A method and system for intestinal microecological management based on digital twins

Abstract

The present invention provides an intestinal microecological management method and system based on digital twins, and a digital twin model based on conditional inference forest, which involves prediction of intestinal microbial abundance development trajectory, disease risk prediction, microbial flora network analysis and clinical intervention decision support, etc. The present invention bins and quantifies the relative abundance data of each microbial group into discrete data, and regards the abundance of each microbial group at each time point as a feature, and constructs a digital twin model CIT‑Net based on conditional inference forest, which contains multiple conditional inference trees, each of which is used to predict the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions; using the CIT‑Net model, the intestinal microbial abundance development trajectory prediction, disease risk prediction, etc. can be performed based on a small amount of initial observation data, providing a basis for early intervention, and improving the treatment effect through personalized intervention suggestions.

Description

Technical Field

[0001] The present invention relates to the fields of microecological health management, bioinformatics, artificial intelligence and medical health, and in particular to the health management of the intestinal microecological system of infants and young children using digital twin technology. Background Art

[0002] In recent years, more and more studies have shown that the intestinal microbiota is closely related to human health, especially in infant development. Studies have shown that the composition and function of the intestinal microbiota can affect brain development and neurobehavior. However, due to the large differences in microbiota between individuals, it is difficult to determine the normal range of microbiota, which limits the understanding of the impact of intestinal flora imbalance on infant development.

[0003] Existing research methods mainly focus on analyzing the correlation between infant gut microbiota and neurodevelopmental defects, for example, by comparing the gut microbiota of children with a certain disease with those without the disease. However, these methods cannot predict the risk of individual diseases or guide clinical interventions.

[0004] In summary, the intestinal microbiota is closely related to human health, especially in infancy and childhood, where the balance of intestinal microorganisms plays an important role in neurodevelopment and disease prevention. Currently, there is a lack of effective tools to predict human intestinal diseases and develop personalized intervention measures. Summary of the invention

[0005] Purpose of the invention: In view of the shortcomings of the existing technology, the purpose of the present invention is to achieve healthy management of infant intestinal microorganisms based on digital twin technology, including prediction of intestinal microbiome development trajectory, prediction of target disease risk, intestinal flora network analysis, clinical intervention recommendations, etc. The specific objectives are as follows:

[0006] Provide a digital twin-based intestinal microbiota developmental trajectory prediction method, which can reliably predict the future developmental trajectory of the infant's intestinal microbiota using a small amount of initial observational data.

[0007] Provide a disease risk prediction method based on digital twins, which can predict the risk of target diseases such as neurodevelopmental defects based on the developmental trajectory of the intestinal microbiota of infants and young children.

[0008] A digital twin-based intestinal flora network analysis method is provided, which can calculate the causal constraint coefficient CC (causal_constraint) coefficient between microbial groups, express the influence of one group on the relative abundance of another group, and draw a network diagram.

[0009] Provide a clinical intervention method based on intestinal microorganisms, which can provide personalized intestinal flora intervention recommendations based on the predicted characteristics of intestinal microbial groups in infants and young children, so as to reduce the risk of diseases such as neurodevelopmental defects.

[0010] Technical solution: To achieve the above-mentioned invention object, the present invention adopts the following technical solution:

[0011] A method for predicting the developmental trajectory of intestinal microbiota based on digital twins, comprising the following steps:

[0012] Process the gene sequencing data of fecal samples to obtain the relative abundance data of each microbial group, and bin and quantify the continuous relative abundance data into discrete data;

[0013] The abundance of each microbial group at each time point is regarded as a feature, and a digital twin model CIT-Net model is constructed based on the conditional inference forest. The CIT-Net model contains multiple conditional inference trees (CIT, Conditional Inference Tree), each of which is used to predict the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions;

[0014] The CIT-Net model is used to predict the trajectory of microbial abundance in the future based on the initial observation data. The prediction process adopts an improved Gibbs sampling algorithm, which uses the conditional probability distribution of each feature in the CIT-Net model to sample and gradually update the feature values ​​of the sample until all missing values ​​are filled.

[0015] Preferably, the binning and quantization processing of continuous type relative abundance data into discrete type data includes:

[0016] Determine the quantification range and, for each observed microbial taxon, calculate the range of its relative abundance values;

[0017] Determine the number of discrete intervals into which the range is divided based on the size of the observed data set and the desired quantitative resolution;

[0018] A mapping function is determined, and for each relative abundance value of each observed microbial group, the mapping function is used to map it to a corresponding interval.

