Disease prediction method and system based on self-adaptive fusion of layered features of microbiological components

Abstract

The invention discloses a disease prediction method and system based on microbiome hierarchical feature self-adaptive fusion, relates to the technical field of microbiome data analysis, and uses a trained disease prediction model to perform disease prediction based on microbiome data. The disease prediction model performs multi-level feature extraction on the feature data of the five classification levels through a hierarchical feature extraction module, and integrates results into multi-level fusion features; a multi-scale feature fusion module is used for processing multi-level fusion features through three parallel paths of a mode stream, a context stream and a content stream, and all-directional characterization of microbiome data is achieved; a dynamic sampling module is introduced in the training process, and the sampling weight of a sample is dynamically adjusted based on a category balance mechanism and an uncertainty evaluation strategy; and finally, performing disease prediction on the multi-scale fusion features through a classifier. According to the method, multi-dimensional feature representation is realized by fully utilizing microorganism classification hierarchy information, the problem of sample imbalance is effectively solved, and the disease prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of microbiome data analysis, and in particular to a disease prediction method and system based on adaptive fusion of hierarchical features of microbiome. Background Art

[0002] With the rapid development of high-throughput sequencing technology, microbiome research has made significant progress in recent years. As a micro-ecosystem inside the human body, the microbial community is closely related to human health, especially the "gut-brain axis" relationship between the intestinal microbiome and the nervous system has been confirmed by many studies. Microbiome data analysis has shown great potential and value in the field of diagnosis of neurological diseases, providing new perspectives and methods for early diagnosis and prognosis assessment.

[0003] However, unlike traditional clinical data, microbiome data has significant particularity, which brings severe challenges to data analysis: first, microbiome data is characterized by high dimensionality, high noise and high sparsity, often containing thousands or even tens of thousands of features, while the sample size is usually small, which can easily lead to model overfitting; second, microbiome data presents obvious hierarchical structural characteristics. The classification hierarchy from phylum, class, order, family to genus reflects the evolutionary relationship and functional characteristics of microorganisms. This structural information is of great value for disease diagnosis, but traditional analysis methods are often unable to effectively utilize it; in addition, microbiome data also has multidimensional characterization requirements. The characteristics of a microbial community can be interpreted from multiple perspectives such as individual characteristics, community structure patterns and environmental dependence. Single-dimensional feature extraction is difficult to fully capture the complex characteristics of the microbiome; finally, in the multi-disease classification scenario, there is usually an imbalance in the number of samples of different diseases. How to maintain the stability of the model while taking into account the diagnostic accuracy of various diseases is another major challenge facing microbiome data analysis.

[0004] Existing disease prediction methods for microbiome data mainly include: classification methods based on machine learning, such as random forests, support vector machines, gradient boosting decision trees, etc.; and emerging deep learning methods, such as multi-layer perceptrons, convolutional neural networks, recurrent neural networks, etc. However, these methods generally have the following limitations: First, they cannot fully utilize the hierarchical knowledge of microbial taxonomy and ignore the information complementarity between different classification levels; second, the feature extraction ability is limited, and it is difficult to simultaneously capture the structured patterns, environmental dependence characteristics and individual characteristics of microbiome data; third, the sample imbalance problem is not considered enough, and the performance is unstable in multi-disease classification scenarios.

[0005] Therefore, there is an urgent need to develop a prediction method that can fully utilize the information of microbial classification levels, realize multi-dimensional feature characterization, and effectively deal with the problem of sample imbalance, so as to improve the diagnostic accuracy and stability of neurological diseases. Summary of the invention

[0006] The purpose of the present invention is to solve the problems in the prior art.

[0007] The technical solution adopted by the present invention to solve the technical problem is: to provide a disease prediction method based on adaptive fusion of hierarchical features of microbial groups, comprising the following steps:

[0008] Preprocess the multi-source microbiome data samples to obtain hierarchical feature data samples at five levels: phylum, class, order, family, and genus;

[0009] Build a disease prediction model and use stratified feature data samples for training. During the training process, combine the category balance mechanism and multi-dimensional uncertainty assessment strategy to dynamically adjust the sample sampling weights.

