A method for donor-recipient matching in gut microbiota transplantation for chronic constipation based on multidimensional microbiota characteristics and machine learning, and its application.

Abstract

This invention relates to a method for donor-recipient matching in intestinal microbiota transplantation for chronic constipation based on multidimensional microbiota characteristics and machine learning, and its application. By collecting 16S rRNA intestinal microbiota data of donors and recipients before intestinal microbiota transplantation (FMT) treatment, as well as retrospective data on the clinical efficacy of FMT, the invention calculates characteristics such as differences in microbiota abundance, metabolic pathway activity, enzyme activity, and species diversity between donors and recipients. Principal component analysis is used to select high-contribution features that can explain 90% of the variance. A random forest machine learning model is used to select feature markers related to treatment efficacy, and the weights of these markers in the model are calculated. Finally, the optimal donor is selected by weighted scoring and ranking of the microbiota characteristics of each donor.

Description

Technical Field

[0001] This invention relates to the field of donor-recipient matching for intestinal flora transplantation, and more particularly to a method for donor-recipient matching for intestinal flora transplantation in chronic constipation based on multidimensional flora characteristics and machine learning, and its application. Background Technology

[0002] Chronic constipation is a common gastrointestinal disease with a rising incidence, severely impacting patients' quality of life and mental health, while also imposing a significant economic burden. Although drug therapy is one of the main management methods, its adverse reactions and poor efficacy in some patients have prompted the search for new treatments. Studies have shown a close link between the gut microbiota and constipation; constipation patients often exhibit abnormal gut microbiota structure, dysregulation, and low microbiota diversity. Adjusting the gut microbiota can shorten colonic transit time, increase defecation frequency, and alleviate symptoms. Fecal microbiota transplantation (FMT), an innovative therapy that transplants healthy donor microbiota into a patient, has shown some effectiveness in the treatment of chronic constipation.

[0003] CN114181855A discloses a gut microbiota preparation for directional proliferation of gut microbiota transplantation, its preparation method, and its application. The preparation includes GAM medium in a volume ratio of 2:2:1:(0.5–1.5), aloe-emodin solution, PBS buffer, and gut microbiota solution. The concentration of the aloe-emodin solution is 9.93–10.13 g / mL, and the mass concentration of the gut microbiota solution is 4–6%.

[0004] Currently, the clinical remission rate of FMT is only about 70%, and some patients do not benefit. Studies have shown that donor screening based on microbial characteristics can improve the success rate of FMT. 16S rRNA sequencing technology is the mainstream method for studying the characteristics of gut microbiota. Through sequencing and related analysis tools, species information and functional characteristics of microorganisms can be obtained, which provides basic support for calculating the matching degree between donors and recipients.

[0005] Improving the success rate of FMT treatment, screening for donor and recipient microbial biomarkers that affect efficacy, and calculating the contribution rate or weight of these biomarkers to treatment effects have become one of the urgent problems to be solved in this field. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for donor-recipient matching intestinal microbiota transplantation for chronic constipation based on multidimensional microbiota characteristics and machine learning, and its application. The aim is to construct an algorithm model for precise matching by analyzing the 16S rRNA intestinal microbiota sequencing data of donors and recipients, thereby helping to select the optimal donor and improving the success rate of FMT treatment for constipation patients.

[0007] To achieve this objective, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a method for donor-recipient matching in intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning, the method comprising the following steps:

[0009] (1) Collect 16S rRNA sequencing data of gut microbiota from donors and recipients, filter and cluster to obtain ASV (Amplicon Sequence Variant) sequences;

[0010] (2) Perform species annotation and functional prediction on ASV sequence data to obtain species diversity index, relative abundance of bacterial community at the species level, metabolic pathway activity and metabolic pathway-related enzyme activity characteristics.

[0011] (3) Subtract the species diversity index, relative abundance of bacterial species, metabolic pathway activity and metabolic pathway-related enzyme activity characteristics of each pair of donors and recipients to obtain the characteristic difference between donors and recipients. A multidimensional characteristic difference matrix is ​​formed with the sample number of donors and recipients as the row and the characteristics as the column.

[0012] (4) Perform principal component analysis (PCA) to reduce the dimensionality of the matrix obtained in step (3) and screen out high contribution features that can explain 90% of the variance;

[0013] (5) The selected features are integrated with the data on whether each pair of donor and recipient responds to clinical treatment, and a matrix is ​​generated with the donor and recipient sample number combination as the row name and the microbial characteristics and clinical indicators as the column names.

