Method and system for filling missing values of blood metabolome of dairy cow

Abstract

The invention discloses a milk cow blood metabolome missing value filling method and system, and belongs to the technical field of metabolic index detection and data analysis. The milk cow blood metabolome missing value filling method comprises the following steps that a whole blood sample of a milk cow is collected, and a plasma metabolome and serum metabolome data are obtained; performing overlapping distribution statistics on the plasma metabolome data, the serum metabolome data and the two metabolome data sets; decomposing a plasma metabolome, a serum metabolome and a plasma and serum multi-data set matrix through non-negative matrix decomposition to obtain an optimal matrix dimension; filling missing values of a plasma metabolome, a serum metabolome and a plasma and serum multi-dataset metabolome; and calculating the filling accuracy of the plasma metabolome, the serum metabolome and the plasma and serum multi-data-set metabolome to obtain the milk cow blood metabolome missing value. According to the filling method provided by the invention, not only is the missing value of the single data set filled, but also the non-overlapping metabolome of the multiple data sets is filled.

Description

Technical Field

[0001] The invention belongs to the technical field of metabolic index detection and data analysis, and in particular relates to a method and system for filling missing values ​​of a dairy cow blood metabolic group. Background Art

[0002] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention, and should not necessarily be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art.

[0003] The cow blood metabolome can be used to study the mechanisms of metabolic pathway activity and to identify metabolic biomarkers of disease and treatment response. By observing how metabolite levels change under various physiological conditions, metabolomics can reveal the role of metabolites in homeostasis, disease, or response to perturbations.

[0004] In current research, a large amount of large-scale metabolomics data is generated using mass spectrometry. In order to accurately identify metabolites, targeted metabolomics studies must be calibrated to identify specific classes of metabolites with similar chemical properties with maximum sensitivity. Therefore, each metabolomics detection platform can only measure a subset of all metabolites in a sample. However, metabolomics analysis performed on different platforms usually obtains fewer overlapping metabolites, and this less overlap greatly limits the merging and comparison of cross-datasets. In the same dataset, some metabolites have low relative abundance, making it relatively difficult to compare with other metabolites, let alone compare with the measured values ​​of the same metabolites in different datasets. The lack of a common measurement scale between metabolites and datasets hinders the use of multi-data metabolomes. Therefore, it is necessary to fill in the missing values ​​of a single dataset, and it is even more necessary to fill in the non-overlapping metabolomes of multiple datasets. Summary of the invention

[0005] In order to address the deficiencies of the prior art, the purpose of the present invention is to provide a method and system for filling missing values ​​in a cow blood metabolome, which not only fills missing values ​​in a single data set, but also fills non-overlapping metabolomes in multiple data sets.

[0006] In order to achieve the above object, the technical solution of the present invention is:

[0007] On the one hand, the present invention provides a method for filling missing values ​​of a dairy cow blood metabolome, comprising the following steps:

[0008] Whole blood samples were collected from dairy cows to obtain plasma metabolome and serum metabolome data;

[0009] Overlap distribution statistics were performed on the plasma metabolome, serum metabolome data, and the two metabolome datasets;

[0010] The plasma metabolome, serum metabolome, and plasma and serum multi-dataset matrices were decomposed by non-negative matrix factorization to obtain the optimal matrix dimension number;

[0011] Fill missing values ​​for plasma metabolome, serum metabolome, and plasma and serum multi-dataset metabolome;

[0012] The filling accuracy of plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome was calculated to obtain the filling method of missing values ​​of dairy cow blood metabolome.

[0013] In one or more embodiments, collecting whole blood samples from dairy cows to obtain plasma metabolome and serum metabolome data specifically includes:

[0014] Anticoagulant tube whole blood samples and procoagulant tube whole blood samples were collected from cows in different dairy farms. The samples were centrifuged to obtain non-hemolyzed plasma or serum, which were stored and tested to obtain the plasma metabolome and serum metabolome of each individual cow. Among them, plasma or serum samples with slight hemolysis, moderate hemolysis and severe hemolysis were discarded.

[0015] In one or more embodiments, the decomposing the plasma metabolome, serum metabolome and plasma and serum multi-dataset matrices by non-negative matrix decomposition to obtain the optimal matrix dimension number specifically includes:

[0016] For different metabolome datasets, the non-negative high-dimensional matrix with low rank approximation was obtained by non-negative matrix factorization (NMF) method. The interval range was set for the metabolites of single datasets of plasma metabolome and serum metabolome, as well as multiple datasets of plasma and serum. The optimal matrix dimension k was evaluated through 10-fold cross validation.

