learning model will be described in detail below.

[0025] (Learning Device) FIG. 4 is a diagram illustrating an example configuration of a learning device according to an embodiment of the present invention. The learning device 100 creates a trained model. The learning device 100 is implemented by a device such as a personal computer, a server, a smartphone, a tablet computer, or an industrial computer. The learning device 100 acquires feed composition information indicating the composition of nutrients in feed. The learning device 100 acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed that feed. The learning device 100 trains a learning model using multiple training datasets including the acquired blood amino acid composition information and the acquired feed composition information, thereby creating a trained model. As an example, the following description will be given of a case in which the learning device 100 trains a learning model using multiple training datasets in which the acquired blood amino acid composition information is used as an input sample and the acquired feed composition information is used as an output sample, thereby creating a trained model.

[0026] For example, the learning device 100 constructs a trained model using algorithms such as least absolute shrinkage and selection operator (LASSO) regression, support vector (SV) regression, convolution neural network (CNN), recurrent neural network (RNN), long short-term memory (LSTM), random forest, neural network, multilayer perceptron (MLP), and deep learning. An input sample is data input to an input layer during training of a learning model. An output sample is data (teacher data) that serves as a correct answer for comparison with an output value output from an output layer during training of a learning model.

[0027] The learning device 100 includes a first acquisition unit 101, a second acquisition unit 102, a learning unit 103, an output unit 105, and a memory unit 110. The first acquisition unit 101 acquires the nutrient composition of the feed indicated by the feed composition information. The nutrients include multiple types of amino acids. The feed composition information includes at least the amino acid composition. The multiple types of amino acids include at least one of Asp (aspartic acid), Thr (threonine), Ser (serine), Gly (glycine), Ala (alanine), Val (valine), Met (methionine), Ile (isoleucine), Leu (leucine), Tyr (tyrosine), Phe (phenylalanine), Lys (lysine), His (histidine), Asn (asparagine), Glu (glutamic acid), Pro (proline), Trp (tryptophan), Gln (glutamine), Cys (cysteine), Cys2 (cystine), and Arg (arginine). Asp (aspartic acid) and Gln (glutamine) are acidic amino acids. Arg (arginine), His (histidine), and Lys (lysine) are basic amino acids.

[0028] The second acquisition unit 102 acquires the blood amino acid composition indicated by the blood amino acid composition information. The blood amino acids include at least one of Asp (aspartic acid), Thr (threonine), Ser (serine), Gly (glycine), Ala (alanine), Val (valine), Met (methionine), Ile (isoleucine), Leu (leucine), Tyr (tyrosine), Phe (phenylalanine), Lys (lysine), His (histidine), Asn (asparagine), Glu (glutamic acid), Pro (proline), Trp (tryptophan), Gln (glutamine), Cys (cysteine), Cys2 (cystine), and Arg (arginine).

[0029] The first acquisition unit 101 and the second acquisition unit 102 may be configured to include an input unit. The input unit inputs information. As an example, the input unit may have an operation unit such as a keyboard and a mouse. In this case, the input unit inputs information according to an operation performed by a user on the operation unit. As another example, the input unit may input information from an external device. The external device may be, for example, a portable storage medium.

[0030] The learning unit 103 acquires feed composition information from the first acquisition unit 101 and acquires blood amino acid composition information from the second acquisition unit 102. The learning unit 103 accepts the acquired feed composition information and blood amino acid composition information as a training dataset. The learning unit 103 creates pairs in which the blood amino acid composition information included in the training dataset is used as an input sample and the feed composition information is used as an output sample. In other words, the learning unit 103 creates a plurality of pairs including an input sample and an output sample.

[0031] The learning unit 103 inputs an input sample into the input layer of the learning model 104 for all pairs, calculates the error between the output value output from the output layer and the output sample (teacher data) corresponding to the input sample, and changes the parameters of the learning model 104 (trains the learning model 104) so ​​as to minimize the error, thereby creating a learned model. In other words, the learning unit 103 uses blood amino acid composition information as an explanatory variable and feed composition information as a target variable, and performs machine learning on the relationship between blood amino acid composition information and feed composition information. The created learned model is output from the output unit 105. The memory unit 110 is realized by a hard disk drive (HDD), flash memory, random access memory (RAM), read-only memory (ROM), etc., and stores information.

[0032] All or part of the first acquisition unit 101, the second acquisition unit 102, the learning unit 103, and the output unit 105 are functional units (hereinafter referred to as software functional units) realized by a processor such as a CPU (Central Processing Unit) executing a program stored in the storage unit 110. Note that all or part of the first acquisition unit 101, the second acquisition unit 102, the learning unit 103, and the output unit 105 may be realized by hardware such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), or an FPGA (Field-Programmable Gate Array), or may be realized by a combination of a software functional unit and hardware.

[0033] (Operation of the learning device 100) Figure 5 is a flowchart showing an example of the operation of the learning device of this embodiment. (Step S1-1) The first acquisition unit 101 acquires one or more pieces of feed composition information indicating the composition of nutrients in the feed. (Step S2-1) The second acquisition unit 102 acquires one or more pieces of blood amino acid composition information indicating the composition of amino acids in the blood of the animal that has been fed the feed.