[0010] Use trained disease prediction models to predict diseases based on microbiome data;

[0011] In the disease prediction model, the hierarchical feature extraction module performs multi-level feature extraction on the hierarchical feature data of five classification levels, and integrates them to obtain multi-level fusion features containing hierarchical information; the multi-scale feature fusion module processes the multi-level fusion features through three parallel pathways, and outputs features containing structured information, environmental dependency information and individual feature information respectively, and then integrates the features output by the three pathways through a dynamic weighted fusion mechanism to obtain multi-dimensional fusion features; the classifier predicts diseases based on the multi-scale fusion features.

[0012] Preferably, the process of obtaining the multi-level fusion features by the hierarchical feature extraction module comprises the following steps:

[0013] The hierarchical feature extraction component extracts features from five levels and combines them to obtain a hierarchical feature representation. , expressed as:

[0014] ;

[0015] ;

[0016] ;

[0017] in, , , , , Represents the characteristics of the five levels: phylum, class, order, family, and genus; the characteristics of each level All taxa at this level The abundance of The vector composed of , represents the data matrix input into the disease prediction model, Indicates the corresponding taxon The eigenvector of It is a taxon The set of all descendant taxa composed of , by recursively using the child node mapping Calculated, for A taxon element in records the set of all descendants; Recorded by taxonomic unit is the set of all sub-classification units of the parent node, through the mapping It is constructed by reverse mapping the parent-child relationship information in; For known data, each microbial taxonomic unit is recorded The five levels and parent node information;

[0018] The classification level weight component dynamically adjusts the feature weights to obtain weighted feature representations , expressed as:

[0019] ;

[0020] ;

[0021] Where X represents the sample data matrix; It is a taxon The base weight of the level, Measures the discriminative power of features. Indicates the sparsity of the feature. and is the equilibrium parameter, represents a diagonal matrix with the adjusted weight vector as the diagonal element;

[0022] The classification relationship modeling component constructs a multidimensional relationship matrix based on the classification path and evolutionary distance between features, and calculates the relationship feature representation through the multidimensional relationship matrix , expressed as:

[0023] ;

[0024] ;

[0025] ;

[0026] in, Representation characteristics and The common path length, and Represent the arrival characteristics and The length of the complete classification path; Representation characteristics and The evolutionary distance between Parameters to control the intensity of distance effects;

[0027] Obtain multi-level fusion features through feature fusion , expressed as:

[0028] ;

[0029] is a learnable fusion weight matrix that is optimized during training by gradient descent.

[0030] Preferably, the multi-scale feature fusion module uses multi-level fusion features as input features , processing multi-level fusion features through three parallel paths, and finally calculating the multi-dimensional fusion features, including the following steps:

[0031] The pattern flow path uses the correlation attention mechanism to process the input features and obtain the pattern flow output features. , expressed as:

[0032] ;

[0033] ;

[0034] ;

[0035] in, , , are the learnable weight matrices for queries, keys, and values, respectively, , , denote the learned query, key, and value projections, respectively. is a scaling factor used to ensure the stability of the gradient flow during training;

[0036] The context flow path uses a global scope processing module to establish long-range dependencies by calculating the similarity matrix between input features, and calculates the upstream and downstream output features through the feedforward neural network of the context flow. , expressed as:

[0037] ;

[0038] ;

[0039] in, represents the global attention operation, and Represent two fully connected layer operations respectively, is the activation function, To prevent overfitting operations;

[0040] The content flow channel processes the input features through the similarity mapping mechanism and the channel focusing module to obtain the content flow output features. , expressed as:

[0041] ;

[0042] ;

[0043] in, and is a learnable weight matrix, is the activation function, is the attention mapping matrix, Indicates channel focus operation;

[0044] Calculate the importance weight of each channel, and then perform weighted combination of the output features of each channel based on the weight to obtain multi-dimensional fusion features , expressed as:

[0045] ;

[0046] ;

[0047] in, represents the learnable fusion weights corresponding to the three pathways, represents the learnable weight vector corresponding to the three pathways, represents the global pooling operation, Represents the output features corresponding to the three pathways; Indicates the mode flow path, Indicates the up and down flow path, Indicates the content flow path.

[0048] Preferably, disease prediction is performed on multi-dimensional fusion features through a classifier, specifically: disease prediction is performed on multi-scale fusion features through a random forest classifier, and classification results and confidence levels are output.

[0049] Preferably, the classification mode can be configured as binary classification or multi-classification according to actual prediction requirements;

[0050] In the binary classification setting, the classification results are distinguished into disease status and non-disease status;

[0051] In a multi-classification setting, the classification results can identify and differentiate into multiple different disease types.

