Method for constructing dietary recommendation model based on diet-gut microbiota-disease database

Abstract

The present invention discloses a method for constructing a dietary recommendation model based on a diet-gut microbiota-disease database, belonging to the technical fields of databases and dietary recommendations. The database of the present invention is constructed by integrating relevant open-source databases and literature, and the content of the database is standardized. The method for constructing the dietary recommendation model includes the following steps: Step 1: Based on the database, modeling is carried out through semantic similarity, Gaussian kernel profile similarity calculation, and graph attention network to construct a diet-gut microbiota and gut microbiota-disease correlation network; Step 2: Based on the qualitative relationships in the database, corresponding to the nodes in the correlation network, a diet-gut microbiota-disease network is constructed; Step 3: The target screening of gut microbiota is carried out through a classifier-optimized model to obtain dietary recommendation results for different diseases. The present invention can provide a data support means in the field of diet-gut microbiota-disease and provide dietary suggestions according to the human health disease status.

Description

Technical Field

[0001] The present invention relates to a method for constructing a dietary recommendation model based on a diet-gut microbiota-disease database, and belongs to the technical fields of databases and dietary recommendations. Background Art

[0002] Microbiota-targeted therapy, that is, preventing and treating special diseases by targeting the regulation of the gut microbiota, is a popular trend in gut microbiota research and translational applications. Drugs, diet, and other factors can improve the functional integrity of the gut microbiota by restoring the gut microbiota environment, thereby promoting host health. Currently, the main intervention methods include narrow-spectrum antibiotics, probiotics, prebiotics, diet, fecal transplantation, etc. Among them, diet intervention stands out among the above measures due to its small side effects and convenient intervention means. In addition, it can be used in combination with other measures to achieve the goal of long-term regulation. Therefore, a comprehensive understanding of the role of diet in the association between microorganisms and diseases is necessary for the development of improved treatment methods. However, the traditional method is to analyze through experimental research, which is time-consuming, costly, and has certain limitations. On this basis, machine learning methods that are time-saving and widely applicable have attracted more and more attention.

[0003] In recent years, various bioinformatics databases have become important auxiliary tools for scientific research. In the field of biological experiments, databases facilitate researchers to retrieve information and improve efficiency. In the field of artificial intelligence, the comprehensiveness and capacity of databases are crucial for the efficiency and performance of model algorithms. So far, there are various microorganism-disease databases on the market (such as Disbiome, GMrepo, HMDAD, etc.). However, although a large number of studies have confirmed the association between diet and gut microbiota, there is still a lack of a systematic diet-microbiota database. In addition, due to the different storage forms and contents of different databases, if different databases are directly combined, data loss will occur. In addition, some databases lack updates, and there are also situations where the data is incomplete. Therefore, based on the current microorganism-disease databases, it is impossible to accurately associate diet and diseases based on the latest research results. The types of diseases and diets that can be associated are relatively few, and the automatic dietary recommendation results for different diseases lack factual basis and still need to rely on the experience of professionals for adjustment. The manual adjustment method is costly and inefficient on the one hand, and highly subjective on the other hand. The recommended results may not be objective, affecting the therapeutic effect of diet therapy. The present invention introduces the microorganism dimension, enhances the interpretability of dietary recommendation results, and provides more theoretical possibilities for the research on the mechanism of gut microbiota and diseases. Summary of the Invention

[0004] In order to improve the accuracy and efficiency of dietary recommendations, the present invention provides a method for constructing a dietary recommendation model based on a diet-gut microbiota-disease database, and the technical solution is as follows:

[0005] The first object of the present invention is to provide a method for constructing a dietary recommendation model, comprising:

[0006] Step 1: Build a diet-gut flora-disease database;

[0007] Step 2: Based on the diet-gut flora-disease database, modeling is performed through semantic similarity, Gaussian kernel profile similarity calculation and graph attention network to construct diet-gut flora and gut flora-disease correlation networks;

[0008] Step 3: Based on the qualitative relationships in the diet-intestinal flora-disease database, corresponding nodes in the correlation network are constructed to construct a diet-intestinal flora-disease network;

[0009] Step 4: Use the classifier optimization model to screen the intestinal flora targets and obtain dietary recommendation results for different diseases.