[0017] Furthermore, for different metabolomics datasets, a low-rank approximate non-negative high-dimensional matrix is ​​obtained by a non-negative matrix decomposition method, specifically: the preprocessed different datasets are merged to form a sparse matrix X, which is then decomposed into two embedding matrices W and H; wherein, when the rows of the X matrix are samples, the columns of the W matrix are used to describe the relative contribution of each embedding vector to the sample and reveal the clustering between samples; when the columns of the X matrix are omics marker features, the rows of the H matrix are used to describe the relative contribution of omics marker features to the embedding vector.

[0018] Furthermore, the interval ranges of metabolites in the single datasets of plasma metabolome and serum metabolome, as well as the multiple datasets of plasma and serum were set, and the optimal matrix dimension k was evaluated through 10-fold cross validation, specifically:

[0019] For the single datasets of plasma metabolome and serum metabolome, as well as the metabolites of plasma and serum multiple datasets, the range interval [1-n] was set, where n is the total number of metabolites. The X matrix was decomposed from 1 to n one by one through 10-fold cross validation, and then the mean absolute error (MAE) was calculated to evaluate the optimal matrix dimension k.

[0020] Furthermore, the optimal matrix dimension k is the one in which the mean absolute error (MAE) is the smallest when the k-th metabolite decomposes the X matrix into the W matrix and the H matrix.

[0021] In one or more embodiments, the filling of missing values ​​of the plasma metabolome, serum metabolome, and plasma and serum multi-dataset metabolome specifically includes:

[0022] The unmeasured omics marker features are formulated as non-negative matrix factorization (NMF) factors for missing value processing. Assuming that the loss function is the term-by-term sum of the losses, the matrix is ​​decomposed by deleting the entries corresponding to the missing items in the loss function; after optimization, both the W and H matrices have no missing entries, and their product Also does not include missing entries, and then After sorting transformation, these predicted entries are the filled-in omics feature values.

[0023] Furthermore, the formula for missing value processing is:

[0024]

[0025] Among them, the dimension of W matrix is The H matrix dimensions are x ij is the missing value of the omics marker, w j is the i-th element of the W matrix, h j is the j-th element of the W matrix, and T is the transpose used in the matrix multiplication calculation.

[0026] In one or more embodiments, the method of calculating the filling accuracy of the plasma metabolome, the serum metabolome, and the plasma and serum multi-dataset metabolome to obtain the filling method of the missing values ​​of the cow blood metabolome specifically includes:

[0027] The single metabolite datasets of the plasma metabolome and serum metabolome, as well as the combined metabolite dataset of the plasma and serum metabolome were filled, respectively. After repeated filling analysis, the median range of the Spearman rank correlation coefficient (ρ(rho)) of the filling accuracy of each metabolite was obtained.

[0028] On the other hand, the present invention provides a system for filling missing values ​​of a cow's blood metabolome, comprising:

[0029] A collection module, which is used to collect whole blood samples from dairy cows to obtain plasma metabolome and serum metabolome data;

[0030] A statistical module, which is used to perform overlapping distribution statistics on plasma metabolome data, serum metabolome data, and two sets of metabolome data sets;

[0031] A non-negative matrix factorization module, which is used to decompose the plasma metabolome, serum metabolome, and plasma and serum multi-dataset matrices through non-negative matrix factorization to obtain the optimal matrix dimension number;

[0032] Filling module, which is used to fill missing values ​​of plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome;

[0033] The calculation accuracy module is used to calculate the filling accuracy of the plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome, and obtain the filling method of the missing values ​​of the cow blood metabolome.

[0034] One or some of the above technical solutions have the following advantages or beneficial effects:

[0035] The present invention provides a method for filling missing values ​​of a cow's blood metabolome. The method comprises the following steps: decomposing a plasma metabolome, a serum metabolome and a plasma and serum multi-dataset matrix through non-negative matrix decomposition to obtain an optimal matrix dimension; filling missing values ​​of the plasma metabolome, the serum metabolome and the plasma and serum multi-dataset metabolome; and calculating the filling accuracy of the plasma metabolome, the serum metabolome and the plasma and serum multi-dataset metabolome to obtain an optimal method for filling missing values ​​of the cow's blood metabolome, which not only realizes the filling of missing values ​​of a single data set, but also realizes the filling of non-overlapping metabolomes of multiple data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0037] Figure 1 A technical roadmap of the method for filling missing values ​​of the blood metabolome of dairy cows in Example 1 of the present invention;