[0014] (6) Input the matrix obtained in step (5), divide it into training set and test set according to the ratio, find the optimal parameter combination after parameter grid search, construct random forest classifier, use training set data for model fitting, use test set data for prediction verification, draw ROC curve, calculate AUC value, and obtain the importance index of each feature.

[0015] (7) Input the recipient data to be paired, pair with each donor, calculate the high contribution feature difference, construct the feature difference matrix, score the donor according to the importance index obtained in step (6) using the formula, sort the calculated donor scores, and the one with the highest score is the best donor.

[0016] In this invention, 16S rRNA gut microbiota data of donors and recipients before fecal microbiota transplantation (FMT) treatment and retrospective data on the clinical efficacy of FMT were collected. Differences in microbiota abundance, metabolic pathway activity, enzyme activity, and species diversity between donors and recipients were calculated. Principal component analysis was used to select high-contribution features that could explain 90% of the variance. A random forest machine learning model was used to screen for feature markers related to treatment efficacy, and the weights of these markers in the model were calculated. Finally, the optimal donor was selected by weighted scoring and ranking of the microbiota features of each donor. This invention aims to construct a precise matching algorithm model by analyzing 16S rRNA gut microbiota sequencing data of donors and recipients to help select the optimal donor and improve the success rate of FMT treatment for constipation patients.

[0017] Preferably, step (1) specifically includes: collecting 16S rRNA sequencing data of gut microbiota of donors and recipients before fecal microbiota transplantation, using QIIME2 software for quality control, low-quality sequence filtering, primer detection and removal, and using the DADA2 algorithm to cluster the sequences to obtain ASV sequences.

[0018] Preferably, step (1) further includes: the recipient receiving donor fecal microbiota transplantation therapy and collecting data on whether the recipient responds to the therapy.

[0019] Preferably, step (2) specifically includes: using the Naive Bayes classifier (class-sklearn) built into the QIIME2 software to compare the ASV sequences with the Silva database (version: Silva-138-99) and the Greengenes2 database (version: 2022.10) to obtain the species annotation results and relative abundance data of the bacterial community at the species level; using the vegan package in R software to calculate the species diversity index (such as Shannon, Simpson, Fisher, etc.) of each sample; and using the PICRUSt2 software to perform functional annotation on the ASV sequences to predict the metabolic pathway activity and metabolic pathway-related enzyme activity of each sample.

[0020] Preferably, step (4) specifically includes: using the sklearn toolkit of Python 3.9 software to perform principal component analysis to reduce the dimensionality of the matrix obtained in step (3), wherein the species diversity index is not reduced in dimensionality, calculating the top few principal components whose cumulative proportion of the explained total variance is greater than 90%, and looping through these principal components, selecting the top 5 features with the highest contribution and a contribution greater than 0.05 in each principal component, merging and removing duplicates to obtain a set of high contribution features.

[0021] Preferably, step (6) specifically includes the following steps:

[0022] (a) The donor and recipient data were divided into training and test sets in a 7:3 ratio using the sklearn toolkit of Python 3.9 software;

[0023] (b) The optimal parameter combination is found by performing a parameter grid search using the GridSearchCV function. The parameters include min_samples_split, min_samples_leaf, max_depth, max_features, and n_estimators. cv is set to 5 and scoring is set to roc_auc. Finally, the parameter combination with the highest score is found.

[0024] (c) Based on the parameters obtained in step (b), construct a random forest classifier, fit the model with the training set data, and perform prediction verification with the test set data. Plot the ROC curve and calculate the AUC value.

[0025] (d) Based on the trained random forest model, obtain the importance index of each feature through the feature_importances attribute.

[0026] Preferably, step (7) specifically includes: inputting the recipient data to be paired, pairing with each donor, calculating the difference in species diversity index, relative abundance of microbial community at the species level, difference in metabolic pathway activity, and difference in activity of metabolic pathway-related enzymes among the high-contribution features, constructing a feature difference matrix, scoring the donors according to the importance index obtained in step (6) using a formula, ranking the calculated donor scores, and selecting the donor with the highest score as the optimal donor. The formula is as follows:

[0027]

[0028] Where, x i w represents the value of the i-th feature in each sample. i Represent each feature x i The corresponding weights, where n represents the total number of features. MinMaxScaler represents the weighted sum of all features, which scales the feature values ​​to the range of 0-1, and m represents the donor's health score constant during clinical examination.

