Maternity folic acid supplement effectiveness prediction method based on machine learning

Abstract

The invention provides a pregnant woman folic acid supplement effectiveness prediction method based on machine learning. The method comprises the following steps: firstly, acquiring genome and metabolome data of a pregnant woman queue, and determining independent SNP (Single Nucleotide Polymorphism) remarkably associated with folic acid and metabolites with whole genome remarkable sites by using whole genome association analysis; then, a dual sample Mendel randomization analysis is performed to identify metabolites having a causal relationship with folate deficiency. And training a machine learning model by using the genome and metabolome data, related SNP and metabolites, and finally forming a folic acid detection model. And inputting the independent SNP of a to-be-detected object and related metabolites into the model to obtain a detection result, thereby evaluating the effectiveness of folic acid supplement. According to the method, by detecting a small amount of SNP and metabolites, an efficient and economical folic acid deficiency prediction means is provided, the analysis process is simplified, and the cost is reduced. Sample obtaining is convenient, operation is easy, and the method can be widely applied to different levels of medical institutions.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning. Background Art

[0002] Folic acid is also called vitamin B9, and is named folic acid because it is rich in green leaves. Folic acid cannot be synthesized in the human body, nor can it be stored in large quantities, so it needs to be supplemented exogenously. There are usually two ways to ingest folic acid exogenously, natural folic acid (Folate) and synthetic folic acid (Folic acid). Natural folic acid mainly comes from plants or animal tissues with high folic acid content in the diet, mainly animal liver, green leaves, beans, etc. In addition to directly ingesting folic acid contained in animal and plant tissues, it can also be obtained by consuming synthetic folic acid drugs or processed foods containing synthetic folic acid. Natural folic acid generally exists in the form of reduced dihydrofolic acid (Dihydrofolic acid, DHF) or tetrahydrofolic acid (Tetrahydrofolate, THF), which contains a polyglutamic acid tail to maintain the biological activity of folic acid in the body; while synthetic folic acid is an oxidized form and only contains a glutamic acid tail chain.

[0003] Folic acid is a methyl donor in the one-carbon cycle. It also acts as a cofactor in many other biochemical reactions and participates in many physiological activities of the human body. The plasma concentration of folic acid is also closely related to the health of the human body. Folic acid deficiency, vitamin B12 deficiency, and high levels of homocysteine ​​(Hcy) in pregnant women are related to neural tube defects (NTDs) in newborns, that is, congenital brain and spinal cord abnormalities. In the body, folic acid can be converted into active folic acid by methyltetrahydrofolate reductase (MTHFR). Active folic acid participates in homocysteine ​​metabolism, reduces the concentration of homocysteine, and thus reduces a series of diseases and physiological problems caused by high homocysteine. Therefore, pregnant women need to maintain high folic acid levels during pregnancy to prevent fetal neural tube defects. Generally, doctors recommend that pregnant women take 400 micrograms of folic acid per day before and during early pregnancy.

[0004] However, in clinical practice, some pregnant women still have low plasma folate levels after taking large amounts of folic acid for a long time. This situation may lead to neural tube defects in newborns, seriously affect the development and health of the fetus, and impose a heavy burden on the family and society. However, the existing detection methods cannot predict whether there is a lack of folic acid during pregnancy during the preparation period, so a prospective method of predicting folic acid deficiency in pregnant women is needed. Since folic acid levels are a trait that is greatly affected by genetic factors and is also affected by other metabolites, detecting single-nucleotide polymorphism (SNP) sites and metabolites related to folic acid can predict the effect of folic acid supplementation.

[0005] Genome-wide Association Study (GWAS) is an analytical method that detects all or most genes between different individuals in a specific species and determines the association between genotype and phenotype through the difference in allele frequencies between individuals with different phenotypes. GWAS can obtain risk single nucleotide polymorphism (SNP) sites that are significantly associated with folate, and then conditional analysis and fine-mapping can obtain independent significant signals and causal variations from the results of GWAS.