[0019] Preferably, the improved Gibbs sampling algorithm is used to generate new data samples or interpolate missing data in the construction of the CIT-Net model; the improved Gibbs sampling algorithm uses the conditional probability distribution inferred from the CIT-Net model to sample from these conditional distributions step by step and iteratively to generate data; the algorithm steps include:

[0020] 1) Initialization: Select a known data sample as the starting point, or start from a sample that contains some known data and some missing data;

[0021] 2) Identify missing data: Determine which data in the sample is missing, using an indicator m j To indicate whether the jth data point is missing;

[0022] 3) Iterative sampling, iteratively perform the following operations on each missing data point: a) Randomly select a missing data point x j ; b) The corresponding conditional distribution Φ(X j ∣X -j =x -j ), x -j Indicates that except x j For all data points except -j and X j Indicates the corresponding variable; c) Update the value of the data point and mark it as non-missing;

[0023] 4) Repeat step 3) and continue iterative sampling until all missing data points are interpolated or the preset number of iterations is reached;

[0024] 5) Convergence check: In the process of generating new samples, check the convergence of the samples to ensure the stability of the sampling process;

[0025] 6) Result output: output the interpolated samples or newly generated samples for further analysis or prediction.

[0026] A disease risk prediction method based on digital twins includes the following steps:

[0027] The constructed CIT-Net model was used to predict the trajectory of microbial abundance over a period of time in the future based on initial observation data of infants and young children;

[0028] The risk of infants and young children developing target diseases is predicted based on the predicted abundance trajectories of each microbial group.

[0029] Preferably, the risk Mδ of certain abnormal deviations in infants and young children in the future is quantified based on the p-distance, so as to track abnormal deviations in the maturation process of the microbiome; the p-distance between two microbial group trajectories is defined as the square root of the Jensen-Shannon divergence of the conditional relative abundance distribution caused by the CIT-Net model for a specific group at a specific time, and the average is taken for all group-timestamp pairs; the probability of a microbial group trajectory x being spontaneously replaced by another trajectory y is estimated based on the p-distance, and the smaller the p-distance, the greater the probability, indicating that the two trajectories are more similar; by comparing the probability of an observed microbial group trajectory x in a poorly performing developmental group with the probability of its occurrence in a typical developmental group, the risk Mδ of an individual's future abnormal deviation from the typical developmental trajectory is estimated.

[0030] A digital twin-based intestinal flora network analysis method includes the following steps:

[0031] The causal constraint CC coefficient is defined to measure the causal constraint effect of one microbial group on the relative abundance of another microbial group while controlling other variables;

[0032] Select the target and source taxa to determine which pairs of microbial taxa interactions need to be quantified;

[0033] Construct a conditional distribution model and use the constructed CIT-Net model to estimate the conditional distribution of the relative abundance of the target group under the given source group and other control variables;

[0034] For each source taxon change, the average change in relative abundance of the target taxon was estimated;

[0035] The marginal effect was calculated by comparing the change in the relative abundance of the target taxa when the relative abundance of the source taxa increased or decreased by one unit;

[0036] Checking time constraints to ensure that changes in the source taxa occur before changes in the target taxa;

[0037] Controlling for confounding factors: When calculating CC coefficients, control for other microbial taxa that may affect the relative abundance of the target taxa to ensure that the estimated effects are specific to the source taxa.

[0038] Conduct a statistical significance test on the estimated CC coefficient to determine whether the source group has a statistically significant effect on the target group;

[0039] Interpret the results of the CC coefficient as the positive or negative influence of one cluster on another;

[0040] The CC coefficient results were visualized to show the dynamic relationships between different microbial groups.

[0041] A microecological health management method based on digital twins, including:

[0042] The digital twin-based intestinal microbiome development trajectory prediction method;

[0043] The disease risk prediction method based on digital twins;

[0044] The intestinal flora network analysis method based on digital twins;

[0045] The intestinal microbiome-based clinical intervention method.

[0046] Based on the same inventive concept, the present invention also provides a microecological health management system based on digital twins, comprising:

[0047] The intestinal flora data feature processing module is used to process the gene sequencing data of fecal samples, obtain the relative abundance data of each microbial group, and bin and quantify the continuous relative abundance data into discrete data;

[0048] The digital twin model construction module is used to treat the abundance of each microbial group at each time point (time granularity is weeks) as a feature, and build a digital twin model CIT-Net model based on the conditional inference forest. The CIT-Net model contains multiple conditional inference trees, each of which is used to predict the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions. In the process of building the conditional inference tree, statistical tests are used to ensure the significance of node segmentation to avoid overfitting;

[0049] The intestinal microbial abundance development trajectory prediction module is used to use the CIT-Net model to predict the abundance trajectory of the microbiome in the future based on the initial observation data. The prediction process uses an improved Gibbs sampling algorithm to sample using the conditional probability distribution of each feature in the CIT-Net model, and gradually updates the feature values ​​of the sample until all missing values ​​are filled.

[0050] The disease risk prediction module is used to predict the risk of infants and young children suffering from target diseases based on the predicted abundance trajectory of each microbial group;

[0051] The microbial community network analysis module is used to draw a CC network diagram based on the CC coefficient between microbial groups and display the directional influence relationship between microbial groups.