[0029] In a second aspect, the present invention provides an apparatus for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning, the apparatus being used to perform the steps in the method for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning described in the first aspect.

[0030] Preferably, the device includes:

[0031] The data collection and processing module is used to collect 16S rRNA sequencing data of gut microbiota from donors and recipients, and obtain ASV sequences after filtering and clustering.

[0032] The species annotation and function prediction module is used to obtain species diversity index, relative abundance of bacterial community at the species level, metabolic pathway activity, and metabolic pathway-related enzyme activity characteristics based on ASV sequence data.

[0033] The multidimensional feature difference matrix module is used to create a multidimensional feature difference matrix with the donor and recipient sample numbers as rows and the feature difference between donors and recipients as columns.

[0034] The principal component analysis module is used to reduce the dimensionality of the multidimensional feature difference matrix and filter out high-contribution features.

[0035] The Microbial Characteristics and Clinical Indicators Matrix module is used to create a matrix with the combination of recipient sample numbers as row names and the screened microbial characteristics and clinical indicators as column names.

[0036] The Random Forest model module is used to build a random forest and obtain the importance index of each feature;

[0037] The donor-recipient matching module is used to input the data of the recipient to be matched, score the donors, and determine the optimal donor.

[0038] Preferably, the data collection and processing module is used to perform the following: collecting 16S rRNA sequencing data of gut microbiota of donors and recipients before fecal microbiota transplantation, performing quality control, low-quality sequence filtering, primer detection and removal using QIIME2 software, and clustering the sequences using the DADA2 algorithm to obtain ASV sequences.

[0039] Preferably, the species annotation and functional prediction module is used to perform the following: using the Naive Bayes classifier built into the QIIME2 software to compare ASV sequences with the Silva database and the Greengenes2 database to obtain the species annotation results and relative abundance data of the bacterial community at the species level; using the vegan package in the R software to calculate the species diversity index of each sample; and using the PICRUSt2 software to perform functional annotation of the ASV sequences to predict the metabolic pathway activity and metabolic pathway-related enzyme activity of each sample.

[0040] Preferably, the multidimensional feature difference matrix module is used to perform the following: subtracting the species diversity index, relative abundance of bacterial species, metabolic pathway activity, and activity of metabolic pathway-related enzymes for each pair of donors and recipients to obtain the feature difference between donors and recipients, and forming a multidimensional feature difference matrix with the donor and recipient sample number as the row and the feature as the column.

[0041] Preferably, the principal component analysis module is used to perform the following: using the sklearn toolkit of Python 3.9 software to perform principal component analysis to reduce the dimensionality of the matrix obtained in step (3), wherein the species diversity index is not reduced in dimensionality, calculating the top few principal components whose cumulative proportion of the explained total variance is greater than 90%, and looping through these principal components, selecting the top 5 features with the highest contribution and a contribution greater than 0.05 in each principal component, merging and removing duplicates to obtain a set of high contribution features.

[0042] Preferably, the microbial characteristics and clinical indicator matrix module is used to perform the following: integrating the screened characteristics with data on whether each pair of donor and recipient responds to clinical treatment, generating a matrix with donor and recipient sample numbers as row names and microbial characteristics and clinical indicators as column names.

[0043] Preferably, the random forest model module is used to perform the following steps:

[0044] (a) The donor and recipient data were divided into training and test sets in a 7:3 ratio using the sklearn toolkit of Python 3.9 software;

[0045] (b) The optimal parameter combination is found by performing a parameter grid search using the GridSearchCV function. The parameters include min_samples_split, min_samples_leaf, max_depth, max_features, and n_estimators. cv is set to 5 and scoring is set to roc_auc. Finally, the parameter combination with the highest score is found.

[0046] (c) Based on the parameters obtained in step (b), construct a random forest classifier, fit the model with the training set data, and perform prediction verification with the test set data. Plot the ROC curve and calculate the AUC value.

[0047] (d) Based on the trained random forest model, obtain the importance index of each feature through the feature_importances attribute.