[0006] Mendelian randomization (MR) is a statistical method that uses genetic variation to estimate the causal relationship between exposure and outcome. It is based on Mendel's law of inheritance and instrumental variable estimation methods, using genetic variation as an instrumental variable to evaluate the causal effect of exposure on outcome. Using metabolites as exposure variables and whether or not folic acid is deficient as the outcome variable, it can be determined whether metabolites have a causal relationship with folic acid deficiency.

[0007] Machine learning is an artificial intelligence technology that can learn from data and make predictions. However, there are currently no reports of folate deficiency detection methods using metabolite and genomic data based on machine learning methods. Summary of the invention

[0008] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning, which can predict folic acid deficiency by detecting a small number of SNP sites and metabolites. The detection method is more efficient and economical, simplifies the analysis process and reduces costs. The easy acquisition of samples and simple operation procedures make the method easy to implement in medical institutions of different levels.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] A method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning, comprising:

[0011] Obtain genomic and metabolomic data of the sample pregnant women cohort;

[0012] Based on the genome-wide association analysis method, determining independent SNPs significantly associated with folic acid according to the genome data;

[0013] Based on the whole genome association analysis method, determining the metabolites with significant sites at the whole genome level according to the metabolome data;

[0014] Two-sample Mendelian randomization analysis was performed based on metabolites with significant loci at the whole genome level to obtain metabolites that have a causal relationship with folate deficiency;

[0015] Training a machine learning model according to the genomic data, the metabolomic data, the independent SNPs significantly associated with folic acid, and the metabolites causally related to folic acid deficiency to obtain a trained folic acid detection model;

[0016] Inputting the independent SNPs significantly associated with folic acid and the metabolites causally related to folic acid deficiency of the target to be tested into the folic acid detection model to obtain the detection results;

[0017] The effectiveness of folic acid supplementation of the target to be tested is determined based on the test results.

[0018] Preferably, obtaining genomic data and metabolomic data of a sample pregnant woman cohort includes:

[0019] Testing the sample pregnant women cohort by single nucleotide polymorphism microarray to obtain the genome data;

[0020] The sample pregnant woman cohort is tested by liquid chromatography-tandem mass spectrometry to obtain the metabolome data.

[0021] Preferably, based on a genome-wide association analysis method, determining independent SNPs significantly associated with folic acid according to the genome data includes:

[0022] Performing quality control, genotype interpolation and genome-wide association analysis on the genomic data to obtain analysis results;

[0023] The analysis results were subjected to conditional analysis and positioning to obtain the 5 independent SNPs significantly associated with folic acid.

[0024] Preferably, the number of the five independent SNPs significantly associated with folic acid is five.

[0025] Preferably, the number of metabolites with significant sites at the whole genome level is 18.

[0026] Preferably, the number of metabolites causally related to folic acid deficiency is 7.

[0027] Preferably, the two-sample Mendelian randomization analysis is performed using metabolites with significant loci at the whole genome level as exposure and whether or not folic acid is deficient as outcome.

[0028] Preferably, the folic acid detection model is any one of a logistic regression algorithm and a random forest algorithm.

[0029] Preferably, the key predictive performance indicators of the folic acid detection model include: accuracy, precision, recall, F1 score and ROC-AUC.

[0030] Preferably, the ROC-AUC value of the logistic regression algorithm is 0.78.

[0031] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0032] The present invention provides a method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning, including: obtaining genome data and metabolome data of a sample pregnant woman cohort; based on a whole genome association analysis method, determining an independent SNP significantly associated with folic acid according to the genome data; based on a whole genome association analysis method, determining metabolites with significant sites at the whole genome level according to the metabolome data; performing a double sample Mendelian randomization analysis according to the metabolites with significant sites at the whole genome level, obtaining metabolites with a causal relationship with folic acid deficiency; training a machine learning model according to the genome data, the metabolome data, the independent SNP significantly associated with folic acid, and the metabolites with a causal relationship with folic acid deficiency, obtaining a trained folic acid detection model; inputting the independent SNP significantly associated with folic acid and the metabolites with a causal relationship with folic acid deficiency of the target to be tested into the folic acid detection model to obtain a test result; determining the effectiveness of folic acid supplementation for the target to be tested according to the test result. The present invention can predict folic acid deficiency by detecting a small amount of SNP sites and metabolites. The detection method is more efficient and economical, simplifies the analysis process, and reduces costs. The easy acquisition of samples and simple operation procedures make this method easy to implement in medical institutions of different levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0034] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0035] Figure 2 The results of the genome-wide association analysis of folate provided in the embodiments of the present invention;