[0052] Preferably, the dynamic adjustment of sample sampling weights is implemented by a dynamic sampling module, comprising the following steps:

[0053] The basic weight is calculated based on the category balancing mechanism and is expressed as:

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] in, Representation sample The base weight of Representation sample Category The comprehensive weight of is the disease level weight, which represents the inverse frequency weight of the category; is the inter-disease weight, which represents the normalized weight of the relative sample distribution between categories; Is the disease category The number of samples in is the total number of samples, is the total number of disease categories;

[0059] The composite uncertainty score is calculated based on the multi-dimensional uncertainty assessment mechanism and is expressed as:

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] in, Representation sample The composite uncertainty score of , and The samples are The normalized entropy value of , confidence difference and inter-class differences , , and All represent weight coefficients; Represents input Belongs to category The predicted probability of Indicates category The unnormalized logic output of Representation sample The probability of being correctly classified; Indicates the first predicted probability;

[0066] The comprehensive weight is calculated by combining the basic weight and the composite uncertainty score, which is expressed as:

[0067] ;

[0068] ;

[0069] in, Indicates the current iteration number, represents the maximum number of iterations, It is a parameter that controls the ratio of uncertainty weights, which gradually increases as training progresses but does not exceed the preset maximum value. .

[0070] Preferably, the dynamic sampling module also includes a temperature parameter adjustment mechanism, which is expressed as:

[0071] ;

[0072] ;

[0073] in, and are the minimum and maximum values ​​of the temperature parameter, It is the rate at which the temperature is reduced. A higher temperature is used in the early stage of training to make the sampling more uniform, and a lower temperature is used in the later stage of training to make the sampling more focused on high-weight samples. Representation sample In the Final sampling weight calculation in iterations.

[0074] Preferably, the method of using the trained disease prediction model to predict diseases based on microbiome data comprises the following steps:

[0075] Collect and preprocess the microbiome data to be tested to obtain hierarchical feature data at five levels: phylum, class, order, family, and genus;

[0076] The hierarchical feature data is input into the trained disease prediction model to obtain the classification results and confidence.

[0077] The present invention also provides a disease prediction system based on adaptive fusion of hierarchical features of microbiome, comprising:

[0078] The preprocessing module preprocesses the multi-source microbiome data samples to obtain hierarchical feature data samples at five levels: phylum, class, order, family, and genus;

[0079] The training module builds a disease prediction model and uses stratified feature data samples for training. During the training process, the category balance mechanism and multi-dimensional uncertainty assessment strategy are combined to dynamically adjust the sample sampling weights.

[0080] The prediction module uses the trained disease prediction model to predict diseases based on microbiome data;

[0081] In the disease prediction model, the hierarchical feature extraction module performs multi-level feature extraction on the hierarchical feature data of five classification levels, and integrates them to obtain multi-level fusion features containing hierarchical information; the multi-scale feature fusion module processes the multi-level fusion features through three parallel pathways, and outputs features containing structured information, environmental dependency information and individual feature information respectively, and then integrates the features output by the three pathways through a dynamic weighted fusion mechanism to obtain multi-dimensional fusion features; the classifier predicts diseases based on the multi-scale fusion features.

[0082] The present invention has the following beneficial effects:

[0083] (1) The present invention explicitly models the hierarchical relationship of microbial taxonomy through a hierarchical feature extraction module, fully utilizes the rich biological information contained in the microbial classification hierarchy, effectively captures the evolutionary and functional connections between phyla, classes, orders, families, and genera, and improves the biological significance and interpretability of feature representation;

[0084] (2) The multi-scale feature fusion module designed in this invention uses three parallel feature processing streams to deeply interpret the microbiome data from three dimensions: structured patterns, environmental dependence, and individual characteristics. This achieves a comprehensive characterization of the microbiome data and significantly enhances the model's ability to identify complex microbiome patterns.

[0085] (3) The dynamic sampling module of the present invention combines the category balance mechanism and the multi-dimensional uncertainty assessment strategy to effectively solve the sample imbalance problem in the multi-disease classification scenario, significantly improve the model's learning ability for minority diseases and difficult samples, and enhance the stability and accuracy of classification;

[0086] (4) The present invention adopts an end-to-end training method, and each module is optimized collaboratively, which reduces the intermediate processing steps, improves the overall performance and generalization ability of the model, and provides more robust technical support for the diagnosis of neurological diseases based on the microbiome.