[0048] Preferably, the donor-recipient matching module is used to perform the following: inputting recipient data to be paired, pairing with each donor, calculating the difference in species diversity index, relative abundance of microbial community at the species level, difference in metabolic pathway activity, and difference in activity of metabolic pathway-related enzymes among the high-contribution features, constructing a feature difference matrix, scoring the donors according to the importance index obtained in step (6) using a formula, ranking the calculated donor scores, and selecting the donor with the highest score as the optimal donor. The formula is as follows:

[0049]

[0050] Where, x i w represents the value of the i-th feature in each sample. i Represent each feature x i The corresponding weights, where n represents the total number of features. MinMaxScaler represents the weighted sum of all features, which scales the feature values ​​to the range of 0-1, and m represents the donor's health score constant during clinical examination.

[0051] Thirdly, the present invention provides a computer-readable storage medium storing a computer program or instructions, which, when executed by a processor, implement the steps in the method for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning as described in the first aspect.

[0052] Fourthly, the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program or instructions, the computer program or instructions being executed by the processor to implement the steps in the method for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning as described in the first aspect.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) This invention successfully developed a highly efficient and accurate donor-recipient matching scoring system by applying machine learning models, which can greatly improve the success rate and clinical efficacy of FMT treatment. From multiple perspectives of the structure and function of the gut microbiota, considering characteristics such as species diversity, microbiota abundance, metabolic pathway activity, and enzyme activity, principal component analysis was used to screen out the features that contribute most to the donor-recipient difference based on a 90% variance explained rate, thereby achieving efficient and accurate feature selection.

[0055] (2) This invention uses the response of FMT clinical treatment as the target variable, uses the random forest machine learning algorithm to build a model and evaluate the weight of features. Through the accurate calculation of feature weights, it ensures that the selected features have significant predictive value for the effect of FMT treatment.

[0056] (3) The present invention constructs a donor-recipient matching scoring system. This system calculates the matching score of each donor-recipient pair by combining the key features selected and their corresponding weights. Based on these scores, the donors are ranked to ensure that the optimal donor is selected, thereby significantly improving the clinical response rate and treatment effect of FMT. Attached Figure Description

[0057] Figure 1The flowchart of the donor-recipient matching method provided by the present invention;

[0058] Figure 2 Calculate the cumulative variance distribution plot for PCA dimensionality reduction;

[0059] Figure 3 The results of dimensionality reduction calculation of high-contribution bacterial communities using PCA are shown in the figure.

[0060] Figure 4 Here is the ROC curve for the random forest model;

[0061] Figure 5 Box plot of clinical treatment response and typing score for paired samples. Detailed Implementation

[0062] To further illustrate the technical means and effects of this invention, the following description, in conjunction with embodiments and accompanying drawings, provides a further explanation of the invention. It is understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it.

[0063] Where specific techniques or conditions are not specified in the examples, they shall be performed in accordance with the techniques or conditions described in the literature in this field, or in accordance with the product instructions. Reagents or instruments whose manufacturers are not specified are all conventional products that can be purchased through legitimate channels.

[0064] Example 1

[0065] This embodiment collected 86 recipient samples in a clinical study. Donor selection was based on the clinical health status score of each donor in the donor bank. All recipients underwent FMT treatment after donor selection. 16S rRNA gut microbiota sequencing data of the recipients and their corresponding donors were collected before FMT treatment, and clinical follow-up observation was conducted after treatment. The final results showed that 68 recipients exhibited a significant treatment response after FMT treatment, while 18 recipients failed to respond. Data processing was performed after collection, as shown in the flowchart below. Figure 1 As shown, it includes the following steps:

[0066] (1) Collect 16S rRNA sequencing data of gut microbiota of donors and recipients before fecal microbiota transplantation. Use QIIME2 software for quality control, low-quality sequence filtering, primer detection and removal. Use DADA2 algorithm to cluster the sequences to obtain ASV sequences.

[0067] (2) Using the built-in Naive Bayes classifier (class-sklearn) in QIIME2 software, ASV sequences were compared with the Silva database (version: Silva-138-99) and the Greengenes2 database (version: 2022.10) to obtain the species annotation results and relative abundance data of bacterial community species. The species diversity index (such as Shannon, Simpson, Fisher, etc.) of each sample was calculated using the vegan package in R software. The ASV sequences were functionally annotated using PICRUSt2 software to predict the metabolic pathway activity and metabolic pathway-related enzyme activity of each sample.

[0068] (3) Subtract the species diversity index, relative abundance of bacterial species, metabolic pathway activity and metabolic pathway-related enzyme activity characteristics of each pair of donor and recipient to obtain the characteristic difference between donor and recipient. A multidimensional characteristic difference matrix is ​​formed with the donor and recipient sample number as the row and the characteristics as the column.