[0038] Figure 2 This is a comparison chart of the hemolysis degree of plasma in normal state, mild hemolysis, moderate hemolysis and severe hemolysis in Example 1 of the present invention;

[0039] Figure 3 The plasma metabolome (data1), serum metabolome (data2) and two metabolome data sets in Example 1 of the present invention;

[0040] Figure 4is the filling accuracy of the plasma metabolome metabolites in Example 1 of the present invention (the red bar graph represents that in 90% of the repeated filling analysis, the filling accuracy ρ (rho)>0 and the Benjamini-Hochberg (BH) corrected q value<0.05);

[0041] Figure 5 is the filling accuracy of serum metabolites in Example 1 of the present invention (the red bar graph represents that in 90% of the repeated filling analysis, the filling accuracy ρ (rho)>0 and the Benjamini-Hochberg (BH) corrected q value <0.05);

[0042] Figure 6 This is the combined filling accuracy of the two groups of metabolites of plasma and serum in Example 1 of the present invention (the red bar graph represents that in 90% of the repeated filling analyses, the filling accuracy ρ (rho)>0 and the Benjamini-Hochberg (BH) corrected q value<0.05). DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to more clearly understand the technical solution of the present invention, the technical solution of the present invention will be described in detail below in conjunction with specific embodiments.

[0044] Example 1

[0045] The present invention provides a method for filling missing values ​​of a dairy cow blood metabolome, comprising the following steps:

[0046] 1. Collect whole blood samples from the tail vein of dairy cows at the dairy farm:

[0047] At dairy farm A, one tube of whole blood sample is collected from each cow in an anticoagulant tube; at dairy farm B, one tube of whole blood sample is collected from each cow in a procoagulant tube, with one tube for each cow in each farm. The sampling personnel shall promptly and clearly record the information of the cattle on the sampling tube, and draw blood from the base of the cattle's tail according to the corresponding cattle (approximate location: one finger away from the anus, the tail vein is about one circle smaller than the little finger. When the blood collection needle is used to start the needle, there will be a little feeling of piercing the plastic wall, and then the needle is pushed all the way to the very hard coccyx, and then the needle is returned, and the needle is suspended in the tail vein. Finally, the vacuum blood collection tube is inserted, and the blood automatically flows out). Both the anticoagulant sampling tube and the procoagulant sampling tube are 10mL anticoagulant tubes. Each blood collection must be guaranteed to be more than 8mL (automatic bleeding, until it is all out). After the collection, gently invert the blood collection tube 4 times to ensure that the anticoagulant fully exerts its anticoagulant effect.

[0048] The sample processing personnel promptly took the blood samples collected by the samplers back to the temporary experimental area for centrifugation: centrifugation at 4°C in the cattle farm laboratory, 1600g for 10 minutes, and obtained non-hemolyzed plasma or serum (i.e. Figure 2The supernatant (i.e. plasma or serum) is pipetted into antifreeze centrifuge tubes (those with screw caps to prevent tube bursting) with a pipette. The remaining plasma or serum is evenly dispensed into other antifreeze centrifuge tubes (as backup) and all numbers are marked. Plasma or serum samples with slight hemolysis, moderate hemolysis, and severe hemolysis are discarded ( Figure 2 ). The collected plasma samples were mailed to the sequencing company on -40℃ dry ice for mass spectrometry detection to obtain the complete metabolome of plasma or serum of each individual dairy cow.

[0049] 2. Overlapping distribution statistics of plasma metabolome and serum metabolome as well as the two metabolome datasets:

[0050] like Figure 3 As shown in the figure, a total of 100 plasma metabolomes were obtained from dairy farm A, and each individual metabolome had 2962 plasma metabolites. A total of 100 serum metabolomes were obtained from dairy farm B, and each individual metabolome had 2559 serum metabolites. Among them, the plasma metabolome dataset (data1) of dairy farm A contains 2084 specific metabolites, and the serum metabolome dataset (data2) of dairy farm B contains 1681 specific metabolites, of which 878 common metabolites were found in the two datasets (data1) and data2 of dairy farms A and B.