[0010] Optionally, the step 1 includes:

[0011] Step 11: Search the public database, download the diet-gut flora and gut flora-disease association data, and convert the obtained json files into a unified content format;

[0012] Step 12: Integrate the database content, screen clinical data, high-throughput technology data, and intestinal microbial data, remove data with a p value greater than 0.05, i.e., data with no significant difference and duplicate data, and correct conflicting data to obtain nodes and their associated information in the three dimensions of diet, intestinal flora, and disease;

[0013] Step 13: Search the literature database and include the diet-gut flora clinical data with significant difference results in the literature into the diet-gut flora-disease database;

[0014] Step 14: Uniformly number and name the fields in the diet-gut flora-disease database.

[0015] Optionally, the step 2 includes:

[0016] Step 21: Constructing diet-gut flora and gut flora-disease adjacency matrices;

[0017]

[0018]

[0019] in, represent microbiota-disease and diet-microbe associations, respectively, and N m 、N d 、Nf respectively represent the quantities of microorganisms, diseases, and diets. If there is a verified association between the microbiota m i and the disease d j , diet f i and the microbiota m i , the values of A(i,j) and B(i,j) are set to 1; otherwise, they are set to 0;

[0020] Step 22: Calculate the Gaussian kernel profile similarity based on the adjacency matrix. The calculation process is as follows:

[0021]

[0022] KD(dx,dy) = exp(-γ d ||IP(d x ) - IP(d y )|| 2 )

[0023] where KD(dx,dy) represents the Gaussian kernel profile similarity between disease x and disease y, and IP(d x ) represents the interaction spectrum of disease x with each microorganism in the adjacency matrix, and γ d is the normalized kernel bandwidth. The γ′ d parameter is set to 0.5 according to relevant research;

[0024] The Gaussian kernel profile similarities KD(mx,my) and KD(fx,fy) of microorganisms and diets are calculated in the same way as the above process;

[0025] Step 23: Calculate the disease semantic similarity based on the disease - symptom associations constructed by combining MeSH terms and the PubMed database, and normalize the similarity of diseases. The calculation process is as follows:

[0026] d j = (w 1,j , w 2,j ,..., w n,j )

[0027]

[0028]

[0029]

[0030] where w i,j represents the association strength between symptom i and disease j, W i,j represents the absolute co - occurrence degree between symptom i and disease j, N represents the number of diseases, and N iDenote the number of diseases containing symptom i, and obtain the vectors d of diseases x and y through cosine similarity calculation x and d y Based on symptoms, the disease similarity score SD(dx,dy) between them is calculated, and finally the disease similarity DS(dx,dy) is determined;

[0031] Step 24: According to the microorganism, disease, diet similarity matrix and adjacency matrix, construct a diet-gut microbiota similarity network and a gut microbiota-disease similarity network through a graph attention network respectively. The calculation process is as follows:

[0032] H (l) = f(H (l-1) ,X)

[0033]

[0034]

[0035]

[0036]

[0037]

[0038] H (l) represents the node embedded in the l-th layer, e ij is the attention coefficient between microorganism i and disease j, μ is a single-layer feedforward neural network parameterized by the trainable weight matrix W, which linearly transforms the input feature h. The initial input feature h is a set of nodes, N represents the number of nodes including a group of features, and further normalizes the attention coefficient e ij to α ij , LeakyReLU represents the activation function, || represents the concatenation operation, is the output feature of each node, minLoss is the minimum loss function, where K m and K d are two feature space vectors, W m and W d are training matrices, A is the training set of the microorganism and disease association matrix, |||| F is the Frobenius norm, λ m and λ d are regularization coefficients, L m and L d are Laplacian regularization matrices.

[0039] Optionally, the step 3 includes:

[0040] Step 31: According to the diet-gut microbiota and gut microbiota-disease similarity networks obtained in Step 2, rank the importance of microorganisms for different diseases based on the correlation scores, and select the microorganisms accounting for 80% of the total as intervention targets;

[0041] Step 32: Combine the qualitative knowledge of diet / disease-gut microbiota in the database and the quantitative knowledge in the similarity network, and associate diet with disease through common gut microorganisms to obtain the recommendation coefficient F of diet i for disease j. i D j , which is calculated as follows:

[0042]

[0043] where F i M k represents the correlation coefficient between diet i and microorganism k, and M k D j represents the correlation coefficient between microorganism k and disease j, and the correlation coefficient is corrected by the qualitative knowledge in the database. For example, for positive correlation, it remains unchanged, and for negative correlation, it needs to be negative.