[0069] (4) Principal component analysis (PCA) was used to reduce the dimensionality of the matrix obtained in step (3) using the sklearn toolkit of Python 3.9. The cumulative variance distribution plot of the PCA dimensionality reduction is shown in the figure below. Figure 2 As shown, the species diversity index was not dimensionality-reduced. A total of 441 bacterial species-level abundance features, 370 metabolic pathway activity features, and 1899 related enzyme activities were used in the dimensionality reduction calculation. The top principal components that explained more than 90% of the total variance were calculated. These principal components were then iterated, and the top 5 features with a contribution greater than 0.05 in each principal component were selected. After merging and removing duplicates, a set of high-contribution features was obtained.

[0070] A total of 21 high-contribution bacterial groups were screened, and the results are as follows: Figure 3 shown, respectively Clostridium_s ensu_stricto_1, Akkermansia, Megamonas, Escherichia-Shigella, Veillonella, Lachnoclostridium, Bacteroides, Prevotella, Fusobacterium, Faecalibacterium, Phascolarcto bacterium, Megasphaera, Collinsella, Blautia, Bifidobacterium, Sutterella, unclass_Enterobacteriaceae_genus, Agathobacter, Alistipes, Parasutterella, Lactobacillus.

[0071] A total of 52 high-contribution metabolic pathway activity signatures were identified: PWY-5676, PWY-6588, DTDPRHAMSYN-PWY, PWYG-321, PWY0-862, PWY-7664, and FASYN-ELON. G-PWY, PWY-5989, PWY-7211, PWY-6282, PWY0-1261, P162-PWY, PWY-7377, FASYN-INITIAL-PWY, PWY-2941, PWY-621, P42-PWY, PWY-59 13. PWY-6612, PWY0-1586, TCA, PWY-7242, PRPP-PWY, FOLSYN-PWY, P461-PWY, PWY-5484, PWY-7237, PWY-5971, GLUCARDEG-PWY, SALVAD EHYPOX-PWY, PWY-6609, HEXITOLDEGSUPER-PWY, GALACTUROCAT-PWY, PWY-7663, NONOXIPENT-PWY, COBALSYN-PWY, P108-PWY, PW Y-6969, PWY-7111, PWY-6147, PWY-6737, PWY0-1296, PWY-7371, PWY-5509, PWY-7539, PWY-6471, P164-PWY, CALVIN-PWY, FUCCAT-PWY, P WY-6749, PWY-7220, PWY-7222.

[0072] A total of 30 high-contribution enzyme activity characteristics were screened: EC: 2.7.1.69, EC: 1.3.99.33, EC: 3.2.1.21, EC: 1.2.7.3, EC: 2.7.7.7, EC: 2.8.2.22, EC: 3.6.4.12, EC: 1.12.1.3, EC: 2.7.13.3, EC: 1.3.5.4, EC: 3.2.1.23, EC: 3.2.1.51, EC: 2.2.1.6, EC: 1.6.5.3, EC: 3 .1.21.3, EC: 5.2.1.8, EC: 5.99.1.2, EC: 4.1.1.37, EC: 3.6.3.34, EC: 3.6.3.17, EC: 3.6.4.13, EC: 3.4.16.4, EC :3.2.1.52, EC:3.1.3.48, EC:5.4.99.2, EC:4.1.2.13, EC:1.97.1.4, EC:2.1.1.72, EC:6.4.1.2, EC:3.4.19.13.

[0073] (5) The 21 bacterial abundance features, 52 metabolic pathway features, 30 enzyme activity features, and 12 diversity index features selected above are integrated with the data on whether each pair of donor and recipient responds to clinical treatment. A matrix is ​​generated with donor and recipient sample number combination as row name, 115 microbial features, and 1 clinical indicator as column name, which is used as input for the machine learning model.

[0074] (6) Input the matrix obtained in step (5) and construct a random forest model to evaluate the feature weights, including the following steps:

[0075] (a) The donor and recipient data were divided into training and test sets in a 7:3 ratio using the sklearn toolkit of Python 3.9 software;

[0076] (b) The optimal parameter combination is found by performing a parameter grid search using the GridSearchCV function. The parameters include min_samples_split, min_samples_leaf, max_depth, max_features, and n_estimators. cv is set to 5 and scoring is set to roc_auc. Finally, the parameter combination with the highest score is found, with n_estimators set to 1500, min_samples_split set to 4, min_samples_leaf set to 3, max_depth set to 10, and max_features set to 5.