[0087] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments, but the present invention is not limited to the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 A method step diagram of an embodiment of the present invention;

[0089] Figure 2 It is a schematic diagram of the process of the hierarchical feature extraction module according to an embodiment of the present invention;

[0090] Figure 3 Schematic diagram of the process of a multi-scale feature fusion module according to an embodiment of the present invention;

[0091] Figure 4 This is a schematic diagram of the flow of a dynamic sampling module according to an embodiment of the present invention;

[0092] Figure 5 This is a performance comparison chart of various nervous system disease diagnosis tasks according to an embodiment of the present invention;

[0093] Figure 6 4 is a system structure diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0094] The data used in the embodiments of the present invention include integrated gut microbiome data from 31 independent studies, covering three neurological diseases: Alzheimer's disease (AD, 6 data sets, 284 cases in the disease group, 264 cases in the control group), Parkinson's disease (PD, 11 data sets, 1267 cases in the disease group, 942 cases in the control group), and autism spectrum disorder (ASD, 14 data sets, 708 cases in the disease group, 584 cases in the control group).

[0095] like Figure 1 As shown, it is a method step diagram of the present invention, comprising the following steps:

[0096] S101, preprocessing the multi-source microbiome data samples to obtain hierarchical feature data samples including five levels: phylum, class, order, family, and genus;

[0097] S102, construct a disease prediction model, use stratified feature data samples for training, combine the category balance mechanism and multi-dimensional uncertainty assessment strategy during the training process, and dynamically adjust the sample sampling weight;

[0098] S103, using the trained disease prediction model to predict diseases based on microbiome data.

[0099] Specifically, the disease prediction model uses a hierarchical feature extraction module to perform multi-level feature extraction on feature data of five classification levels, and integrates multi-level fusion features; uses a multi-scale feature fusion module to process multi-level fusion features through three parallel pathways of pattern flow, context flow and content flow, and then integrates the features output by the three pathways through a dynamic weighted fusion mechanism to achieve a comprehensive characterization of microbiome data; finally, a classifier is used to predict diseases based on the multi-scale fusion features.

[0100] Specifically, the preprocessing process involves several key steps: first, the quality of the original sequencing data is assessed using the FastQC tool, and the initial 10 nucleotides of each read are removed to reduce systematic bias; ASV inference is performed using the DADA2 process, which can distinguish between true biological variations and sequencing errors at a single-base resolution level; the SILVA 138.1 database is used for taxonomic annotation, and a minimum bootstrap confidence threshold of 80% is set; ASV sequences of chloroplasts or mitochondria and ASVs with a frequency of less than 1% are excluded; finally, the taxonomic levels of all data sets are uniformly aggregated to the genus level, and standardized through sparse sampling to eliminate the impact of differences in sequencing depth between different samples.

[0101] Specifically, the process of predicting by the disease prediction model includes the following steps:

[0102] S201, preprocessing the multi-source microbiome data to be detected to obtain hierarchical feature data including five levels: phylum, class, order, family, and genus;

[0103] S202, using a hierarchical feature extraction module to perform multi-layer feature extraction on the hierarchical feature data, and fusing to obtain multi-level fusion features (i.e., multi-level feature fusion representation);

[0104] S203, using a multi-scale feature fusion module to process multi-level feature fusion representation through three parallel paths of pattern stream, context stream and content stream, and then integrating the features output by the three paths through a dynamic weighted fusion mechanism to obtain a multi-scale fusion feature (i.e., a multi-dimensional feature representation);

[0105] S204, finally, the classifier is used to perform disease prediction on the multi-dimensional feature representation.