[0044] Optionally, Step 4 includes:

[0045] Step 41: Analyze the feature importance scores of the gut microbiota of the target disease through the metagenomic cohort data of the target disease using a classifier;

[0046] Step 42: Take the intersection of the gut microbiota targets in Step 31 with the gut microbiota targets obtained by the classifier, and redefine the correlation coefficient M k D j through the feature importance scores, so as to improve the accuracy of the gut microbiota targets and the correlation coefficients, and finally obtain an optimized dietary recommendation strategy for different diseases.

[0047] Optionally, the public database includes: Disbiome, GMrepo, HMDAD, MASI, gutMDisoder, Amadis.

[0048] Optionally, in Step 13, "gut microbiota", "fecal microbiota", "Intestinalmicrobiota" and "16S", "Metagenome", "Metagenomic" are used as keywords and combined in pairs for online retrieval in the literature database.

[0049] Optionally, in step 14, diseases are queried one by one through disease terms in the Disease Ontology database, numbered by DOID, and the detection techniques are uniformly named. The gut microbiota is uniformly numbered and named through NCBI.

[0050] The second object of the present invention is to provide an electronic device, including a memory and a processor;

[0051] The memory is used to store a computer program;

[0052] The processor is used to implement the method according to any one of claims 1 to 8 when executing the computer program.

[0053] The third object of the present invention is to provide a computer-readable storage medium, characterized in that a computer program is stored on the storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

[0054] The beneficial effects of the present invention are as follows:

[0055] The diet recommendation method based on diet-gut microbiota-disease of the present invention systematically and comprehensively links diet to human health through gut microbiota, which is of great significance for dietary recommendation targeting the human gut microbiota and health status.

[0056] The present invention combines qualitative knowledge and quantitative knowledge, proposes a new framework based on graph attention network for dietary recommendation for diseases, and introduces metagenomic data and classifier optimization model, which can select the most relevant gut microbiota targets for each disease, improve the interpretability and recommendation reliability of the model, and further improve the accuracy of dietary recommendation.

[0057] Based on the latest research results, the present invention constructs a more comprehensive diet-gut microbiota-disease database, which can facilitate the retrieval of relevant knowledge by scientific researchers. With sufficient experimental data support, hypotheses can be proposed and more in-depth research can be carried out, and it is also convenient to be applied to other machine learning fields. The dietary recommendation model constructed based on this database integrates the most advanced research results of diet-microbiota-disease and can provide more effective dietary recommendation results. Description of the Drawings

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained without creative efforts based on these drawings.

[0059] Figure 1 It is a flow chart of the present invention for constructing a diet-intestinal flora-disease database and a dietary recommendation model. DETAILED DESCRIPTION

[0060] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0061] Embodiment 1:

[0062] This embodiment provides a method for constructing a diet-intestinal flora-disease database. Figure 1 , the method comprising:

[0063] Step 1: Obtain open databases in related fields through retrieval, including Disbiome, GMrepo, HMDAD, MASI, gutMDisoder, and Amadis, a total of 6 databases, download diet-gut flora and gut flora-disease association data through online websites, and convert the given json files into a unified content format.

[0064] Step 2: Integrate the database content. Specifically, remove animal data such as mice and rabbits and in vitro experimental data; only retain human data, remove samples of blood, saliva, skin and other parts, and only retain intestinal flora data; remove low-throughput sequencing technology (such as bacterial culture method, non-sequencing-based molecular technology, etc.); remove antibiotics, fecal microbiota transplantation and other data; finally, screen the data and eliminate data with a p-value greater than 0.05, i.e. data with no significant difference and duplicate data.

[0065] Since most of the data in different databases come from manual inclusion by researchers, inclusion errors may occur, resulting in contradictions in the changes in intestinal flora under the same disease or diet in some of the databases from the same reference. In response to this situation, we corrected it by consulting the original text. The final database contains entries such as diet, intestinal flora, Taxonomy ID, intestinal flora classification, disease, DOID, detection technology, P value, changes in flora abundance, references, etc.

[0066] Specifically, after screening 3,193 records of gutMDisorder, there were 1,704 gut microbiota-disease records. For MASI with 13,191 records, since most were drug-gut microbiota records, there were 166 gut microbiota-disease records after screening. For HMADA with 483 records, there were 206 gut microbiota-disease records after screening. For Amadis with 5,578 records, there were 666 gut microbiota-disease records after screening. For GMrepo with 4,060 records, there were 3,644 gut microbiota-disease data after screening. For Disbiome with 10,816 records, there were 4,842 gut microbiota-disease records after screening. After integrating the databases, there were 11,228 gut microbiota-disease records in FGMDIS. The diet-gut microbiota data consisted of 611 records from gutMDisorder plus 513 records from MASI, for a total of 1,124 records, and there were 379 records left after screening.