[0077] (c) Based on the parameters obtained in step (b), construct a random forest classifier, fit the model using the training set data, and validate the predictions using the test set data, such as... Figure 4 Plot the ROC curve as shown and calculate the AUC value;

[0078] (d) Based on the trained random forest model, obtain the importance index of each feature through the feature_importances attribute to provide a basis for subsequent donor selection.

[0079] (7) The donor is scored according to the importance index obtained in step (6) using the following formula:

[0080]

[0081] Where, x i w represents the value of the i-th feature in each sample. i Represent each feature x i The corresponding weights, where n represents the total number of features. MinMaxScaler represents the weighted sum of all features, which scales the feature values ​​to the range of 0-1, and m represents the donor's health score constant during clinical examination.

[0082] Based on clinical feedback regarding the response to FMT treatment, the matching scores of the response group and the non-responding group were compared, and the statistical t-test was used to assess the significant differences between the two groups. The results are as follows: Figure 5 As shown in the figure. The analysis results showed that the matching score of the treatment response group was significantly higher than that of the non-response group, with a statistical p-value of 0.00295. This result indicates that the method provided by this invention can effectively improve the matching accuracy between donors and recipients, thereby significantly improving the clinical success rate of FMT treatment.

[0083] Example 2

[0084] A 56-year-old female patient (hereinafter referred to as Ms. A) visited the hospital due to long-term chronic constipation. She only had bowel movements 1-2 times a week, accompanied by significant abdominal bloating and discomfort. Traditional drug treatments and other methods had little effect on her. The doctor suggested that she try FMT as a treatment option and use the matching scoring system provided by this invention to select the most suitable donor in order to obtain the best treatment results.

[0085] Ms. A's gut microbiota profile data was analyzed using QIIME2 and PICRUSt2, followed by matching and scoring with all donors in the donor database. The scoring system automatically calculated a feature difference matrix between each donor and Ms. A, covering multiple dimensions such as microbial species diversity, abundance differences, and metabolic pathway activity, and calculated a matching score based on a built-in formula. The final scoring results showed that Ms. A's scores with donors ranged from 60 to 95 points, with donor D-102 scoring 93 points, higher than other donors, and thus selected as the optimal donor. Donor D-102's gut microbiota was highly matched with Ms. A in terms of species diversity and functional characteristics, and its health examination score also indicated that the donor was in good clinical condition. This scientific and comprehensive matching process ensured that Ms. A would benefit best from FMT treatment.

[0086] After identifying the optimal donor, Ms. A underwent FMT treatment. Following treatment, her bowel movement frequency significantly improved from 1-2 times per week to once daily, and her abdominal bloating symptoms also improved markedly. During the subsequent 3-month follow-up, Ms. A's symptoms continued to improve, and no significant side effects or complications occurred. Through the donor-recipient matching method provided by this patent, Ms. A successfully selected the most suitable donor, significantly improving the effectiveness of FMT treatment and further demonstrating the device's effectiveness in increasing FMT success rates and improving patients' clinical symptoms.

[0087] In summary, this invention provides a method for donor-recipient matching intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning. By applying this method, donors with higher matching degrees to recipients can be screened more accurately, improving the overall effect of FMT treatment and providing reliable technical support and scientific basis for future clinical practice.

[0088] The applicant declares that the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention fall within the protection and disclosure scope of the present invention.