[0052] and a clinical intervention decision support module, which is used to determine personalized intervention plans for infants and young children based on the predicted abundance trajectories of each microbial group.

[0053] A computer program product comprises a computer program / instruction, wherein the computer program / instruction implements the steps of the above-mentioned methods when executed by a processor.

[0054] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0055] 1. The present invention performs bin-based quantification processing on the relative abundance data of microbial groups, and uses the CIT-Net model composed of conditional inference trees to train a digital twin model. The features of each dimension are processed by a conditional inference tree, which has high interpretability. The structure of each tree is relatively simple and easy to understand, which can help researchers deeply understand the organizational laws and key driving factors of microbial ecosystems.

[0056] 2. The CIT-Net model constructed by the present invention has good prediction performance. By learning multivariate time-aware dependencies, it has strong prediction ability and can effectively predict the abundance dynamics of microbial groups. It also shows better prediction performance than existing methods and further predicts disease risks based on developmental trajectories, providing a basis for early intervention.

[0057] 3. The CIT-Net model constructed by the present invention uses conditional inference trees to learn complex dependencies. Each tree can learn complex dependencies between microbial groups across any time point without assuming any restrictions in advance and avoiding overfitting.

[0058] 4. The present invention can help doctors better understand the dynamic changes of intestinal microbial communities in infants and young children and provide support for clinical decision-making.

[0059] 5. The present invention can provide personalized intervention suggestions such as supplementing specific strains or adjusting diet based on the prediction results and clinical information to improve the treatment effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flow chart of a method for predicting the developmental trajectory of intestinal microbiota based on digital twins provided in an embodiment of the present invention.

[0061] Figure 2 The diagram is a conditional inference tree structure diagram exemplified in an embodiment of the present invention.

[0062] Figure 3 This is a Bacteroides trajectory prediction diagram used as an example in an embodiment of the present invention.

[0063] Figure 4 This is a flora trajectory prediction diagram exemplified in an embodiment of the present invention.

[0064] Figure 5 A flow chart of the intestinal flora network analysis method provided in an embodiment of the present invention.

[0065] Figure 6 This is a network analysis diagram exemplified in an embodiment of the present invention, where the color of the edge represents the positive and negative regulatory influence, and the boldness of the edge represents the size of the CC coefficient.

[0066] Figure 7 This is a diagram showing the prediction effect of a single sample and a single strain in an embodiment of the present invention.

[0067] Figure 8 A flow chart of a microecological health management method based on digital twins provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0069] Example 1

[0070] like Figure 1 As shown, a method for predicting the developmental trajectory of intestinal microbiota based on digital twins disclosed in an embodiment of the present invention comprises the following steps:

[0071] Step S101, processing the gene sequencing data of the stool sample to obtain the relative abundance data of each microbial group, and binning and quantizing the continuous type of relative abundance data into discrete type data.

[0072] In this step, stool samples and clinical information are collected from the target population and gene sequencing is performed. The sequencing data is processed using bioinformatics tools to obtain the relative abundance data of each microbial group, and then the binning quantification technology is used to process the relative abundance data of the microbial groups. The specific data processing steps include: determining the quantification range, for each observed microbial group, calculating the range of its relative abundance values; determining the number of discrete intervals to divide the range according to the size of the observed data set and the required quantification resolution; determining the mapping function, for each relative abundance value of each observed microbial group, using the mapping function to map it to the corresponding interval.

[0073] Exemplarily, the detailed steps of binning and quantization processing are as follows:

[0074] Step S1011, determine the quantification range, and for each observed microbial group, calculate the range of its relative abundance value (maximum and minimum values). According to the size of the observed data set and the required quantification resolution, determine how many discrete intervals the range is divided into (for example, divide the range into 26 intervals). Calculate the width of each interval and set it to Δ = 1.002*(range) / k, where k is the number of intervals.

[0075] Step S1012, determine the mapping function, select a suitable mapping function, and map each relative abundance value to the corresponding interval. The specific quantification process is as follows:

[0076] 1) Determine the number of quantization levels k: Select an appropriate number of quantization levels k, usually k = 26;

[0077] 2) Calculate bin width Δ: Calculate bin width Δ for each microbial group using the formula: Δ = 1.002 × range (βt) / k, where range (βt) is the relative abundance range of the microbial group at all observation time points, and k is the number of quantitative levels.

[0078] 3) Calculate the minimum value m: Calculate the minimum value m of each microbial group using the formula: m = min(βt)-0.001×range(βt), where min(βt) is the minimum relative abundance of the microbial group at all observation time points.

[0079] 4) Quantification: Use the mapping function ψ to quantize each relative abundance value to a discrete bin value, the formula is: ψ(bt) = Ai if and only if bt∈(m+(i-1)Δ,m+iΔ].

[0080] Step S1013, mapping relative abundance values: for each relative abundance value of each observed microbial group, use a mapping function to map it to a corresponding interval; use letters of the alphabet (AZ) as labels for the intervals.