[0051] 3. Decompose the plasma metabolome and serum metabolome as well as the plasma and serum multi-dataset matrices through non-negative matrix factorization (NMF) to obtain the optimal matrix dimension number:

[0052] For different metabolomics datasets, the non-negative matrix factorization (NMF) method is used to obtain a low-rank approximate non-negative high-dimensional matrix, that is, the pre-processed different datasets d i The merged matrix X is formed into a sparse matrix, which is then decomposed into two embedding matrices W and H. In the NMF method, when the rows of the X matrix are samples, the columns of the W matrix are used to describe the relative contribution of each embedding vector to the sample and can reveal the clustering between samples. When the columns of the X matrix are omics marker features, the rows of the H matrix are used to describe the relative contribution of omics marker features to the embedding vector.

[0053] The range interval [1-n] was set for the metabolites of the plasma metabolome and serum metabolome single datasets, as well as the plasma and serum multiple datasets, where n is the total number of metabolites. The X matrix was decomposed from 1 to n one by one through 10-fold cross validation, and then the mean absolute error (MAE) was calculated. The optimal matrix dimension k was evaluated, that is, when the X matrix was decomposed into the W matrix and the H matrix at the kth metabolite, the MAE was the smallest.

[0054] 4. Fill in missing values ​​of plasma metabolome and serum metabolome as well as missing values ​​of plasma and serum multi-dataset metabolome:

[0055] Missing value handling was performed by formulating unmeasured omics marker features into NMF factors.

[0056]

[0057] Among them, the dimension of W matrix is The H matrix dimensions are x ij is the missing value of the omics marker, w j is the i-th element of the W matrix, h j is the jth element of the W matrix, and T is the transpose used for matrix multiplication. Assuming that the loss function is the term-by-term sum of the losses, the matrix can be decomposed by deleting the entries corresponding to the missing items in the loss function. After optimization, both the W and H matrices have no missing entries, and their product Also does not include missing entries, and then After sorting transformation, these predicted entries are the filled-in omics feature values.

[0058] 5. Calculate the filling accuracy of plasma metabolome and serum metabolome as well as plasma and serum multi-dataset metabolome, and obtain the best filling method for missing values ​​of dairy cow blood metabolome:

[0059] By filling in missing metabolites (approximately 10% of the metabolites were missing, i.e., 290 metabolites were missing) in a single dataset of plasma metabolome metabolites (n = 2962), after 200 repeated filling analyses, the median range of the filling accuracy ρ (rho) for each metabolite was -0.11 to 0.89, with an average of 0.41 ( Figure 4 ).

[0060] The missing metabolites (approximately 10% of the metabolites were missing, i.e., 260 metabolites were missing) were filled in a single dataset of serum metabolome metabolites (n = 2559). After 200 repeated filling analyses, the median range of the filling accuracy ρ (rho) for each metabolite was -0.20 to 0.88, with an average of 0.37 ( Figure 5 ).

[0061] The missing metabolites (approximately 10% of the metabolites were missing, i.e., 460 metabolites were missing) were filled in for the combined metabolite datasets of the two metabolite groups of plasma and serum (n = 4643). After 200 repeated filling analyses, the median range of the filling accuracy ρ (rho) for each metabolite was -0.22 to 0.88, with an average of 0.41 ( Figure 6 ).

[0062] Example 2

[0063] The present invention provides a system for filling missing values ​​of a dairy cow's blood metabolome, which is characterized by comprising:

[0064] A collection module, which is used to collect whole blood samples from dairy cows to obtain plasma metabolome and serum metabolome data;

[0065] A statistical module, which is used to perform overlapping distribution statistics on plasma metabolome data, serum metabolome data, and two sets of metabolome data sets;

[0066] A non-negative matrix factorization module, which is used to decompose the plasma metabolome, serum metabolome, and plasma and serum multi-dataset matrices through non-negative matrix factorization to obtain the optimal matrix dimension number;

[0067] Filling module, which is used to fill missing values ​​of plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome;

[0068] The calculation accuracy module is used to calculate the filling accuracy of the plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome, and obtain the filling method of the missing values ​​of the cow blood metabolome.

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for filling missing values ​​in the blood metabolome of dairy cows, characterized in that: The following steps are involved: Whole blood samples were collected from dairy cows to obtain plasma metabolome and serum metabolome data; Overlap distribution statistics were performed on the plasma metabolome, serum metabolome data, and the two metabolome datasets; The plasma metabolome, serum metabolome, and plasma and serum multi-dataset matrices were decomposed by non-negative matrix factorization to obtain the optimal matrix dimension number; Fill missing values ​​for plasma metabolome, serum metabolome, and plasma and serum multi-dataset metabolome; The filling accuracy of plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome was calculated to obtain the filling method of missing values ​​of dairy cow blood metabolome.

2. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 1, characterized in that: The whole blood sample of the dairy cow is collected to obtain the plasma metabolome and serum metabolome data, specifically including: Anticoagulant tube whole blood samples and procoagulant tube whole blood samples were collected from cows in different dairy farms. The samples were centrifuged to obtain non-hemolyzed plasma or serum, which were stored and tested to obtain the plasma metabolome and serum metabolome of each individual cow. Among them, plasma or serum samples with slight hemolysis, moderate hemolysis and severe hemolysis were discarded.

3. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 1, characterized in that: The plasma metabolome, serum metabolome and plasma and serum multi-dataset matrices are decomposed by non-negative matrix decomposition to obtain the optimal matrix dimension number, specifically including: For different metabolome datasets, a low-rank approximate non-negative high-dimensional matrix was obtained through non-negative matrix factorization method. The interval range was set for the metabolites of single datasets of plasma metabolome and serum metabolome, as well as multiple datasets of plasma and serum. The optimal matrix dimension k was evaluated through 10-fold cross validation.

4. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 3, characterized in that: For different metabolomics datasets, a low-rank approximate non-negative high-dimensional matrix is ​​obtained by a non-negative matrix decomposition method, specifically: The preprocessed different data sets are merged to form a sparse matrix X, which is then decomposed into two embedding matrices W and H. When the rows of the X matrix are samples, the columns of the W matrix are used to describe the relative contribution of each embedding vector to the sample and reveal the clustering between samples. When the columns of the X matrix are omics marker features, the rows of the H matrix are used to describe the relative contribution of omics marker features to the embedding vector.

5. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 3, characterized in that: The interval ranges of metabolites in the single datasets of plasma metabolome and serum metabolome, as well as the multiple datasets of plasma and serum were set, and the optimal matrix dimension k was evaluated through 10-fold cross validation, specifically: For the single datasets of plasma metabolome and serum metabolome, as well as the metabolites of plasma and serum multiple datasets, the range interval [1-n] was set, where n is the total number of metabolites; the X matrix was decomposed from 1 to n one by one through 10-fold cross validation, and then the mean absolute error was calculated to evaluate the optimal matrix dimension k.

6. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 5, characterized in that: The optimal matrix dimension k is the minimum mean absolute error obtained when the X matrix is ​​decomposed into the W matrix and the H matrix for the kth metabolite.

7. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 1, characterized in that: The missing values ​​of the plasma metabolome, serum metabolome, and plasma and serum multi-dataset metabolome are filled in, specifically including: The unmeasured omics marker features are formulated as non-negative matrix decomposition factors for missing value processing. Assuming that the loss function is the term-by-term sum of the losses, the matrix is ​​decomposed by deleting the entries corresponding to the missing items in the loss function; after optimization, both the W and H matrices have no missing entries, and their product Also does not include missing entries, and then After sorting transformation, these predicted entries are the filled-in omics feature values.

8. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 7, characterized in that: The formula for missing value processing is: Among them, the dimension of W matrix is The H matrix dimensions are x ij is the missing value of the omics marker, w j is the i-th element of the W matrix, h j is the j-th element of the W matrix, and T is the transpose used in the matrix multiplication calculation.

9. The method for filling missing values ​​of the blood metabolome of dairy cows according to claim 1, characterized in that: The method of calculating the filling accuracy of the plasma metabolome, the serum metabolome, and the plasma and serum multi-dataset metabolome to obtain the filling method of the missing values ​​of the dairy cow blood metabolome specifically includes: The single metabolite datasets of the plasma metabolome and serum metabolome, as well as the combined metabolite dataset of the two metabolomes of plasma and serum were filled, and after repeated filling analysis, the median range of the Spearman rank correlation coefficient of the filling accuracy of each metabolite was obtained.

10. A system for filling missing values ​​in the blood metabolome of dairy cows, characterized in that: include: A collection module, which is used to collect whole blood samples from dairy cows to obtain plasma metabolome and serum metabolome data; A statistical module, which is used to perform overlapping distribution statistics on plasma metabolome data, serum metabolome data, and two sets of metabolome data sets; A non-negative matrix factorization module, which is used to decompose the plasma metabolome, serum metabolome, and plasma and serum multi-dataset matrices through non-negative matrix factorization to obtain the optimal matrix dimension number; Filling module, which is used to fill missing values ​​of plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome; The calculation accuracy module is used to calculate the filling accuracy of the plasma metabolome, serum metabolome and plasma and serum multi-dataset metabolome, and obtain the filling method of the missing values ​​of the cow blood metabolome.

Images