[0034] (Step S3-1) The learning unit 103 acquires one or more pieces of feed composition information from the first acquisition unit 101, and acquires one or more pieces of blood amino acid composition information from the second acquisition unit 102. Based on the acquired one or more pieces of feed composition information and blood amino acid composition information, the learning unit 103 associates the nutrient composition in the feed indicated by the feed composition information with the blood amino acid composition of the animal that was fed that feed indicated by the blood amino acid composition information, and accepts the association as a learning dataset.

[0035] The learning unit 103 creates one or more pairs of input samples, each pair using blood amino acid composition information included in the training dataset and feed composition information as an output sample. For all pairs of input samples and output samples included in each of the one or more training datasets, the learning unit 103 inputs the input sample into the input layer of the training model 104, calculates the error between the output value output from the output layer and the output sample (teacher data) corresponding to the input sample, and changes the parameters of the training model 104 so as to minimize the error (trains the training model 104). (Step S4-1) The output unit 105 acquires the training model 104 from the training unit 103. The output unit 105 outputs the acquired training model 104 as a trained model. The trained model output by the output unit 105 is output to, for example, the estimation device 200 described below. The output unit 105 may also output the trained model to the storage unit 110, thereby storing the trained model in the storage unit 110.

[0036] In the above-described embodiment, as an example, the learning device 100 trains a learning model using one or more learning datasets in which the acquired blood amino acid composition information is used as an input sample and the acquired feed composition information is used as an output sample, and creates a trained model. However, this example is not limited to this. For example, the learning device 100 may train a learning model using one or more learning datasets in which the acquired feed composition information is used as an input sample and the acquired blood amino acid composition information is used as an output sample, and create a trained model.

[0037] (Specific example of processing by the learning device 100) A specific example of processing by the learning unit 103 of the learning device 100 of this embodiment will be described. As an example, a case where the learning unit 103 constructs a learning model using Lasso regression and a case where the learning model is constructed using SV regression will be described. A case where the learning unit 103 constructs a learning model using Lasso regression will be described. Lasso regression is a type of linear multiple regression analysis. FIG. 6 is an image diagram showing an example of a regression equation for multiple regression analysis. The intake of an amino acid (y) is represented by the amount of amino acid in blood (x) and a constant term. In FIG. 6, the intake is represented by "Intake". "Intake_Ile" is the intake of Ile (isoleucine), "Intake_Leu" is the intake of Leu (leucine), and "Intake_Lys" is the intake of Lys (lysine). The same applies to other amino acids.

[0038] When one or more amino acids contained in blood amino acids are input, the intake amount of one type of amino acid is calculated. Here, an example of blood amino acids is represented by a vector of 20 variables. A similar equation is created for each of the 20 types of ingested amino acids. This allows a regression equation to be created for calculating the intake amount of amino acids from blood amino acids.

[0039] The dataset used in this study is described below. Rats were fed a low-amino acid (5AA) diet supplemented with only one amino acid for one week. Specifically, three rats were fed each of a control amino acid mixed diet (CN diet) containing amino acids equivalent to 15% casein, 5AA, and 5AA supplemented with each of 20 ingested amino acids (5AA + Ile, 5AA + Leu, etc.). A total of 66 rats were fed.

[0040] One amino acid was reduced to one-third the amount in the CN diet and fed to rats for one week. Specifically, CN, 5AA, and CN were fed three of each of the 20 amino acids (ΔIle, Δleu, ...) reduced to one-third the amount in each diet. A total of 66 rats were fed. Data were obtained for a total of 132 rats.

[0041] The blood amino acid amounts and the amino acid composition in the feed obtained from these experiments were used for machine learning. The blood amino acid amount (x) was taken as the rate of change from CN. The rate of change from CN is expressed by formula (1): Rate of change from CN = (amino acid amount - average CN) / average CN (1) The same formula (1) applies to the amount of amino acid in the feed (y).

[0042] Using Lasso regression, two-thirds of the data set was used for machine learning to create a regression equation, and the remaining one-third was used to verify accuracy. Verification was performed by deriving the coefficient of determination R2. The coefficient of determination R2 is shown in Equation (2). The maximum value of the coefficient of determination R2 is 1, and it may be a negative number. In general, the coefficient of determination R2 is preferably 0.6 or more.

[0043] FIG. 7 shows an example of the results of verifying a regression equation created using Lasso regression. In addition to the case where blood amino acid levels and the amino acid composition in the feed were used for machine learning, for comparison, the case where blood amino acid levels obtained in a similar experiment and one week's worth of ingested amino acids were used for machine learning is also shown. In FIG. 7, (1) shows the coefficient of determination R2 (Lasso_Train) when the blood amino acid levels and the amino acid composition in the feed were used for machine learning, and the coefficient of determination R2 (Lasso_Test) when the regression equation was created, and (2) shows the coefficient of determination R2 (Lasso_Train) when the blood amino acid levels and one week's worth of ingested amino acids were used for machine learning, and the coefficient of determination R2 (Lasso_Test) when the regression equation was created, and the coefficient of determination R2 (Lasso_Test) when the regression equation was verified.