[0081] Use quantitative data for data analysis: When building the CIT-Net model, use the quantified data as input. When making predictions, risk assessments, and intervention designs, the conditional distribution generated by the CIT-Net model can be used to convert the quantitative data back into continuous relative abundance values. The steps for dequantization are as follows: a. Determine the quantification range; b. Find the corresponding interval; c. Calculate the middle value of the interval as the dequantized value.

[0082] Step S102, the abundance of each microbial group at each time point is regarded as a feature, and a digital twin model CIT-Net model is constructed based on the conditional inference forest. The CIT-Net model contains multiple conditional inference trees, each of which is used to predict the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions.

[0083] In this step, assume that the time points in the sample data are T1, T2, ..., T N , the microbial groups are Z1, Z2, …, Z M, where N is the total number of time points and M is the total number of microbial groups; each conditional inference tree in the constructed CIT-Net model is used to predict the microbial group Z m At time point T n The relative abundance of m∈{1,2,…,M},n∈{1,2,…,N}. In this step, a CIT-Net model containing all features is constructed. Each feature corresponds to a conditional probability distribution and is modeled using a conditional inference tree to represent the relative abundance distribution of the feature given other features. The conditional inference tree is used as a component of the CIT-Net model to ensure the significance of each node segmentation through explicit statistical tests to avoid overfitting; the CIT-Net model contains multiple conditional inference trees to form a complex network structure that can accurately capture the interaction patterns between different microbial groups. Figure 2 FIG. 4 is a schematic diagram of the structure of the conditional recommendation tree in this embodiment.

[0084] Step S103, using the CIT-Net model, starting from the initial observation data, predict the trajectory of microbial abundance in the future. The prediction process uses an improved Gibbs sampling algorithm, using the conditional probability distribution of each feature in the CIT-Net model for sampling, and gradually updates the feature values ​​of the sample until all missing values ​​are filled.

[0085] In this step, the prediction process uses an improved Gibbs sampling algorithm, which uses the conditional probability distribution of each feature in the CIT-Net model for sampling, so that the sampling results are consistent with the prior knowledge of the model. Starting from a known sample, through iterative sampling, the characteristic values ​​of the sample are gradually updated until all missing values ​​are filled. Although the improved sampling algorithm is similar to Gibbs sampling in principle and both use conditional probability distribution for sampling, the improved algorithm uses a fixed conditional probability distribution and starts from a known sample, which can converge faster.

[0086] Specifically, the improved Gibbs sampling algorithm is described as follows:

[0087] Starting from a known sample, we can iteratively update its index by sampling the corresponding conditional distribution in CIT-Net. We then continue to sample the next index, this time using the value generated in the previous step. This process can be used both to generate new real samples that reflect model dependencies and to fill in missing values ​​that may exist in the data. The improved Gibbs sampling routine is:

[0088] 1) Initialization: Select a known data sample as the starting point, or start from a sample that contains some known data and some missing data;

[0089] 2) Identify missing data: Determine which data in the sample is missing, usually using an indicator m j To represent, where m j =1 means the jth data point is missing;

[0090] 3) Iterative sampling, iteratively perform the following operations on each missing data point: a) Randomly select a missing data point x j , where m j = 1; b) The corresponding conditional distribution Φ(X j ∣X -j =x -j ), x -j Indicates that except x j For all data points except -j and X j Indicates the corresponding variable; c) Update the value of the data point and mark it as non-missing;

[0091] 4) Repeat step 3) and continue iterative sampling until all missing data points are interpolated or the preset number of iterations is reached;

[0092] 5) Convergence check: In the process of generating new samples, the convergence of the samples can be checked to ensure the stability of the sampling process;

[0093] 6) Result output: output the interpolated samples or newly generated samples, which can be used for further analysis or prediction.

[0094] It also uses iterative samples from the full conditional distribution to generate samples from the joint distribution in an asymptotic manner. However, unlike Gibbs sampling, Improved Gibbs Sampling uses a fixed approximate conditional distribution inferred by CIT-Net and initialized from known samples, which can make sampling converge faster (without the warm-up period when Gibbs sampling is initialized arbitrarily).

[0095] Figure 3 and Figure 4 Shown are the Bacteroides trajectory prediction graph and the flora trajectory prediction graph obtained based on the CIT-Net model in this embodiment.

[0096] Example 2

[0097] An embodiment of the present invention discloses a disease risk prediction method based on digital twins. Based on Example 1, it predicts the trajectory of microbial abundance in the future based on a small amount of initial observation data of infants and young children; and predicts the risk of infants and young children suffering from target diseases based on the predicted abundance trajectory of each microbial group.

[0098] Specifically, the risk of deviation from the normal trajectory can be calculated based on the predicted developmental trajectory of the intestinal microbial population of infants and young children; the characteristics of the intestinal microbial population can be analyzed and associated with known disease data to predict the possibility of infants and young children suffering from intestinal diseases, etc.