[0067] Step 3: Use "gut microbiota", "fecal microbiota", "Intestinal microbiota" and "16S", "Metagenome", "Metagenomic" as keywords to conduct online searches in the literature database in pairs. After removing duplicates, integrate and enter 902 relevant literatures, and include the clinical data of diet-gut microbiota with significant differences in results to construct a diet-gut microbiota-disease database.

[0068] Specifically, the inclusion criteria are that the intervention method is dietary intervention and is a single-factor intervention; the intervention results need to have significant differences; the research needs to be a clinical double-blind experiment; the intervention group has an ideal control group.

[0069] Step 4: In order to standardize the naming in the database, query diseases one by one through the disease terms in the Disease Ontology database, number the diseases with DOID, and finally integrate 322 diseases after querying into 281 diseases, and unify the naming of detection technologies. Through NCBI, uniformly number and name the gut microbiota.

[0070] Example Two

[0071] This example provides a method for constructing a dietary recommendation model. See Figure 1 , and the method includes:

[0072] Step 1: Based on the database constructed in Example One, conduct modeling through semantic similarity, Gaussian kernel profile similarity calculation, and graph attention network to construct a diet-gut microbiota and gut microbiota-disease correlation network. Specifically:

[0073] Step 1.1: Construct the diet-gut microbiota and gut microbiota-disease adjacency matrices;

[0074]

[0075]

[0076] Among them, respectively represent the microbiota-disease and diet-microbiota associations, N m , N d , N f respectively represent the numbers of microorganisms, diseases, and diets. If there is a verified association between the microbiota m i and the disease d j , diet f i and the microbiota m i , the values of A(i,j) and B(i,j) are set to 1; otherwise, they are set to 0.

[0077] Specifically, due to the influence of different factors such as individual differences, regional differences, and eating habits, there are also contradictory situations of gut microbiota changes in the same diet or disease in different references. To make the constructed model more reliable, these entries are deleted when constructing the adjacency matrix according to the database to ensure the rigor of the original data.

[0078] Step 1.2: Calculate the Gaussian kernel profile similarity based on the adjacency matrix. The calculation process is as follows:

[0079]

[0080] KD(dx,dy) = exp(-γ d ||IP(d x ) - IP(d y )|| 2 )

[0081] where IP(d x ) represents the interaction spectrum of disease x with each microorganism in the adjacency matrix, and γ d is the normalized kernel bandwidth. The γ' d parameter is set to 0.5 according to relevant research.

[0082] The Gaussian kernel profile similarities KD(mx,my) and KD(fx,fy) of microorganisms and diets are the same as the above calculation process.

[0083] Step 1.3: Calculate the disease semantic similarity based on the disease-symptom associations constructed by combining MeSH terms and the PubMed database, and normalize the similarity of diseases. The calculation process is as follows:

[0084] dj =(w 1,j , w 2,j ,..., w n,j )

[0085]

[0086]

[0087]

[0088] where w i,j represents the association strength between symptom i and disease j, W i,j represents the absolute co-occurrence degree between symptom i and disease j, N represents the number of diseases, N i represents the number of diseases containing symptom i, and the disease similarity score SD(dx, dy) based on symptoms between the vectors d x and d y of diseases x and y is calculated, and finally the disease similarity DS(dx, dy) is determined.

[0089] Step 1.4: According to the above microbial, disease, and diet similarity matrices and adjacency matrices, construct a diet-gut microbiota similarity network and a gut microbiota-disease similarity network through a graph attention network respectively. The calculation process is as follows:

[0090] H (l) = f(H (l-1) , X)

[0091]

[0092]

[0093]

[0094]

[0095]

[0096] H (l) represents the nodes embedded in the l-th layer, e ij is the attention coefficient between microorganism i and disease j, μ is a single-layer feedforward neural network parameterized by the trainable weight matrix W, which linearly transforms the input feature h. The initial input feature h is a set of nodes, N represents the number of nodes including a group of features, and the attention coefficient e ij is further normalized to α ij by the softmax function, LeakyReLU represents the activation function, and || represents the concatenation operation. For the output features of each node, minLoss is the minimum loss function, where K m and K d are two feature space vectors, W m and W d are training matrices, A is the training set of the microorganism-disease association matrix, |||| F is the Frobenius norm, λ m and λ d are regularization coefficients, L m and L d are Laplacian regularization matrices.