[0044] 7, the coefficient of determination R2 when creating the regression equation and the coefficient of determination R2 when verifying the regression equation are shown to be higher when the blood amino acid levels and the amino acid composition in the diet are used for machine learning than when the blood amino acid levels and a week's worth of ingested amino acids are used for machine learning. When the blood amino acid levels and the amino acid composition in the diet are used for machine learning, the coefficient of determination R2 values ​​when verifying the regression equation are 0.6 or higher, except for Ser (serine) and Gly (glycine).

[0045] Next, comparative examples for each amino acid will be described with reference to Figures 8A to 8C. Figure 8A is a diagram showing Comparative Example 1 for each amino acid. Figure 8A shows Thr (threonine) as an example of an amino acid. In Figure 8A, (1) shows a case where blood amino acid levels and dietary amino acid composition are used for machine learning, and (2) shows a case where blood amino acid levels and a week's worth of ingested amino acids are used for machine learning. In (1), the left figure shows data when creating a regression equation, and the right figure shows data when verifying the regression equation. The coefficient of determination R2 when creating the regression equation is 0.918, and the coefficient of determination R2 when verifying the regression equation is 0.878. In (2), the left figure shows data when creating the regression equation, and the right figure shows data when verifying the regression equation. The coefficient of determination R2 when creating the regression equation is 0.876, and the coefficient of determination R2 when verifying the regression equation is 0.748.

[0046] FIG. 8B shows Comparative Example 2 for each amino acid. FIG. 8B shows Asn (asparagine) as an example of an amino acid. In FIG. 8B, (1) shows a case where blood amino acid levels and feed amino acid composition were used for machine learning, and (2) shows a case where blood amino acid levels and a week's worth of ingested amino acids were used for machine learning. In (1), the left figure shows data when creating a regression equation, and the right figure shows data when verifying the regression equation. The coefficient of determination R2 when creating the regression equation was 0.814, and the coefficient of determination R2 when verifying the regression equation was 0.707. In (2), the left figure shows data when creating the regression equation, and the right figure shows data when verifying the regression equation. The coefficient of determination R2 when creating the regression equation was 0.733, and the coefficient of determination R2 when verifying the regression equation was 0.557.

[0047] FIG. 8C shows Comparative Example 3 for each amino acid. FIG. 8C shows Ser (serine) as an example of an amino acid. In FIG. 8C, (1) shows a case where the amount of amino acids in the blood and the composition of amino acids in the feed were used for machine learning, and (2) shows a case where the amount of amino acids in the blood and a week's worth of ingested amino acids were used for machine learning. In (1), the left figure shows data when creating a regression equation, and the right figure shows data when verifying the regression equation. The coefficient of determination R2 when creating the regression equation was 0.741, and the coefficient of determination R2 when verifying the regression equation was 0.570.

[0048] In (2), the left figure shows the data when the regression equation was being created, and the right figure shows the data when the regression equation was being verified. The coefficient of determination R2 when the regression equation was being created was 0.624, and the coefficient of determination R2 when the regression equation was being verified was 0.419.

[0049] Next, a case where the learning unit 103 constructs a learning model using SV regression (SVR) will be described. SVR is a type of machine learning that can create linear / nonlinear models. Its performance is reasonable and it is effective even with a small number of samples, for example, less than 1,000. Below, a case where a nonlinear model is created will be described as an example.

[0050] The process of selecting features in support vector regression is explained below. (1) Create a support vector regression model. (2) Keep one feature (x, amino acid) whose importance you want to check, and replace the others with the average value. (3) Calculate the predicted value (predicted value of the objective variable y) using the model created in (1). (4) Perform simple regression with the actual y on the horizontal axis and the predicted value y calculated in (3) on the vertical axis. (5) The slope of the simple regression line in (4) is taken as the sensitivity. (6) Remove the feature (amino acid) with the lowest absolute value of sensitivity. Steps (1) to (6) are then repeated.

[0051] FIG. 9 is a diagram showing an example of the verification results of a regression formula created using support vector regression. In addition to the verification results of the regression formula created using support vector regression, an example of the verification results of a regression formula created using Lasso regression is also shown for comparison. In FIG. 9, (1) shows the coefficient of determination R2 (SVR_rbf_train) when creating the regression formula using SVR and the coefficient of determination R2 (SVR_rbf_test) when verifying the regression formula. (2) shows the coefficient of determination R2 (Lasso_train) when creating the regression formula using Lasso regression and the coefficient of determination R2 (Lasso_test) when verifying the regression formula. FIG. 9 shows that the verification results of the regression formula created using SVR have a higher coefficient of determination R2 value than the verification results of the regression formula created using Lasso regression, in terms of the coefficient of determination R2 when creating the regression formula and the coefficient of determination R2 when verifying the regression formula. The results of verifying the regression equation created using SVR show that the coefficient of determination R2 when creating the regression equation and the coefficient of determination R2 when verifying the regression equation are both 0.6 or greater.