[0099] In some embodiments, based on the abundance trajectory of each microbial group at each time point predicted for the infant to be evaluated, a specific model for predicting disease risk can be selected from existing machine learning algorithms, and a neural network model, a random forest classifier, etc. can be used.

[0100] For example, the p-distance between two microbial abundance trajectories can be calculated based on the CIT-Net model as a measure of the similarity between the trajectories. The Mδ risk index of the target disease is defined to measure the ratio of the probability of the current microbial abundance trajectory appearing in the neurodeficient population to the probability of appearing in the normal development population. The Mδ risk index is used to train a random forest classifier to predict whether infants and young children will have the risk of the target disease.

[0101] In this section, we use p-distance to define a risk indicator Mδ to track abnormal deviations in the microbiota maturation process. The p-distance is described as follows:

[0102] set up and Denote the CIT-Net model on distribution P and Q respectively, assuming that x = (x1, ..., xn) and y = (y1, ..., yn) are samples of X ~ P and Y ~ Q respectively. Then, the p distance between x and y is defined as:

[0103]

[0104] in represents the Jensen-Shannon distance induced by the Jensen-Shannon divergence. To simplify the notation, the subscript can be omitted and simply written as θ p Instead of θ p,p , which can be shortened to θ if the distribution is clear in the context.

[0105] Note that the Jensen-Shannon distance is a metric on a set of probability distributions (a metric in a metric space), and θ inherits non-negativity and symmetry and obeys the triangle inequality, so the p-distance is a (pseudo-)metric on Σ. Note that being a pseudo-metric means that we may have θ(x,y)=0 even if x≠y; i.e., different vectors can induce the same distribution and have zero distance. This is actually desirable, since we don't want our distances to be sensitive to biologically unimportant variations.

[0106] This definition provides a new way to measure the differences between sample vectors, which reflect possible biological changes in microbial ecosystems. The concept of p-distance is helpful for evaluating similarities and differences between samples in microbiome data analysis.

[0107] P risk: Assume that P and Q are the problem queue and the normal queue, respectively, and train CIT-Net models ΦP and ΦQ respectively. Let x=(x1,...,xn) represent a sample realization of the random variable X=(X1,...,Xn). Let V P represents the vector where all features sampled from P are missing; V Q represents a vector with all features missing from Q. Then, we define the P risk of x as:

[0108]

[0109] By constructing CIT-net models for two subcohorts corresponding to typical and suboptimal development, respectively, the clinically meaningful similarity between two stable microbial taxon trajectories x, y is quantified as the probability that an observed trajectory x is randomly replaced by another y. Therefore, in our approach, the higher the probability Pr(x→y), the more similar they are. We estimate Pr(x→y) by formulating the intrinsic distance between microbial taxon trajectories. The p-distance θ(x,y) between two microbial taxon trajectories is defined as the square root of the Jensen-Shannon divergence of the conditional relative abundance distribution induced by CIT-Net for a particular taxon at a particular time, averaged over all taxon-timestamp pairs. The p-distance approximates the log-likelihood of spontaneous changes in microbial taxon trajectories, that is, θ(x,y) ≈ constant × logPr(x→y). (Qualitatively, if two trajectories lead to healthy outcomes, i.e., are interchangeable, then we want to be sure that they are similar, while if a trajectory ultimately leads to an adverse outcome, then our goal is to be able to identify clinically meaningful differences.)

[0110] The significance of p-distance: 1. Similarity measurement, p-distance can measure the similarity between two microbial community trajectories. The smaller the distance, the more similar the two trajectories are; the larger the distance, the less similar the two trajectories are; 2. Probability estimation, based on p-distance, the probability of a microbial community trajectory x being spontaneously replaced by another trajectory y can be estimated. The smaller the p-distance, the greater the probability, indicating that the two trajectories are more similar; 3. Risk measurement, by comparing the probability of an observed microbial community trajectory x in a poorly performing developmental group with the probability of its occurrence in a typical developmental group, the risk Mδ of an individual's future abnormal deviation from the typical developmental trajectory can be estimated. The higher the Mδ, the greater the risk of abnormal development of the individual.

[0111] Example 3

[0112] An intestinal flora network analysis method based on digital twins disclosed in an embodiment of the present invention calculates the causal constraint coefficient CC (CausalConstraint) between microbial groups on the basis of the CIT-Net model constructed in Example 1, which represents the influence of one group on the relative abundance of another group, and draws a network diagram for network analysis.

[0113] CC network analysis can explore the bidirectional influence between different microbial groups, that is, the impact of a certain group on the abundance change of another group, calculate the CC coefficient between microbial groups, draw a CC network diagram, and show the directional influence relationship between microbial groups.