[0106] Specifically, the S202 is as follows: Figure 2 As shown in Figure 1, the hierarchical feature extraction module adopts a bottom-up feature processing strategy, starting from the attribute level and gradually building up to the gate-level feature representation, and finally obtains. This module contains three key feature processing components:

[0107] (1) Hierarchical feature extraction component: through mapping function and subset functions Construct a complete taxonomic relationship network. For each taxonomic unit , Returns its full taxonomic path (e.g., M("Lactobacillus") = {"Bacteria", "Firmicutes", "Bacilli", "Lactobacillales", "Lactobacillaceae", "Lactobacillus"}). Returns all its sub-collections (e.g. R("Lactobacillaceae") = {"Lactobacillus", "Pediococcus", ...}). Hierarchical feature representation It is composed of the features of each classification level:

[0108] ;

[0109] Among them, the characteristics of each level is a vector consisting of the aggregate abundance of all taxa at this level:

[0110] ;

[0111] Each taxon The abundance of Calculated as the sum of the abundances of all its descendant taxa:

[0112] ;

[0113] in, It is a taxon The set of all descendants (including child nodes, grandchild nodes, etc.) of Perform calculations.

[0114] (2) Classification level weight component: An adaptive weighted feature scheme is introduced, and the basic weight is designed to increase from phylum to genus level: 0.6 for phylum level, 0.7 for class level, 0.8 for order level, 0.9 for family level, and 1.0 for genus level. At the same time, a dynamic weight adjustment mechanism is designed to adjust the weight according to the sparsity and discriminability of the features:

[0115] ;

[0116] in, It is a taxon The base weight of the level, Measures the discriminative power of a feature, determined by calculating the coefficient of variation of the feature between different disease categories. Indicates the sparsity of the feature, which is determined by calculating the proportion of zero values ​​of the feature. and are balance parameters and are set to 0.3 and 0.2 respectively.

[0117] Feature representation generated by the classification level weight component Calculated as:

[0118] ;

[0119] in is the original feature matrix, represents a diagonal matrix with the adjusted weight vector as its diagonal elements.

[0120] (3) Taxonomic relationship modeling component: Construct a multidimensional relationship matrix to quantify the taxonomic distance between features by analyzing the taxonomic paths between feature pairs and determining the most recent common ancestor. The relationship strength between features is calculated using the following formula:

[0121] ;

[0122] in, Representation characteristics and The common path length, and Represent the arrival characteristics and The complete classification path length.

[0123] To further improve the accuracy of relationship modeling, the module also introduces an enhanced relationship metric that takes evolutionary distance into account:

[0124] ;

[0125] in, Representation characteristics and The evolutionary distance between Set this to 0.5 to control the strength of the distance effect.

[0126] Feature representation generated by the classification relationship modeling component Calculated as:

[0127] ;

[0128] The final feature representation This is achieved through a multi-level feature fusion strategy, expressed as:

[0129] ;

[0130] in, is the original feature weighted by the classification level weight, It is a hierarchical aggregation feature generated by the hierarchical feature extraction algorithm. It is the relational feature generated based on the classification relation matrix. is a learnable fusion weight matrix that is optimized during training by gradient descent.

[0131] Specifically, the S203 is as follows: Figure 3 As shown in the figure, the multi-scale feature fusion module interprets the microbiome data in multiple dimensions through three parallel feature processing streams:

[0132] (1) Pattern Stream: Responsible for identifying and extracting structured patterns in microbial communities. This stream first passes through , , Transformation projects the input features, and then uses the correlation attention mechanism to process the transformed features and receive the input features , the core attention calculation formula is:

[0133] ;

[0134] ;

[0135] ;

[0136] in , , are the learnable weight matrices for queries, keys, and values, respectively, , , denote the learned query, key and value projections, respectively, and the scaling factor Ensure the stability of gradient flow during training.

[0137] (2) Context Stream: Focuses on the environmental dependency characteristics of microbial communities. The Global Range Processing module is used to establish long-range dependencies by calculating the similarity matrix between features. The feedforward neural network calculation expression of the context stream is:

[0138] ;

[0139] ;

[0140] in represents the global attention operation, and There are two fully connected layers. is the activation function, It is an operation to prevent overfitting.

[0141] (3) Content Stream: focuses on processing the individual feature information of microorganisms. Features are processed through the similarity mapping mechanism and channel focusing module. The similarity calculation process can be expressed as:

[0142] ;

[0143] ;

[0144] in and is a learnable weight matrix, is the activation function, is the attention mapping matrix, Indicates a channel focus operation.

[0145] The three feature streams are dynamically weighted and combined through the feature fusion module. The fusion process first calculates the importance weight of each stream, and then combines the weighted features:

[0146] ;

[0147] ;

[0148] in , , represents the learnable fusion weights determined by the input features, is a learnable weight vector, represents the global pooling operation, represents the output features of the corresponding feature stream, It is the final output feature after fusion.