[0052] Next, comparative examples for each amino acid will be described with reference to FIGS. 10A to 10C. FIG. 10A shows Comparative Example 1 for each amino acid. FIG. 10A shows Thr (threonine) as an example of an amino acid. In FIG. 10A, (1) shows the case of a regression equation created using SVR, and (2) shows the case of a regression equation created using Lasso regression. In (1), the left figure shows data when the regression equation was created, and the right figure shows data when the regression equation was verified. The coefficient of determination R2 when the regression equation was created was 1.000, and the coefficient of determination R2 when the regression equation was verified was 0.992. In (2), the left figure shows data when the regression equation was created, and the right figure shows data when the regression equation was verified. The coefficient of determination R2 when the regression equation was created was 0.918, and the coefficient of determination R2 when the regression equation was verified was 0.878.

[0053] FIG. 10B shows Comparative Example 2 for each amino acid. FIG. 10B shows Ser (serine) as an example of an amino acid. In FIG. 10B, (1) shows the case of a regression equation created using SVR, and (2) shows the case of a regression equation created using Lasso regression. In (1), the left figure shows data when the regression equation was created, and the right figure shows data when the regression equation was verified. The coefficient of determination R2 when the regression equation was created was 0.992, and the coefficient of determination R2 when the regression equation was verified was 0.892. In (2), the left figure shows data when the regression equation was created, and the right figure shows data when the regression equation was verified. The coefficient of determination R2 when the regression equation was created was 0.741, and the coefficient of determination R2 when the regression equation was verified was 0.570.

[0054] FIG. 10C shows Comparative Example 3 for each amino acid. FIG. 10C shows Pro (proline) as an example of an amino acid. In FIG. 10C, (1) shows the case of a regression equation created using SVR, and (2) shows the case of a regression equation created using Lasso regression. In (1), the left figure shows data when the regression equation was created, and the right figure shows data when the regression equation was verified. The coefficient of determination R2 when the regression equation was created was 0.941, and the coefficient of determination R2 when the regression equation was verified was 0.672.

[0055] In (2), the left figure shows the data when the regression equation was being created, and the right figure shows the data when the regression equation was being verified. The coefficient of determination R2 when the regression equation was being created was 0.860, and the coefficient of determination R2 when the regression equation was being verified was 0.611.

[0056] (Estimation Device) Fig. 11 is a diagram showing an example of an estimation device according to this embodiment. The estimation device 200 according to this embodiment receives animal-related information. The animal-related information includes animal identification information and blood amino acid composition information indicating the composition of amino acids in the animal's blood. The estimation device 200 acquires the blood amino acid composition information included in the animal-related information, and acquires feed composition information indicating the composition of nutrients in feed based on the acquired blood amino acid composition information and a trained model.

[0057] The trained model is a machine-learned model of the relationship between feed composition information, which indicates the composition of nutrients in the feed, and blood amino acid composition information, which indicates the composition of amino acids in the blood of the animal that has been fed that feed. The estimation device 200 outputs the animal identification information and the acquired blood amino acid composition information.

[0058] The estimation device 200 is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, an industrial computer, etc. The estimation device 200 includes an input unit 201, a blood amino acid composition acquisition unit 202, a feed composition acquisition unit 203, an output unit 205, and a memory unit 210.

[0059] The input unit 201 inputs information. As an example, the input unit 201 may have an operation unit such as a keyboard and a mouse. In this case, the input unit 201 inputs information according to an operation performed by a user on the operation unit. As another example, the input unit 201 may input information from an external device. The external device may be, for example, a portable storage medium. Animal-related information is input to the input unit 201.

[0060] The blood amino acid composition acquisition unit 202 acquires animal-related information from the input unit 201. The blood amino acid composition acquisition unit 202 acquires animal identification information and blood amino acid composition information included in the acquired animal-related information, and accepts the acquired animal identification information and blood amino acid composition information.

[0061] The blood amino acid composition information indicates the composition of amino acids in the blood of the animal fed the feed, including at least one of Asp (aspartic acid), Thr (threonine), Ser (serine), Gly (glycine), Ala (alanine), Val (valine), Met (methionine), Ile (isoleucine), Leu (leucine), Tyr (tyrosine), Phe (phenylalanine), Lys (lysine), His (histidine), Asn (asparagine), Glu (glutamic acid), Pro (proline), Trp (tryptophan), Gln (glutamine), Cys (cysteine), Cys2 (cystine), and Arg (arginine).

[0062] The feed composition acquisition unit 203 acquires animal identification information and blood amino acid composition information from the blood amino acid composition acquisition unit 202. The feed composition acquisition unit 203 is equipped with a trained model 204. The trained model 204 is a machine-learned version of the correspondence between the nutrient composition in the feed indicated by the feed composition information and the amino acid composition in the blood indicated by the blood amino acid composition information. An example of the trained model 204 is created by the learning device 100. In this case, the trained model output by the learning device 100 is input to the estimation device 200 via a network or a medium, and acquired by the feed composition acquisition unit 203. Note that the estimation device 200 may include the learning device 100. In other words, the estimation device 200 may create the trained model 204. In this case, the feed composition acquisition unit 203 acquires the trained model 204 from the learning device 100. The feed composition acquisition unit 203 inputs the acquired blood amino acid composition information into the trained model 204, and acquires feed composition information indicating the composition of nutrients in the feed output by the trained model 204 for the input blood amino acid composition information. The feed composition information includes at least the amino acid composition.