[0114] The CC coefficient focuses on the marginal effect of the relative abundance change of the source microbial group on the target microbial group. It is a coefficient that represents the marginal contribution of the relative abundance change of Xj to the relative abundance change of Xi. The calculation formula is: CC coefficient = marginal contribution / change in relative abundance of Xi. The calculation process is as follows Figure 5 As shown, the steps are as follows:

[0115] Step S301, define the CC coefficient, which is used to measure the causal constraint effect of one microbial group (source group) on the relative abundance of another microbial group (target group) while controlling other variables. This effect is causal and irreversible in time (future changes will not affect the past);

[0116] Step S302, select target and source groups, determine which pair of microbial groups' interactions need to be quantified. Select a target group, whose relative abundance changes will be studied, and a source group, which is considered to have an impact on the target group;

[0117] Step S303, constructing a conditional distribution model, using the CIT-net model to estimate the conditional distribution of the relative abundance of the target group under the conditions of the given source group and other control variables;

[0118] Step S304, analyzing the quantitative impact, estimating the average change in the relative abundance of the target group for each small change in the source group. This usually involves calculating the expected value or mean of the conditional distribution;

[0119] Step S305, calculating the marginal effect, by comparing the change in the relative abundance of the target group when the relative abundance of the source group increases or decreases by one unit, to calculate the marginal effect. This can be done by taking the derivative of the conditional distribution or using an approximate method;

[0120] Step S306, check the time constraint. Since the CC coefficient is time-local, it is necessary to ensure that the correct time sequence is considered during the analysis, that is, the change of the source group occurs before the change of the target group.

[0121] Step S307, controlling confounding factors. When calculating the CC coefficient, it is necessary to control other microbial groups that may affect the relative abundance of the target group to ensure that the estimated effect is specific to the source group;

[0122] Step S308, significance test, performing a statistical significance test on the estimated CC coefficient to determine whether the influence of the source group on the target group is statistically significant;

[0123] Step S309, result interpretation, interpreting the CC coefficient result as the positive or negative influence of one group on another group. A positive coefficient may indicate a promoting effect, while a negative coefficient may indicate an inhibiting effect;

[0124] Step S310, visualization and application, visualize the results of the CC coefficient to show the dynamic relationship between different microbial groups and apply it to further ecological analysis or clinical decision-making.

[0125] In this example, network analysis using the CC network diagram can reveal the bidirectional (positive and negative) effects of abundance fluctuations between different populations.

[0126] To better account for the inherent asymmetric directional influences between taxa in ecosystems, we calculated a causal CC coefficient for each pair of observed interacting microbial taxa. The CC coefficient represents the up-regulation / down-regulation influence of a source microbial taxon on a target microbial taxon, where the regulatory effects are causally constrained in time (the future cannot influence the past). Using this concept, we can examine the “influence network” revealed by CIT-Net, allowing us to visualize the main differences between the digital twins inferred for the optimal and suboptimal cohorts. After calculating the set of CC values ​​for each model, a directed graph of asymmetric relationships can be plotted. As Figure 6 The CC network analysis diagram of the example is shown, where the color of the edge represents the positive and negative regulatory influence, and the boldness of the edge represents the size of the CC coefficient.

[0127] Example 4

[0128] The embodiment of the present invention discloses a clinical intervention method based on intestinal microorganisms. Based on the first embodiment, based on a small amount of initial observation data of infants and young children, the microbial abundance trajectory in the future is predicted; according to the predicted abundance trajectory of each microbial group, the personalized intervention plan for infants and young children is determined. For example, early supplementation of Bacteroidetes or Actinomycetes. Considering the impact of dietary factors on Mδ risk, other possible intervention plans are explored; the ranking of the best intervention plans can be provided.

[0129] Example 5

[0130] The embodiment of the present invention discloses a microecological health management method based on digital twins, which integrates the intestinal microbiome development trajectory prediction, disease risk prediction, intestinal flora network analysis and other methods in embodiments 1 to 4. Figure 8 As shown, the main steps include:

[0131] Step S501, data collection: collecting stool samples from the target population and performing gene sequencing.

[0132] Step S502, data processing: using bioinformatics tools to process the sequencing data, obtain the relative abundance data of each microorganism, and bin and quantify the continuous type of relative abundance data into discrete type data.

[0133] Step S503, model construction: taking the abundance of each microbial group of each infant at each time point as a feature, constructing a digital twin model CIT-Net model based on the conditional inference forest, each tree predicts the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions, and deeply recursively learns the interactions between microbial groups.

[0134] Step S504, prediction of microbial abundance trajectory: Using the CIT-Net model, starting from a small amount of initial observation data, predict the trajectory of microbial abundance in the future. The digital twin model can predict the microbial abundance at any point in the future.

[0135] Step S505, target disease risk prediction: combining the predicted microbial group trajectory data and clinical data to assess the risk of intestinal diseases in infants and young children.

[0136] Step S506, intestinal microbial flora network analysis: calculate the CC coefficient between microbial groups, and draw a CC network diagram to show the directional influence relationship between microbial groups.