[0036] Figure 3 The results of Mendelian randomization analysis of 18 metabolites and folic acid provided in the embodiments of the present invention;

[0037] Figure 4 The ROC curve of the logistic regression model provided by the embodiment of the present invention;

[0038] Figure 5 The ROC curve of the random forest model provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning, comprising:

[0042] Step 100: Obtaining genome data and metabolome data of a sample pregnant woman cohort;

[0043] Step 200: Based on the genome-wide association analysis method, determine the independent SNPs significantly associated with folate according to the genome data;

[0044] Step 300: Based on the whole genome association analysis method, determine the metabolites with significant sites at the whole genome level according to the metabolome data;

[0045] Step 400: performing a two-sample Mendelian randomization analysis based on metabolites with significant loci at the whole genome level to obtain metabolites that are causally related to folate deficiency;

[0046] Step 500: training a machine learning model based on genomic data, metabolomic data, independent SNPs significantly associated with folic acid, and metabolites causally related to folic acid deficiency to obtain a trained folic acid detection model;

[0047] Step 600: inputting the independent SNPs significantly associated with folic acid and the metabolites causally related to folic acid deficiency of the target to be tested into a folic acid detection model to obtain a detection result;

[0048] Step 700: Determine the effectiveness of folic acid supplementation for the target according to the test results.

[0049] In order to implement the prediction method of the present invention, this embodiment constructed a cohort of 2005 pregnant women from the Affiliated Obstetrics and Gynecology Hospital of Fudan University, and obtained the cohort genome and metabolome data through single nucleotide polymorphism microarray (SNP array), i.e., SNP chip, and liquid chromatography-tandem mass spectrometry (LC-MS-MS).

[0050] The Society of Obstetricians and Gynaecologists of Canada (SOGC) proposed that in order to prevent neural tube defects in newborns, pregnant women need to supplement folic acid to a serum folic acid level of 28-30 nmol / L. In fact, clinically, less than 32 nmol / ml is also considered low. The folic acid deficiency standard finally adopted in this application is ≤14.7 ng / ml (33 nmol / l), and the corresponding healthy control group is ≥17.7 ng / ml (40 nmol / l).

[0051] There are two genotyping methods used for genome-wide association study (GWAS) analysis, SNP chip and whole genome sequencing. Whole genome sequencing can obtain all the genetic information of an individual, but the cost is relatively high. This application chose the low-cost but more efficient SNP chip. The genomic data obtained from the SNP chip was quality controlled by PLINKv1.9, and genotype interpolation was performed using Beagle 5.4 to expand the genotype information. Then fastGWA software was used for GWAS analysis. The number of samples entering the GWAS analysis was 1978 (994 cases in the folate deficiency group and 984 cases in the control group), and the number of SNPs was 4,744,363. Figure 2 As shown in the figure, the GWAS analysis results (P value < 1e-5) were subjected to GCTA conditional analysis, FINEMAP and SuSiE fine positioning, and 5 independent SNPs significantly associated with folate were obtained. The 5 independent SNPs obtained by the whole genome association analysis of folate are shown in Table 1.

[0052] Table 1

[0053]

[0054] One of the SNPs, rs34607504, is an expression quantitative trait locus (eQTL) of the oxidized folate receptor (Folate Receptor 1, FOLR1), and is located on the enhancer of the FOLR1 gene. Combined with the phenotypic information associated with the SNP, the A to G mutation of this SNP will lead to a decrease in plasma folate. At the same time, the eQTL information shows that its A to G mutation leads to increased FOLR1 expression. FOLR1 has a stronger transport capacity for oxidized folate than reduced folate, which can be directly converted into active folate; the oxidized folate it transports into cells is also less effective in lowering homocysteine ​​than reduced folate. Therefore, this site is considered a risk site for folate deficiency. For patients with mutations at this site, it can be recommended to use reduced folate or active folate instead of oxidized synthetic folate.