[0063] The multiple types of amino acids include at least one of Asp (aspartic acid), Thr (threonine), Ser (serine), Gly (glycine), Ala (alanine), Val (valine), Met (methionine), Ile (isoleucine), Leu (leucine), Tyr (tyrosine), Phe (phenylalanine), Lys (lysine), His (histidine), Asn (asparagine), Glu (glutamic acid), Pro (proline), Trp (tryptophan), Gln (glutamine), Cys (cysteine), Cys2 (cystine), and Arg (arginine).

[0064] The output unit 205 acquires the animal identification information and the feed composition information from the feed composition acquisition unit 203. The output unit 205 outputs the acquired animal identification information and the feed composition information. For example, the output unit 205 may output the animal identification information and the feed composition information by voice, or may output the animal identification information and the feed composition information by displaying them on a display unit (not shown). The output unit 205 may also associate the animal identification information and the feed composition information and store them in the memory unit 210. The memory unit 210 is realized by a HDD, flash memory, RAM, ROM, etc., and stores information.

[0065] All or part of the input unit 201, blood amino acid composition acquisition unit 202, feed composition acquisition unit 203, and output unit 205 are functional units (hereinafter referred to as software functional units) that are realized, for example, by a processor such as a CPU executing a program stored in the storage unit 210. Note that all or part of the input unit 201, blood amino acid composition acquisition unit 202, feed composition acquisition unit 203, and output unit 205 may be realized by hardware such as an LSI, an ASIC, or an FPGA, or may be realized by a combination of software functional units and hardware.

[0066] (Operation of the estimation device 200) Figure 12 is a flowchart showing an example of the operation of the estimation device of the embodiment. With reference to Figure 12, the operation of the estimation device 200 after the feed composition acquisition unit 203 acquires the trained model 204 will be described. (Step S1-2) Animal-related information is input to the input unit 201. (Step S2-2) The blood amino acid composition acquisition unit 202 acquires the animal-related information from the input unit 201. The blood amino acid composition acquisition unit 202 acquires the animal identification information and blood amino acid composition information included in the acquired animal-related information, and accepts the acquired animal identification information and blood amino acid composition information.

[0067] (Step S3-2) The feed composition acquisition unit 203 acquires animal identification information and blood amino acid composition information from the blood amino acid composition acquisition unit 202. The feed composition acquisition unit 203 inputs the acquired blood amino acid composition information into the trained model 204, and acquires feed composition information indicating the composition of nutrients in the feed output by the trained model 204 for the input blood amino acid composition information. (Step S4-2) The output unit 205 acquires the animal identification information and feed composition information from the feed composition acquisition unit 203. The output unit 205 outputs the acquired animal identification information and feed composition information.

[0068] In the above-described embodiment, the estimation device 200 acquires animal identification information included in the animal-related information and blood amino acid composition information indicating the composition of amino acids in the animal's blood. The estimation device 200 then acquires feed composition information indicating the composition of nutrients in the feed based on the acquired blood amino acid composition information and the trained model. However, this example is not limiting. For example, the estimation device 200 may acquire animal identification information included in the animal-related information and feed composition information indicating the composition of nutrients in the feed, and then acquire the blood amino acid composition information based on the acquired feed composition information and the trained model. In the above-described embodiment, as an example, a case in which feature selection processing is performed in support vector regression is described. However, this example is not limiting. The feature selection processing may be omitted in support vector regression. Performing feature selection processing in support vector regression can improve accuracy.

[0069] According to the estimation device 200 of the embodiment, by using at least one of Lasso regression and support vector regression with blood amino acid composition information as an explanatory variable and feed composition information as a response variable, it is possible to calculate a feed (diet) composition that will modify the blood amino acid profile to a target profile, thereby making it possible to propose an amino acid composition of a feed (diet) that will improve a deteriorated blood amino acid profile.

[0070] (Modifications) Modifications of the embodiment will be described. As an example, the cases where a low-arginine diet, a low-methionine diet, and a low-lysine diet are fed will be described. The case where a low-arginine diet is fed will be described. When CN and low-arginine are fed, fatty liver is induced and blood LDL is lowered. When a high-fat diet is fed, fatty liver is suppressed and blood LDL does not change. The case where a low-methionine diet is fed will be described. When CN is fed, normal liver is induced and blood LDL does not change. When a low-arginine diet is fed, fatty liver is suppressed and blood LDL does not change. When a high-fat diet is fed, fatty liver is induced and blood LDL is lowered. The case where a low-lysine diet is fed will be described. When CN is fed, normal liver is induced and blood LDL does not change. When a high-fat diet is fed, hyperfatty liver is induced and blood LDL is lowered.

[0071] From the above, it can be seen that the optimal method for modifying the blood amino acid profile (intervention method) should differ depending on the health condition of the animal. This shows that personalized nutritional medicine (Precision Nutrition) is very important. In the modified embodiment, the explanation will continue with the case where a human (subject) is applied as an example of an animal.

[0072] In a modified embodiment, subjects are clustered based on their blood amino acid profiles using unsupervised machine learning known as self-organizing map (SOM) analysis. The SOM classifies subjects based on the patterns of 20 types of amino acids in their blood, and displays each individual's gender, obesity (obesity level), blood triglyceride (TG), fatty liver, blood LDL, and blood HDL in a visualized graph such as a heat map. Based on this SOM, a diet is proposed to alter the blood amino acid profile to a healthy one.