[0137] Step S507, personalized intervention decision support: formulate a personalized intervention plan for each infant.

[0138] Example 6

[0139] The embodiment of the present invention discloses a microecological health management system based on digital twins, which mainly includes intestinal flora data feature processing, digital twin model construction, intestinal microbial abundance development trajectory prediction, disease risk prediction, microbial flora network analysis module and clinical intervention decision support module.

[0140] The intestinal flora data feature processing module processes the gene sequencing data of fecal samples to obtain the relative abundance data of each microbial group, and bins and quantifies the continuous relative abundance data into discrete data for more efficient analysis and modeling.

[0141] The digital twin model construction module considers the abundance of each microbial group at each time point as a feature, and constructs the digital twin model CIT-Net model based on the conditional inference forest. The CIT-Net model contains multiple conditional inference trees, each of which is used to predict the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions. The system can use all the features in multiple dimensions to build a highly simulated digital twin, solve the problem of scarcity of problem data through transfer learning, and truly realize the two functions of "reflecting the real with the virtual" and "controlling the real with the virtual".

[0142] The intestinal microbial abundance development trajectory prediction module uses the CIT-Net model to predict the microbial abundance trajectory in the future based on the initial observation data, and can fill in missing data and generate new samples through sampling.

[0143] The disease risk prediction module assesses the risk of infants and young children developing target diseases based on the predicted abundance trajectories of each microbial community and clinical data.

[0144] The microbial flora network analysis module calculates the CC coefficient between microbial groups, draws CC network diagrams, and displays the directional influence relationship between microbial groups.

[0145] The clinical intervention decision support module can specify personalized intervention plans for each infant based on the predicted abundance trajectories of each microbial group.

[0146] Example 7

[0147] A computer program product disclosed in an embodiment of the present invention includes a computer program / instruction, wherein the computer program / instruction implements the steps of each method in Embodiments 1 to 5 when executed by a processor.

Claims

1. A method for predicting the developmental trajectory of intestinal microbiota based on digital twins, characterized in that: The steps include: Process the gene sequencing data of fecal samples to obtain the relative abundance data of each microbial group, and bin and quantify the continuous relative abundance data into discrete data; The abundance of each microbial group at each time point is regarded as a feature, and a digital twin model CIT-Net model is constructed based on the conditional inference forest. The CIT-Net model contains multiple conditional inference trees, each of which is used to predict the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions; Using the CIT-Net model, starting from the initial observation data, the trajectory of microbial abundance in the future is predicted; the prediction process uses an improved Gibbs sampling algorithm, using the conditional probability distribution of each feature in the CIT-Net model for sampling, and gradually updates the feature values ​​of the sample until all missing values ​​are filled; The improved Gibbs sampling algorithm is used to generate new data samples or interpolate missing data in the construction of the CIT-Net model. The improved Gibbs sampling algorithm uses the conditional probability distribution inferred from the CIT-Net model to generate data by sampling from these conditional distributions step by step and iteratively. The algorithm steps include: 1) initialization: selecting a known data sample as a starting point, or starting from a sample containing some known data and some missing data; 2) identifying missing data: determining which data in the sample is missing, using an indicator m j To indicate whether the jth data point is missing; 3) Iterative sampling, iteratively perform the following operations on each missing data point: a) Randomly select a missing data point x j ; b) The corresponding conditional distribution Φ(X j ∣X -j =x -j ), x -j Indicates that except x j For all data points except -j and X j Indicates the corresponding variable; c) Update the value of the data point and mark it as non-missing; 4) Repeat step 3) and continue iterative sampling until all missing data points are interpolated or the preset number of iterations is reached; 5) Convergence check: in the process of generating new samples, check the convergence of the samples to ensure the stability of the sampling process; 6) Output the results: output the interpolated samples or the newly generated samples for further analysis or prediction.

2. The method for predicting the developmental trajectory of intestinal microbiota based on digital twins according to claim 1, characterized in that: The step of binning and quantifying the continuous type of relative abundance data into discrete type of data includes: Determine the quantification range and, for each observed microbial taxon, calculate the range of its relative abundance values; Determine the number of discrete intervals into which the range is divided based on the size of the observed data set and the desired quantitative resolution; A mapping function is determined, and for each relative abundance value of each observed microbial group, the mapping function is used to map it to a corresponding interval.

3. A disease risk prediction method based on digital twins, characterized in that: The steps include: Using the CIT-Net model constructed according to the method of claim 1 or 2, based on the initial observation data of infants and young children, predicting the trajectory of microbial abundance in the future; The risk of infants and young children developing target diseases is predicted based on the predicted abundance trajectories of each microbial group.