[0149] Specifically, in S102, the disease prediction model also dynamically adjusts the sample sampling weight based on the dynamic sampling module combined with the category balance mechanism and the multi-dimensional uncertainty evaluation strategy to enhance the model's learning ability for minority diseases and difficult samples. Figure 4 As shown in Figure 1, the dynamic sampling module includes a category balancing mechanism and a multi-dimensional uncertainty assessment mechanism:

[0150] (1) Category balancing mechanism: Calculate category-specific weights for disease samples based on their inverse frequency in the dataset. Disease level weights for samples Defined as:

[0151] ;

[0152] in Is the disease category The number of samples in is the total number of samples, is the total number of disease categories.

[0153] For the imbalance between diseases, a normalization factor is introduced to consider the relative sample distribution between disease categories and calculate the weight between diseases. :

[0154] ;

[0155] in Represents the number of samples in each disease category. The logarithmic formula provides a smooth adjustment to prevent over-weighting while compensating for imbalance between diseases.

[0156] Based on this, each sample is assigned a weight of the corresponding category to form the basic sampling weight:

[0157] ;

[0158] ;

[0159] (2) Multi-dimensional uncertainty assessment mechanism: Based on the model’s probability distribution of sample categories, the network’s logical output is converted into a probability estimate for each disease category through a Softmax transformation:

[0160] ;

[0161] in Indicates category The unnormalized logic output of Represents input Belongs to category The predicted probability of .

[0162] Based on this probability distribution, three complementary uncertainty assessment approaches are implemented:

[0163] Information entropy uncertainty: measures the dispersion of probability mass in sample prediction distribution, defined as:

[0164] ;

[0165] Confidence uncertainty: measures the gap between the model prediction result and the true category, defined as:

[0166] ;

[0167] in Representation sample The probability of being correctly classified.

[0168] Inter-class difference uncertainty: used to identify difficult samples near the decision boundary, evaluated by calculating the difference between the probabilities of adjacent categories in the predicted probability distribution:

[0169] ;

[0170] in Indicates the first A predicted probability.

[0171] These three uncertainty measures are combined into a composite uncertainty score:

[0172] ;

[0173] in , and They are the normalized entropy value, confidence difference and inter-class difference, respectively. , and is the weight coefficient of each indicator, satisfying In actual implementation, setting , and .

[0174] In order to achieve a smooth transition from category balance to hard sample focus, an iteratively dependent temporal adaptation mechanism is introduced:

[0175] ;

[0176]

[0177] in Indicates the current iteration number, represents the maximum number of iterations, It is a parameter that controls the ratio of uncertainty weights, which gradually increases as training progresses but does not exceed the preset maximum value. (set to 0.8), Set to 0.2, is the base weight, It is the normalized comprehensive uncertainty index.

[0178] In order to further optimize the sampling process of microbiome data, a temperature parameter adjustment mechanism was introduced. The temperature parameter controls the smoothness of the sampling weight distribution. In the early stage of training, a higher temperature is used to make the sampling more uniform, ensuring that the model has a basic understanding of various microbiome patterns; in the later stage of training, the temperature is lowered to make the sampling more focused on high-weight samples, refining the model's understanding of microbiome characteristics:

[0179] ;

[0180] in, and are the minimum and maximum values ​​of the temperature parameter, is the temperature reduction rate. In actual implementation, set , , .

[0181] Final Sample In the The sampling weight in the iteration is calculated as:

[0182] ;

[0183] Specifically, in S204, the fusion features are predicted for diseases by a random forest classifier, and the classification results and confidence levels are output. The classifier uses a random forest algorithm, which implements classification by constructing multiple decision trees and integrating their prediction results. In this embodiment, the number of decision trees is set to 500, the maximum tree depth is unlimited, the Gini impurity index is used for feature selection, and the minimum number of leaf node samples is 5. Each decision tree of the random forest randomly selects samples with replacement from the original training set during training, and randomly selects a feature subset (usually the square root of the total number of features) when the node is split. This randomness enables the model to have better generalization ability and robustness. In addition, the random forest can also output the probability estimate that each sample belongs to each category, which is convenient for subsequent uncertainty analysis. During the training process, the 5-fold cross-validation method is used to evaluate the model performance, and the grid search method is used to optimize the key hyperparameters, including the number of decision trees, the maximum number of features, and the minimum number of split samples.