Claims

1. A method for donor-recipient matching in intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning, characterized in that, The method includes the following steps: (1) Collect 16S rRNA sequencing data of gut microbiota from donors and recipients, filter and cluster to obtain ASV sequences; (2) Perform species annotation and functional prediction on ASV sequence data to obtain species diversity index, relative abundance of bacterial community at the species level, metabolic pathway activity and metabolic pathway-related enzyme activity characteristics. (3) Subtract the species diversity index, relative abundance of bacterial species, metabolic pathway activity and metabolic pathway-related enzyme activity characteristics of each pair of donors and recipients to obtain the characteristic difference between donors and recipients. A multidimensional characteristic difference matrix is ​​formed with the donor and recipient sample number as the row and the characteristics as the column. (4) Perform principal component analysis to reduce the dimensionality of the matrix obtained in step (3) and screen out the high contribution features that can explain 90% of the variance; (5) The selected features are integrated with the data on whether each pair of donor and recipient responds to clinical treatment, and a matrix is ​​generated with donor and recipient sample number combination as row name and microbial characteristics and clinical indicators as column name. (6) Input the matrix obtained in step (5), divide it into training set and test set according to the ratio, find the optimal parameter combination after parameter grid search, construct random forest classifier, use training set data for model fitting, use test set data for prediction verification, draw ROC curve, calculate AUC value, and obtain the importance index of each feature. (7) Input the recipient data to be paired, pair with each donor, calculate the high contribution feature difference, construct the feature difference matrix, score the donor according to the importance index obtained in step (6) using the formula, sort the calculated donor scores, and the one with the highest score is the best donor; The formula is as follows: in, This represents the value of the i-th feature in each sample. Represent each feature The corresponding weights, where n represents the total number of features. Represents the weighted sum of all features. This represents scaling the feature values ​​to the range of 0-1, where m represents the donor's health status score constant during clinical examination.

2. The method according to claim 1, characterized in that, Step (1) specifically includes: collecting 16S rRNA sequencing data of gut microbiota of donors and recipients before fecal microbiota transplantation, using QIIME2 software for quality control, low-quality sequence filtering, primer detection and removal, and using the DADA2 algorithm to cluster the sequences to obtain ASV sequences.

3. The method according to claim 1, characterized in that, Step (1) also includes: the recipient receiving donor fecal microbiota transplantation therapy and collecting data on whether the recipient responds to the therapy.

4. The method according to claim 1, characterized in that, Step (2) specifically includes: using the Naive Bayes classifier built into the QIIME2 software to compare the ASV sequences with the Silva database and the Greengenes2 database to obtain the species annotation results and relative abundance data of the bacterial community at the species level; using the vegan package in the R software to calculate the species diversity index of each sample; and using the PICRUSt2 software to perform functional annotation on the ASV sequences to predict the metabolic pathway activity and metabolic pathway-related enzyme activity of each sample.

5. The method according to claim 1, characterized in that, Step (4) specifically includes: using the sklearn toolkit of Python 3.9 software to perform principal component analysis to reduce the dimensionality of the matrix obtained in step (3), while the species diversity index is not reduced in dimensionality. The top principal components with a cumulative proportion of total variance explained greater than 90% are calculated, and these principal components are iterated. The top 5 features with the highest contribution and a contribution greater than 0.05 in each principal component are selected, and after merging and removing duplicates, a set of high contribution features is obtained.

6. The method according to claim 1, characterized in that, Step (6) specifically includes the following steps: (a) The donor and recipient data were divided into training and test sets in a 7:3 ratio using the sklearn toolkit of Python 3.9 software; (b) The optimal parameter combination is found by performing a parameter grid search using the GridSearchCV function. The parameters include min_samples_split, min_samples_leaf, max_depth, max_features and n_estimators. cv is set to 5 and scoring is set to roc_auc. Finally, the parameter combination with the higher score is found. (c) Based on the parameters obtained in step (b), construct a random forest classifier, fit the model with the training set data, and perform prediction verification with the test set data. Plot the ROC curve and calculate the AUC value. (d) Based on the trained random forest model, obtain the importance index of each feature through the feature_importances attribute.

7. The method according to claim 1, characterized in that, Step (7) specifically includes: inputting the recipient data to be paired, pairing with each donor, calculating the difference in species diversity index, relative abundance of microbial community species, metabolic pathway activity, and metabolic pathway-related enzyme activity among the high contribution features, constructing a feature difference matrix, scoring the donors according to the importance index obtained in step (6) using a formula, ranking the calculated donor scores, and selecting the donor with the highest score as the optimal donor.

8. A device for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning, characterized in that, The device is used to perform the steps in the method for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning as described in any one of claims 1 to 7.

9. The apparatus according to claim 8, characterized in that, The device includes: The data collection and processing module is used to collect 16S rRNA sequencing data of gut microbiota from donors and recipients, and obtain ASV sequences after filtering and clustering. The species annotation and function prediction module is used to obtain species diversity index, relative abundance of bacterial community at the species level, metabolic pathway activity, and metabolic pathway-related enzyme activity characteristics based on ASV sequence data. The multidimensional feature difference matrix module is used to build a multidimensional feature difference matrix with the donor and recipient sample numbers as rows and the feature difference between donors and recipients as columns. The principal component analysis module is used to reduce the dimensionality of the multidimensional feature difference matrix and filter out high-contribution features. The Microbial Characteristics and Clinical Indicators Matrix module is used to create a matrix with the combination of recipient sample numbers as row names and the screened microbial characteristics and clinical indicators as column names. The Random Forest model module is used to build a random forest and obtain the importance index of each feature; The donor-recipient matching module is used to input the data of the recipient to be matched, score the donors, and determine the optimal donor.