[0097] Step 2: Based on the qualitative relationships in the database, corresponding to the nodes in the correlation network, thereby constructing a diet-gut microbiota-disease network;

[0098] Step 2.1: According to the diet-gut microbiota and gut microbiota-disease similarity networks obtained in Step 1, rank the importance of 1382 microorganisms for different diseases based on the correlation scores, and select the microorganisms accounting for 80% of the total as intervention targets.

[0099] Step 2.2: Combining the qualitative knowledge of diet / disease-gut microbiota in the database and the quantitative knowledge in the similarity network, associate diet with disease through common gut microorganisms, and obtain the recommendation coefficient F i D j for diet i for disease j, calculated as follows:

[0100]

[0101] where, F i M k represents the correlation coefficient between diet i and microorganism k, M k D j represents the correlation coefficient between microorganism k and disease j, and the correlation coefficient is corrected by the qualitative knowledge of the database, such as remaining unchanged for positive correlation and becoming negative for negative correlation.

[0102] Step 3: Screen the gut microbiota targets through the classifier optimization model to obtain the dietary recommendation results for different diseases.

[0103] Step 3.1: Since the number of gut microbiota in gut microbiota-disease is too large, which is 1382, even if microorganisms accounting for 80% of the total are selected as intervention targets, there is still a lack of precision. To optimize the recommendation performance of the model, further analyze the feature importance scores of the gut microbiota of the target disease through the metagenomic cohort data of the target disease using 6 classifiers such as random forest, extreme gradient boosting, gradient boosting, adaptive boosting, and support vector machine. Specifically, select the metagenomic data of three cohorts, PRJNA422434, PRJEB1220, and PRJEB12123, which correspond to type 2 diabetes, ulcerative colitis, and obesity respectively. Among the 6 classifiers, the random forest has the best performance.

[0104] Step 3.2: Take the intersection of the gut microbiota targets obtained by the classifier with the gut microbiota targets in Step 2.1, and redefine the correlation coefficient M through the feature importance scores k D j , so as to improve the precision of the gut microbiota targets and the correlation coefficients, and finally obtain the optimized dietary recommendation strategies for different diseases.

[0105] The present invention integrates the existing open-source diet-gut microbiota-disease database, and at the same time supplements the diet-gut microbiota literature information, provides data support for the interaction of diet-gut microbiota-disease, provides dietary suggestions and target analysis for different diseases, and provides a technical means for scientific researchers to explore the influence mechanism of gut microbiota on human health and the prediction and application of diet in the field of gut microbiota.

[0106] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk, etc.