[0184] To verify the effectiveness of the present invention, an experimental evaluation was conducted using the integrated intestinal microbiome data from 31 independent studies. The performance evaluation was conducted using a 5-fold cross-validation method, and a variety of evaluation indicators such as accuracy, precision, recall, F1 score, and AUC-ROC were used for comprehensive evaluation. The experimental results are shown in Figure 2. Figure 5 As shown in the figure, the method (MSAFI) of the embodiment of the present invention is significantly better than the existing methods in all evaluation indicators. In terms of overall classification accuracy, MSAFI reaches 68.96%±0.90%, F1 score is 66.53%±0.74%, and AUC-ROC value is as high as 90.96%±0.75%. It has obvious advantages over traditional machine learning methods such as random forest (accuracy 65.64%±1.95%, F1 score 63.79%±1.60%).

[0185] For details, see Figure 6 FIG. 1 is a system structure diagram of an embodiment of the present invention, including:

[0186] A preprocessing module 601 preprocesses the multi-source microbiome data samples to obtain hierarchical feature data samples including five levels: phylum, class, order, family, and genus;

[0187] A training module 602 constructs a disease prediction model and performs training using stratified feature data samples;

[0188] Prediction module 603, using the trained disease prediction model to perform disease prediction based on microbiome data;

[0189] The present invention makes full use of microbial classification hierarchical information to achieve multi-dimensional feature characterization, effectively address the problem of sample imbalance, and improve the diagnostic accuracy and stability of nervous system diseases.

[0190] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A disease prediction method based on adaptive fusion of hierarchical features of microbiome, characterized in that: The following steps are involved: Preprocess the multi-source microbiome data samples to obtain hierarchical feature data samples at five levels: phylum, class, order, family, and genus; Build a disease prediction model and use stratified feature data samples for training. During the training process, combine the category balance mechanism and multi-dimensional uncertainty assessment strategy to dynamically adjust the sample sampling weights. Use trained disease prediction models to predict diseases based on microbiome data; In the disease prediction model, the hierarchical feature extraction module performs multi-level feature extraction on the hierarchical feature data of the five classification levels, and integrates them to obtain multi-level fusion features containing hierarchical information; the multi-scale feature fusion module processes the multi-level fusion features through three parallel pathways, and outputs features containing structural information, environmental dependency information, and individual feature information respectively, and then integrates the features output by the three pathways through a dynamic weighted fusion mechanism to obtain multi-dimensional fusion features; The classifier predicts diseases based on multi-scale fusion features.

2. The disease prediction method based on adaptive fusion of hierarchical features of microbiome according to claim 1, characterized in that: The process of obtaining multi-level fusion features by the hierarchical feature extraction module includes the following steps: The hierarchical feature extraction component extracts features from five levels and combines them to obtain a hierarchical feature representation. , expressed as: ； ； ； in, , , , , Represents the characteristics of the five levels: phylum, class, order, family, and genus; the characteristics of each level All taxa at this level The aggregation abundance The vector composed of , represents the data matrix input into the disease prediction model, Indicates the corresponding taxon The eigenvector of It is a taxon The set of all descendant taxa composed of , by recursively using the child node mapping Calculated, for A taxon element in records the set of all descendants; Recorded by taxonomic unit is the set of all sub-classification units of the parent node, through the mapping It is constructed by reverse mapping the parent-child relationship information in; For known data, each microbial taxonomic unit is recorded The five levels and parent node information; The classification level weight component dynamically adjusts the feature weights to obtain weighted feature representation , expressed as: ； ； Where X represents the sample data matrix; It is a taxon The base weight of the level, Measures the discriminative power of features. Indicates the sparsity of the feature. and is the equilibrium parameter, represents a diagonal matrix with the adjusted weight vector as the diagonal element; The classification relationship modeling component constructs a multidimensional relationship matrix based on the classification path and evolutionary distance between features, and calculates the relationship feature representation through the multidimensional relationship matrix , expressed as: ； ； ； in, Representation characteristics and The common path length, and Represent the arrival characteristics and The length of the complete classification path; Representation characteristics and The evolutionary distance between Parameters to control the intensity of distance effects; Obtain multi-level fusion features through feature fusion , expressed as: ； is a learnable fusion weight matrix that is optimized during training by gradient descent.