[0073] 13A is a diagram showing an example of the operation of the estimation device according to the modified example of the embodiment. FIG. 13A is a diagram showing an example of a self-organizing map. The self-organizing map includes an input layer and a competitive layer. The input layer is composed of one neuron. The input vectors are a combination of blood amino acid concentrations (xclass) and other disease factor values ​​(x １ class, x ２ class,...,x Ｊ class) (J is an integer greater than 0), and is expressed by formula (3), where J represents a unit included in the competitive layer. The competitive layer consists of many units. The weight vector (W ｍ ) is expressed by equation (4).

[0074] The input layer and the competitive layer are fully connected and have no connection weights. Furthermore, there are no connections between the competitive layers, and they are adjacent to each other. The neurons in the competitive layer have weight vectors, which are updated to approach the input vector presented. The distance between the input vector and the weight vector of the unit is calculated. Only the distance of the amino acid concentration is used, and other values ​​are not taken into account. Distance (Distance ｉｍ ) is expressed by equation (5). The unit with the smallest distance to the input vector is selected as the winning unit. ｉ，ｗｉｎｎｅｒ is expressed as in equation (6). As shown in equation (7), the winning unit and its neighboring units are changed.

[0075] FIG. 13B is a diagram illustrating an example of the operation of an estimation device according to a modified embodiment. FIG. 13B shows an example of the results of clustering blood amino acid profile data from health checkups of multiple subjects using SOM analysis. Physiological activity indices (disease indices) not used in the clustering are shown in a heat map. Disease indices include gender, obesity, TG, fatty liver, LDL, HDL, etc. FIG. 13B shows, as an example, heat maps for disease index 1, disease index 2, disease index 3, disease index 4, disease index 5, and disease index 6. The heat maps show the correspondence between blood amino acid composition and disease state. This allows users to see at a glance on the same map what diseases subjects with similar blood amino acid profiles have.

[0076] For example, in a heat map of disease index 5, the blood amino acid composition information of a subject is "A," and since the disease index 5 is high at "A," we will explain the case where we want to lower it. The arrow indicates the direction from "A" to a disease index 5 lower than "A." The arrow direction is preferably a direction that is expected to improve based on disease index 5 as well as at least one of disease index 2, disease index 3, disease index 4, and disease index 6. Even if deterioration is expected, a direction that minimizes the deterioration is preferable. For example, "A" is changed to "B," which has a lower disease index 5 than "A." "B" corresponds to disease-improving blood amino acid composition information that indicates the composition of blood amino acids for improving the disease.

[0077] The estimation device derives the difference vector between the starting point "A" and the end point "B" of the arrow. The estimation device estimates disease-improving blood amino acid composition information based on the blood amino acid composition information and the derived difference vector. The estimation device inputs the disease-improving blood amino acid composition information as an explanatory variable into the trained model, and obtains feed composition information indicating the composition of nutrients in the feed output by the trained model for the input explanatory variables. This allows the amino acid composition of the target diet to be known, making it possible to prevent disease onset through personalized nutrition.

[0078] (Estimation Device) FIG. 14 is a diagram illustrating an example of an estimation device according to a modified embodiment. The estimation device 300 according to the modified embodiment receives animal-related information. The animal-related information includes animal identification information and blood amino acid composition information indicating the composition of amino acids in the animal's blood. The estimation device 300 acquires the blood amino acid composition information included in the animal-related information, and estimates disease-improving blood amino acid composition information indicating the composition of blood amino acids required to improve the disease based on the acquired blood amino acid composition information and disease information indicating the correspondence between the blood amino acid composition and the disease state. The estimation device 300 acquires feed composition information indicating the composition of nutrients in feed based on the blood amino acid composition information, the disease-improving blood amino acid composition information, and a trained model. The estimation device 300 outputs the animal identification information and the acquired feed composition information.

[0079] The estimation device 300 is realized by a device such as a personal computer, a server, a smartphone, a tablet computer, an industrial computer, etc. The estimation device 300 includes an input unit 301, a blood amino acid composition acquisition unit 302, a feed composition acquisition unit 303, an output unit 305, an estimation unit 306, and a memory unit 310.

[0080] The input unit 301, blood amino acid composition acquisition unit 302, output unit 305, and memory unit 310 are similar to the input unit 201, blood amino acid composition acquisition unit 202, output unit 205, and memory unit 210 described above, respectively, and therefore will not be described again. The estimation unit 306 acquires animal identification information and blood amino acid composition information from the blood amino acid composition acquisition unit 302. The estimation unit 306 stores disease information indicating the correspondence between the composition of blood amino acids and the state of the disease. The estimation unit 306 estimates disease-improving blood amino acid composition information indicating the composition of blood amino acids required to improve the disease, based on the acquired blood amino acid composition information and disease information.