[0107] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, 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 constructing a dietary recommendation model, characterized in that: The method comprises: Step 1: Build a diet-gut flora-disease database; Step 2: Based on the diet-gut flora-disease database, modeling is performed through semantic similarity, Gaussian kernel profile similarity calculation and graph attention network to construct diet-gut flora and gut flora-disease correlation networks; Step 3: Based on the qualitative relationships in the diet-intestinal flora-disease database, corresponding nodes in the correlation network are constructed to construct a diet-intestinal flora-disease network; Step 4: Screen the intestinal flora targets through the classifier optimization model to obtain dietary recommendation results for different diseases; The step 2 comprises: Step 21: Constructing diet-gut flora and gut flora-disease adjacency matrices; in, represent microbiota-disease and diet-microbe associations, respectively, and N m 、N d 、N f represent the number of microorganisms, diseases, and diets, respectively. If the microbiome m i and diseases j , Diet i and microbiota i If there is a verified association between them, the values ​​of A(i,j) and B(i,j) are set to 1, otherwise they are 0; Step 22: Calculate the Gaussian kernel profile similarity based on the adjacency matrix. The calculation process is as follows: KD(dx,dy)=exp(-γ d ||IP(d x )-IP(d y )|| 2 ) Among them, KD(dx,dy) represents the Gaussian kernel profile similarity between disease x and disease y, IP(d x ) represents the interaction spectrum between disease x and each microorganism in the adjacency matrix, γ d is the normalized kernel bandwidth, γ′ d The parameter was set to 0.5 based on relevant studies; The Gaussian kernel profile similarity KD(mx,my) and KD(fx,fy) of microorganisms and diets are calculated in the same way as above; Step 23: Calculate the disease semantic similarity based on the disease-symptom association constructed by combining the MeSH vocabulary and the PubMed database, and normalize the disease similarity. The calculation process is as follows: d j =(w 1,j ,w 2,j ,...,w n,j ) Among them, w i,j represents the strength of association between symptom i and disease j, W i,j represents the absolute co-occurrence between symptom i and disease j, N represents the number of diseases, and N i Represents the number of diseases containing symptom i, and the vector d of disease x and disease y is obtained by cosine similarity calculation x and d y The symptom-based disease similarity score SD(dx,dy) is used to determine the disease similarity DS(dx,dy); Step 24: Based on the microorganism, disease, diet similarity matrix and adjacency matrix, the diet-gut flora similarity network and the gut flora-disease similarity network are constructed through the graph attention network. The calculation process is as follows: H (l) =f(H (l-1) ,X) H (l) represents the node embedded in the lth layer, e ij is the attention coefficient between microorganism i and disease j, μ is a single-layer feedforward neural network parameterized by a trainable weight matrix W, which linearly transforms the input feature h; the initial input feature h is a collection of nodes, N represents the number of nodes including a set of features, and the attention coefficient e is further transformed by the softmax function. ij Normalized to α ij , LeakyReLU represents the activation function, || represents the connection operation, is the output feature of each node, minLoss is the minimum loss function, where K m and K d are two feature space vectors, W m and W d is the training matrix, A is the training set of microorganism and disease association matrix, |||| F is the Frobenius norm, λ m and λ d is the regularization coefficient, L m and L d is the Laplace regularization matrix; The step 3 comprises: Step 31: According to the diet-gut flora and gut flora-disease similarity networks obtained in step 2, the importance of microorganisms for different diseases is ranked by correlation scores, and microorganisms accounting for 80% of the total are selected as intervention targets; Step 32: Combine the qualitative knowledge of diet / disease-gut flora in the database and the quantitative knowledge in the similarity network, associate diet and disease through common intestinal microorganisms, and obtain the recommendation coefficient F of diet i for disease j i D j , calculated as follows: Among them, F i M k represents the correlation coefficient between diet i and microorganism k, M k D j It represents the correlation coefficient between microorganism k and disease j, and the correlation coefficient is corrected by the qualitative knowledge of the database. If it is positively correlated, it remains unchanged, and if it is negatively correlated, it must be a negative number.

2. The method according to claim 1, characterized in that The step 1 comprises: Step 11: Search the public database, download the diet-gut flora and gut flora-disease association data, and convert the obtained json files into a unified content format; Step 12: Integrate the database content, screen clinical data, high-throughput technology data, and intestinal microbial data, remove data with a p value greater than 0.05, i.e., data with no significant difference and duplicate data, and correct conflicting data to obtain nodes and their associated information in the three dimensions of diet, intestinal flora, and disease; Step 13: Search the literature database and include the diet-gut flora clinical data with significant difference results in the literature into the diet-gut flora-disease database; Step 14: Uniformly number and name the fields in the diet-gut flora-disease database.

3. The method according to claim 2, characterized in that The step 4 comprises: Step 41: Using the metagenomic cohort data of the target disease, using the classifier to perform feature importance score analysis on the intestinal flora of the target disease; Step 42: Take the intersection of the intestinal flora targets obtained by the classifier with the intestinal flora targets in step 31, and redefine the correlation coefficient M by the feature importance score k D j , thereby improving the accuracy of intestinal flora targets and correlation coefficients, and ultimately obtaining optimized dietary recommendation strategies for different diseases.

4. The method according to claim 2, characterized in that: The public databases include: Disbiome, GMrepo, HMDAD, MASI, gutMDisoder, and Amadis.

5. The method according to claim 2, characterized in that: In step 13, "gut microbiota", "fecal microbiota", "Intestinal microbiota" and "16S", "Metagenome" and "Metagenomic" are used as keywords in pairs to perform online retrieval of literature databases.

6. The method according to claim 2, characterized in that In step 14, diseases are searched one by one through disease terms in the Disease Ontology database, diseases are numbered with DOID, and detection technologies are uniformly named. Intestinal flora are uniformly numbered and named through NCBI.

7. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the method according to any one of claims 1 to 6 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images