4. The disease risk prediction method based on digital twins according to claim 3 is characterized in that: Based on the p-distance, the risk Mδ of certain abnormal deviations in infants and young children in the future is quantified to track abnormal deviations in the maturation of the microbiome; the p-distance between two microbial group trajectories is defined as the square root of the Jensen-Shannon divergence of the conditional relative abundance distribution caused by the CIT-Net model for a specific group at a specific time, and the average is taken for all group-timestamp pairs; the probability of a microbial group trajectory x being spontaneously replaced by another trajectory y is estimated based on the p-distance. The smaller the p-distance, the greater the probability, indicating that the two trajectories are more similar; by comparing the probability of an observed microbial group trajectory x in a poorly performing developmental group with the probability of its occurrence in a typical developmental group, the risk Mδ of an individual's future abnormal deviation from the typical developmental trajectory is estimated.

5. A clinical intervention method based on intestinal microorganisms, characterized in that: The steps include: Using the CIT-Net model constructed according to the method of claim 1 or 2, based on the initial observation data of infants and young children, predicting the trajectory of microbial abundance in the future; Determine personalized intervention plans for infants and young children based on the predicted abundance trajectories of each microbial community.

6. A method for intestinal flora network analysis based on digital twins, characterized in that: The steps include: The causal constraint CC coefficient is defined to measure the causal effect of one microbial group on the relative abundance of another microbial group while controlling other variables; Select the target and source taxa to determine which pairs of microbial taxa interactions need to be quantified; Constructing a conditional distribution model, using the CIT-Net model constructed in the method according to any one of claims 1 to 3 to estimate the conditional distribution of the relative abundance of the target group under the conditions of the given source group and other control variables; For each source taxon change, the average change in relative abundance of the target taxon was estimated; The marginal effect was calculated by comparing the change in the relative abundance of the target taxa when the relative abundance of the source taxa increased or decreased by one unit; Checking time constraints to ensure that changes in the source taxa occur before changes in the target taxa; Controlling for confounding factors: When calculating CC coefficients, control for other microbial taxa that may affect the relative abundance of the target taxa to ensure that the estimated effects are specific to the source taxa. Conduct a statistical significance test on the estimated CC coefficient to determine whether the source group has a statistically significant effect on the target group; Interpret the results of the CC coefficient as the positive or negative influence of one cluster on another; The CC coefficient results were visualized to show the dynamic relationships between different microbial groups.

7. A microecological health management method based on digital twins, characterized in that: include: The method for predicting the developmental trajectory of intestinal microbiota based on digital twins according to claim 1 or 2; The disease risk prediction method based on digital twins according to claim 3 or 4; The intestinal flora network analysis method based on digital twin according to claim 6; The intestinal microbiome-based clinical intervention method according to claim 5.

8. A microecological health management system based on digital twins, characterized in that: include: The intestinal flora data feature processing module is used to process the gene sequencing data of fecal samples, obtain the relative abundance data of each microbial group, and bin and quantify the continuous relative abundance data into discrete data; A digital twin model construction module is used to regard the abundance of each microbial group at each time point as a feature, and to construct a digital twin model CIT-Net model based on a conditional inference forest. The CIT-Net model contains multiple conditional inference trees, each of which is used to predict the relative abundance of a specific microbial group at a specific time point, and considers the relative abundance of other microbial groups as conditions; The intestinal microbial abundance development trajectory prediction module is used to use the CIT-Net model to predict the microbial abundance trajectory in the future based on the initial observation data; the prediction process adopts an improved Gibbs sampling algorithm, uses the conditional probability distribution of each feature in the CIT-Net model for sampling, and gradually updates the characteristic values ​​of the sample until all missing values ​​are filled; the improved Gibbs sampling algorithm is used to generate new data samples or interpolate missing data separately in the construction of the CIT-Net model; the improved Gibbs sampling algorithm uses the conditional probability distribution inferred from the CIT-Net model to sample from these conditional distributions step by step and iteratively to generate data; the algorithm steps include: 1) initialization: select a known data sample as the starting point, or start from a sample containing some known data and some missing data; 2) identify missing data: determine which data in the sample is missing, and use an indicator m j To indicate whether the jth data point is missing; 3) Iterative sampling, iteratively perform the following operations on each missing data point: a) Randomly select a missing data point x j ; b) The corresponding conditional distribution Φ(X j ∣X -j =x -j ), x -j Indicates that except x j For all data points except -j and X j Indicates the corresponding variable; c) Update the value of the data point and mark it as non-missing; 4) Repeat step 3) and continue iterative sampling until all missing data points are interpolated or the preset number of iterations is reached; 5) Convergence check: In the process of generating new samples, check the convergence of the samples to ensure the stability of the sampling process; 6) Output the results: output the interpolated samples or the newly generated samples for further analysis or prediction; The disease risk prediction module is used to predict the risk of infants and young children suffering from target diseases based on the predicted abundance trajectory of each microbial group; The microbial community network analysis module is used to draw a CC network diagram based on the CC coefficient between microbial groups and display the directional influence relationship between microbial groups; and a clinical intervention decision support module, which is used to determine personalized intervention plans for infants and young children based on the predicted abundance trajectories of each microbial group.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images