10. The apparatus according to claim 9, characterized in that, The data collection and processing module is used to perform the following: collecting 16S rRNA sequencing data of gut microbiota from donors and recipients before fecal microbiota transplantation; using QIIME2 software for quality control, low-quality sequence filtering, primer detection and removal; and using the DADA2 algorithm to cluster the sequences to obtain ASV sequences.

11. The apparatus according to claim 9, characterized in that, The species annotation and functional prediction module is used to perform the following: using the Naive Bayes classifier built into the QIIME2 software to compare ASV sequences with the Silva and Greengenes2 databases to obtain the species annotation results and relative abundance data of the bacterial community at the species level; using the vegan package in the R software to calculate the species diversity index of each sample; and using the PICRUSt2 software to perform functional annotation of ASV sequences and predict the metabolic pathway activity and metabolic pathway-related enzyme activity of each sample.

12. The apparatus according to claim 9, characterized in that, The multidimensional feature difference matrix module is used to perform the following: subtracting the species diversity index, relative abundance of bacterial species, metabolic pathway activity, and activity of metabolic pathway-related enzymes for each pair of donors and recipients to obtain the feature difference between donors and recipients, forming a multidimensional feature difference matrix with the donor and recipient sample number as the row and the feature as the column.

13. The apparatus according to claim 9, characterized in that, The principal component analysis module is used to perform the following: using the sklearn toolkit of Python 3.9 software to perform principal component analysis to reduce the dimensionality of the matrix obtained in step (3), wherein the species diversity index is not reduced in dimensionality, calculating the top few principal components whose cumulative proportion of the explained total variance is greater than 90%, and looping through these principal components, selecting the top 5 features with the highest contribution and a contribution greater than 0.05 in each principal component, merging and removing duplicates to obtain a set of high contribution features.

14. The apparatus according to claim 9, characterized in that, The microbial characteristics and clinical indicator matrix module is used to perform the following: integrating the screened characteristics with data on whether each pair of donor and recipient responds to clinical treatment, generating a matrix with donor and recipient sample numbers as row names and microbial characteristics and clinical indicators as column names.

15. The apparatus according to claim 9, characterized in that, The random forest model module is used to perform the following steps: (a) The donor and recipient data were divided into training and test sets in a 7:3 ratio using the sklearn toolkit of Python 3.9 software; (b) The optimal parameter combination is found by performing a parameter grid search using the GridSearchCV function. The parameters include min_samples_split, min_samples_leaf, max_depth, max_features and n_estimators. cv is set to 5 and scoring is set to roc_auc. Finally, the parameter combination with the higher score is found. (c) Based on the parameters obtained in step (b), construct a random forest classifier, fit the model with the training set data, and perform prediction verification with the test set data. Plot the ROC curve and calculate the AUC value. (d) Based on the trained random forest model, obtain the importance index of each feature through the feature_importances attribute.

16. The apparatus according to claim 9, characterized in that, The donor-recipient matching module is used to perform the following: inputting recipient data to be paired, pairing with each donor, calculating the difference in species diversity index, relative abundance of microbial community at the species level, difference in metabolic pathway activity, and difference in activity of metabolic pathway-related enzymes among the high-contribution features, constructing a feature difference matrix, scoring the donors according to the importance index obtained in step (6) using a formula, ranking the calculated donor scores, and selecting the donor with the highest score as the optimal donor. The formula is as follows: in, This represents the value of the i-th feature in each sample. Represent each feature The corresponding weights, where n represents the total number of features. Represents the weighted sum of all features. This represents scaling the feature values ​​to the range of 0-1, where m represents the donor's health status score constant during clinical examination.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program or instructions, which, when executed by a processor, implement the steps of the method for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning as described in any one of claims 1 to 7.

18. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores computer programs or instructions, which, when executed by the processor, implement the steps in the method for donor-recipient matching of intestinal flora transplantation for chronic constipation based on multidimensional flora characteristics and machine learning, as described in any one of claims 1 to 7.

Images