[0055] The metabolome data obtained by liquid chromatography-tandem mass spectrometry overlapped with the genomic data for 522 samples (236 in the folate deficiency group and 286 in the control group). Figure 3 As shown in Table 2, GWAS was performed using the same method as above, and 18 metabolites with significant sites at the whole genome level were found (Pvalue<5e-8). The 18 metabolites were used as exposures, and whether or not folate deficiency was used as the outcome. A two-sample Mendelian randomization analysis was performed, and 7 metabolites with a causal relationship with folate deficiency were obtained (Pvalue<0.05), as shown in Table 2.

[0056] Table 2

[0057]

[0058] Combining these genome and metabolome information, a machine learning method is used to model whether folic acid is deficient, so as to predict the risk of folic acid deficiency. This application uses logistic regression (Logistic Regression) and random forest (RandomForest, RF) algorithms to build a model. All features are standardized. The samples used are 522 samples with both metabolome data and genomic data. 80% are randomly selected as training sets and 20% as test sets. The prediction results are comprehensively compared from multiple key performance indicators such as accuracy, precision, recall rate, F1 score and ROC-AUC. Figure 4 , Figure 5 ,As shown in Tables 3 and 4, the logistic regression model has better prediction performance,,with ROC-AUC reaching 0.78.

[0059] Table 3

[0060]

[0061] Table 4

[0062]

[0063] According to the aforementioned steps, in practice, the data of 5 SNPs and 7 metabolites of a single individual are measured by SNP chip and specific metabolite kit, and the constructed logistic regression model can be used to predict whether the individual is folate deficient.

[0064] As an optional implementation, the prediction steps of this embodiment are as follows:

[0065] S1. Sample collection: Collect blood samples from subjects.

[0066] S2.SNP site detection: Use SNP chip technology to detect 5 specific SNP sites related to folate metabolism in the subjects.

[0067] S3. Metabolite testing: Use a specific metabolite kit to test the subjects for 7 key metabolites that are causally related to folate deficiency.

[0068] S4. Data input: The subjects’ SNP sites and metabolite detection data are standardized and input into the established prediction model.

[0069] S5. Model application: Using the constructed logistic regression model, a Python script was used to calculate whether the subject was deficient in folate based on the input SNP and metabolite data.

[0070] S6. Interpretation of results: If the calculation result shows that the subject is deficient in folic acid, it is considered that the subject is at risk of still being deficient in folic acid after taking a large amount of folic acid supplements.

[0071] S7. Clinical recommendation: If the rs34607504 locus in the subject's test results is AG or GG genotype, it is recommended to take active folic acid or reduced folic acid instead of synthetic folic acid.

[0072] Corresponding to the above method, this embodiment also provides a system for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning, including:

[0073] A data acquisition module is used to obtain genomic data and metabolomic data of a sample pregnant woman cohort;

[0074] A SNP determination module, used to determine independent SNPs significantly associated with folate based on genome data based on a genome-wide association analysis method;

[0075] A metabolite determination module is used to determine metabolites with significant sites at the whole genome level based on the metabolome data based on the whole genome association analysis method;

[0076] The randomization analysis module is used to perform two-sample Mendelian randomization analysis based on metabolites with significant loci at the whole genome level to obtain metabolites that are causally related to folate deficiency;

[0077] A model building module is used to train a machine learning model based on genomic data, metabolomic data, independent SNPs significantly associated with folate, and metabolites causally related to folate deficiency to obtain a trained folate detection model;

[0078] A detection module is used to input the independent SNPs significantly associated with folic acid and the metabolites causally related to folic acid deficiency of the target to be tested into the folic acid detection model to obtain the detection results;

[0079] The effectiveness determination module is used to determine the effectiveness of folic acid supplementation of the target to be tested based on the test results.