[0081] The feed composition acquisition unit 303 acquires animal identification information and disease-improving blood amino acid composition information from the estimation unit 306. The feed composition acquisition unit 303 is equipped with a trained model 304. The trained model 304 is a machine-learned version of the correspondence between the nutrient composition in the feed indicated by the feed composition information and the amino acid composition in the blood indicated by the blood amino acid composition information. For example, the trained model 304 is created by the learning device 100. In this case, the trained model output by the learning device 100 is received by the estimation device 300 via a network or medium, and acquired by the feed composition acquisition unit 303. Note that the estimation device 300 may include the learning device 100. In other words, the estimation device 300 may create the trained model 304. In this case, the feed composition acquisition unit 303 acquires the trained model 304 from the learning device 100.

[0082] The feed composition acquisition unit 303 inputs the acquired disease-improving blood amino acid composition information as blood amino acid composition information into the trained model 304, and acquires feed composition information indicating the composition of nutrients in the feed output by the trained model 304 for the input disease-improving blood amino acid composition information. The feed composition information includes at least the amino acid composition.

[0083] All or part of the input unit 301, blood amino acid composition acquisition unit 302, feed composition acquisition unit 303, output unit 305, and estimation unit 306 are functional units (hereinafter referred to as software functional units) that are realized, for example, by a processor such as a CPU executing a program stored in the storage unit 310. Note that all or part of the input unit 301, blood amino acid composition acquisition unit 302, feed composition acquisition unit 303, output unit 305, and estimation unit 306 may be realized by hardware such as an LSI, an ASIC, or an FPGA, or may be realized by a combination of software functional units and hardware.

[0084] (Operation of the estimation device 300) Figure 15 is a flowchart showing an example of the operation of the estimation device of the modified embodiment. With reference to Figure 15, the operation of the estimation device 300 after the feed composition acquisition unit 303 acquires the trained model 304 will be described. Steps S1-3 to S2-3 and S5-3 are the same as steps S1-2 to S2-2 and S4-2 in Figure 12, and therefore will not be described here.

[0085] (Step S3-3) The estimation unit 306 acquires animal identification information and blood amino acid composition information from the blood amino acid composition acquisition unit 302. The estimation unit 306 estimates disease-improving blood amino acid composition information indicating the composition of blood amino acids for improving the disease based on the acquired blood amino acid composition information and disease information. (Step S4-3) The feed composition acquisition unit 303 acquires animal identification information and disease-improving blood amino acid composition information from the estimation unit 306. The feed composition acquisition unit 303 inputs the acquired disease-improving blood amino acid composition information as blood amino acid composition information into the trained model 304, and acquires feed composition information indicating the composition of nutrients in the feed output by the trained model 304 for the input disease-improving blood amino acid composition information.

[0086] In addition to the effects of the estimation device 200 of the embodiment, the estimation device 300 of the modified embodiment can calculate the composition of a feed (meal) that changes the blood amino acid profile from a diseased state to one that can ameliorate the disease. This makes it possible to propose the amino acid composition of a feed (meal) that will obtain a blood amino acid profile that will ameliorate the disease.

[0087] (1) A learning device according to this embodiment includes a first acquisition unit that acquires feed composition information indicating the composition of nutrients in feed, a second acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed the feed, and a learning unit that generates a trained model that has learned the correspondence between the nutrient composition in the feed indicated by the feed composition information and the blood amino acid composition indicated by the blood amino acid composition information. This configuration allows the learning device to generate a trained model that has learned the correspondence between the nutrient composition in the feed indicated by the feed composition information and the blood amino acid composition indicated by the blood amino acid composition information. This allows the learning device to estimate the blood amino acid composition from the nutrient composition in the feed, and the nutrient composition in the feed from the blood amino acid composition.

[0088] (2) In the learning device, the learning unit generates a trained model by Lasso regression on feed composition information and blood amino acid composition information. By configuring in this manner, the learning device can generate a trained model by Lasso regression on feed composition information and blood amino acid composition information, thereby enabling estimation with higher accuracy than when a trained model is generated by Lasso regression on one week's worth of ingested amino acids and blood amino acid composition information instead of feed composition information and blood amino acid composition information.

[0089] (3) In the learning device, the learning unit generates a trained model by support vector regression for the feed composition information and the blood amino acid composition information. By configuring in this way, the learning device can generate a trained model by support vector regression for the feed composition information and the blood amino acid composition information, thereby enabling estimation with higher accuracy compared to when a trained model is generated by Lasso regression.

[0090] (4) In the learning device, the nutrients include multiple types of amino acids, and the feed composition information includes at least an amino acid composition. This configuration allows the learning device to generate a trained model that learns the correspondence between the nutrient composition in feed that includes at least the amino acid composition indicated by the feed composition information and the blood amino acid composition indicated by the blood amino acid composition information. Therefore, it is possible to estimate the blood amino acid composition from the nutrient composition in feed that includes at least an amino acid composition, and to estimate the nutrient composition in feed that includes at least an amino acid composition from the blood amino acid composition.

[0091] (5) The estimation device according to this embodiment includes a blood amino acid composition acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed with feed, and a feed composition acquisition unit that acquires feed composition information indicating the composition of nutrients in the feed, which is output by a trained model trained by the learning device described in (1) when the blood amino acid composition information is provided to the trained model. This configuration allows the estimation device to acquire feed composition information indicating the composition of nutrients in the feed, which is output by the trained model when the blood amino acid composition information is provided to the trained model, and therefore allows the estimation device to propose a composition of nutrients in the feed that will improve a deteriorated blood amino acid composition.

[0092] (6) The estimation device according to this embodiment includes a feed composition acquisition unit that acquires feed composition information indicating the composition of nutrients in the feed, and a blood amino acid composition acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of the animal fed by providing the feed composition information to a trained model trained by the learning device described in (1). This configuration allows the estimation device to acquire blood amino acid composition information indicating the composition of amino acids in the blood of the animal fed by providing the feed composition information to the trained model, which is output by the trained model. Therefore, the estimation device can propose a blood amino acid composition from the composition of nutrients in the feed.

[0093] (7) The estimation device according to this embodiment includes an estimation unit that estimates a nutrient composition in a feed for improving the disease based on a trained model trained by the learning device described in (1), disease information indicating a correspondence between blood amino acid composition and the state of the disease, and information indicating the blood amino acid composition for improving the disease. This configuration allows the estimation device to estimate a nutrient composition in a feed for improving the disease based on the trained model, the disease information, and the information indicating the blood amino acid composition for improving the disease, thereby proposing a nutrient composition in a feed for improving the disease.

[0094] (8) In the estimation device, the nutrients include multiple types of amino acids, and the feed composition information includes at least an amino acid composition. By configuring in this manner, the estimation device can obtain feed composition information indicating the composition of nutrients in feed containing multiple types of amino acids, which is output by the trained model when blood amino acid composition information is provided to the trained model, and can therefore propose a nutrient composition in feed containing multiple types of amino acids that improves a deteriorated blood amino acid composition.

[0095] Although the embodiments of the present invention have been described above, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be embodied in various other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the spirit of the invention. These embodiments and their modifications are intended to be included within the scope and spirit of the invention, as well as within the scope of the invention and its equivalents as set forth in the claims.

[0096] The learning device 100, the estimation device 200, and the estimation device 300 each have a built-in computer. The processing steps of each of the devices are stored in the form of a program on a computer-readable recording medium, and the computer reads and executes this program to perform the above processing.

[0097] Here, computer-readable recording media refers to magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memories, etc. Alternatively, the computer program may be distributed to a computer via a communication line, and the computer that receives the distribution may execute the program. The program may also be intended to realize some of the functions described above. Furthermore, the program may be a so-called differential file (differential program) that can realize the functions described above in combination with a program already recorded on a computer device.

[0098] DESCRIPTION OF SYMBOLS 100... learning device, 101... first acquisition unit, 102... second acquisition unit, 103... learning unit, 104... learning model, 105... output unit, 110... storage unit, 200, 300... estimation device, 201, 301... input unit, 202, 302... blood amino acid composition acquisition unit, 203, 303... feed composition acquisition unit, 204, 304... trained model, 205, 305... output unit, 306... estimation unit, 210, 310... storage unit

Claims

1. A learning device comprising: a first acquisition unit that acquires feed composition information indicating the composition of nutrients in feed; a second acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed the feed; and a learning unit that generates a trained model that has learned the correspondence between the composition of nutrients in the feed indicated by the feed composition information and the composition of amino acids in the blood indicated by the blood amino acid composition information.

2. The learning device according to claim 1, wherein the learning unit generates a trained model by Lasso regression on the feed composition information and the blood amino acid composition information.

3. The learning device according to claim 1, wherein the learning unit generates a trained model by support vector regression on the feed composition information and the blood amino acid composition information.

4. The learning device according to claim 1, wherein the nutrients include a plurality of types of amino acids, and the feed composition information includes at least an amino acid composition.

5. An estimation device comprising: a blood amino acid composition acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed with feed; and a feed composition acquisition unit that acquires feed composition information indicating the composition of nutrients in the feed, which is output by a trained model trained by the learning device described in claim 1 when the trained model is supplied with the blood amino acid composition information.

6. An estimation device comprising: a feed composition acquisition unit that acquires feed composition information indicating the composition of nutrients in feed; and a blood amino acid composition acquisition unit that acquires blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed the feed, which is output by a trained model trained by the learning device described in claim 1 when the trained model is provided with the feed composition information.

7. An estimation device comprising an estimation unit that estimates the nutrient composition in feed to ameliorate the disease based on a trained model trained by the learning device described in claim 1, disease information indicating the correspondence between the composition of amino acids in the blood and the state of the disease, and information indicating the composition of amino acids in the blood to ameliorate the disease.

8. The estimation device according to any one of claims 5 to 7, wherein the nutrients include a plurality of types of amino acids, and the feed composition information includes at least an amino acid composition.

9. A program for causing a computer to perform the following steps: acquiring feed composition information indicating the composition of nutrients in feed; acquiring blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed the feed; and generating a trained model that has learned the correspondence between the composition of nutrients in the feed indicated by the feed composition information and the composition of amino acids in the blood indicated by the blood amino acid composition information.

10. A program for causing a computer to perform the following steps: obtain blood amino acid composition information indicating the composition of amino acids in the blood of an animal fed with feed; provide the blood amino acid composition information to a trained model that has learned the correspondence between the composition of nutrients in feed and the composition of amino acids in the blood of the animal fed with the feed; and obtain feed composition information indicating the composition of nutrients in the feed, output by the trained model.