[0080] The beneficial effects of the present invention are as follows:

[0081] (1) The present invention can predict folic acid deficiency by detecting a small amount of SNP sites and metabolites. The detection method is more efficient and economical, simplifies the analysis process and reduces costs. The easy acquisition of samples and simple operation procedures make the method easy to implement in medical institutions of different levels.

[0082] (2) The present invention uses a machine learning algorithm to construct a model for predicting folate deficiency, wherein the ROC-AUC value of the logistic regression model reaches 0.78, indicating that its predictive performance is relatively good.

[0083] (3) Folic acid is essential for the health of pregnant women and fetuses. Folic acid deficiency is related to neural tube defects in newborns, and the detection method of the present invention can predict whether a pregnant woman may be deficient in folic acid during pregnancy. This is of great significance for preventing neural tube defects in newborns and other health problems related to folic acid deficiency. This method is also applicable to folic acid deficiency detection in other populations. By detecting SNP sites and metabolites related to folic acid, the present invention provides a forward-looking prediction method for different populations and provides the possibility for personalized medicine. Through personalized folic acid supplementation, health problems related to folic acid deficiency can be more effectively prevented and treated.

[0084] (4) The present invention is based on scientific methods such as GWAS and Mendelian randomization, which ensures the scientificity and reliability of the prediction results. There is no method for detecting folate deficiency based on machine learning methods using metabolites and genomic data in this application, which shows that this application is also innovative.

[0085] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0086] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning, characterized in that: include: Obtain genomic and metabolomic data of the sample pregnant women cohort; Based on the genome-wide association analysis method, determining independent SNPs significantly associated with folic acid according to the genome data; Based on the whole genome association analysis method, determining the metabolites with significant sites at the whole genome level according to the metabolome data; Two-sample Mendelian randomization analysis was performed based on metabolites with significant loci at the whole genome level to obtain metabolites that have a causal relationship with folate deficiency; Training a machine learning model according to the genomic data, the metabolomic data, the independent SNPs significantly associated with folic acid, and the metabolites causally related to folic acid deficiency to obtain a trained folic acid detection model; Inputting the independent SNPs significantly associated with folic acid and the metabolites causally related to folic acid deficiency of the target to be tested into the folic acid detection model to obtain the detection results; The effectiveness of folic acid supplementation of the target to be tested is determined based on the test results.

2. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 1, characterized in that: Obtain genomic and metabolomic data of the sample pregnant women cohort, including: Testing the sample pregnant women cohort by single nucleotide polymorphism microarray to obtain the genome data; The sample pregnant woman cohort is tested by liquid chromatography-tandem mass spectrometry to obtain the metabolome data.

3. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 1, characterized in that: Based on the whole genome association analysis method, independent SNPs significantly associated with folic acid were determined according to the genome data, including: Performing quality control, genotype interpolation and genome-wide association analysis on the genomic data to obtain analysis results; The analysis results were subjected to conditional analysis and positioning to obtain the 5 independent SNPs significantly associated with folic acid.

4. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 1, characterized in that: The number of the five independent SNPs significantly associated with folic acid is five.

5. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 1, characterized in that: The number of metabolites with significant sites at the whole genome level is 18.

6. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 1, characterized in that: The number of metabolites causally related to folic acid deficiency is 7.

7. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 1, characterized in that: The two-sample Mendelian randomization analysis was performed using metabolites with significant loci at the genome-wide level as exposures and folate deficiency as the outcome.

8. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 1, characterized in that: The folic acid detection model is any one of a logistic regression algorithm and a random forest algorithm.

9. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 8, characterized in that: The key predictive performance indicators of the folic acid detection model include: accuracy, precision, recall, F1 score and ROC-AUC.

10. The method for predicting the effectiveness of folic acid supplementation for pregnant women based on machine learning according to claim 9, characterized in that: The ROC-AUC value of the logistic regression algorithm was 0.78.

Images