3. The disease prediction method based on adaptive fusion of hierarchical features of microbiome according to claim 1, characterized in that: The multi-scale feature fusion module uses multi-level fusion features as input features. , processing multi-level fusion features through three parallel paths, and finally calculating the multi-dimensional fusion features, including the following steps: The pattern flow path uses the correlation attention mechanism to process the input features and obtain the pattern flow output features. , expressed as: ； ； ； in, , , are the learnable weight matrices for queries, keys, and values, respectively, , , denote the learned query, key, and value projections, respectively. is a scaling factor used to ensure the stability of the gradient flow during training; The context flow path uses a global scope processing module to establish long-range dependencies by calculating the similarity matrix between input features, and calculates the upstream and downstream output features through the feedforward neural network of the context flow. , expressed as: ； ； in, represents the global attention operation, and Represent two fully connected layer operations respectively, is the activation function, To prevent overfitting operations; The content flow channel processes the input features through the similarity mapping mechanism and the channel focusing module to obtain the content flow output features. , expressed as: ； ； in, and is a learnable weight matrix, is the activation function, is the attention mapping matrix, Indicates channel focus operation; Calculate the importance weight of each channel, and then perform weighted combination of the output features of each channel based on the weight to obtain multi-dimensional fusion features , expressed as: ； ； in, represents the learnable fusion weights corresponding to the three pathways, represents the learnable weight vector corresponding to the three pathways, represents the global pooling operation, Represents the output features corresponding to the three pathways; Indicates the mode flow path, Indicates the up and down flow path, Indicates the content flow path.

4. The disease prediction method based on adaptive fusion of hierarchical features of microbiome according to claim 1, characterized in that: Disease prediction is performed on multi-dimensional fusion features through classifiers. Specifically, disease prediction is performed on multi-scale fusion features through random forest classifiers, and classification results and confidence levels are output.

5. The disease prediction method based on adaptive fusion of hierarchical features of microbiome according to claim 1, characterized in that: The classification mode can be configured as binary classification or multi-classification according to actual prediction requirements; In the binary classification setting, the classification results are distinguished into disease status and non-disease status; In a multi-classification setting, the classification results can identify and differentiate into multiple different disease types.

6. The disease prediction method based on adaptive fusion of hierarchical features of microbiome according to claim 1, characterized in that: The dynamic adjustment of the sample sampling weight is implemented by a dynamic sampling module, including the following steps: The basic weight is calculated based on the category balancing mechanism and is expressed as: ； ； ； ； in, Representation sample The base weight of Representation sample Category The comprehensive weight of is the disease level weight, which represents the inverse frequency weight of the category; is the inter-disease weight, which represents the normalized weight of the relative sample distribution between categories; Is the disease category The number of samples in is the total number of samples, is the total number of disease categories; The composite uncertainty score is calculated based on the multi-dimensional uncertainty assessment mechanism and is expressed as: ； ； ； ； ； in, Representation sample The composite uncertainty score of , and The samples are The normalized entropy value of , confidence difference and inter-class differences , , and All represent weight coefficients; Represents input Belongs to category The predicted probability of Indicates category The unnormalized logic output of Representation sample The probability of being correctly classified; Indicates the first prediction probability; The comprehensive weight is calculated by combining the basic weight and the composite uncertainty score, which is expressed as: ； ； in, Indicates the current iteration number, represents the maximum number of iterations, It is a parameter that controls the ratio of uncertainty weights, which gradually increases as training progresses but does not exceed the preset maximum value. .

7. The disease prediction method based on adaptive fusion of hierarchical features of microbiome according to claim 6, characterized in that: The dynamic sampling module also includes a temperature parameter adjustment mechanism, which is expressed as: ； ； in, and are the minimum and maximum values ​​of the temperature parameter, is the rate at which the temperature is reduced. In the early stage of training, a higher temperature is used to make the sampling more uniform, and in the later stage of training, the temperature is reduced to make the sampling more focused on high-weight samples. Representation sample In the Final sampling weight calculation in iterations.

8. The disease prediction method based on adaptive fusion of hierarchical features of microbiome according to claim 1, characterized in that: The method of using the trained disease prediction model to predict diseases based on microbiome data includes the following steps: Collect and preprocess the microbiome data to be tested to obtain hierarchical feature data at five levels: phylum, class, order, family, and genus; The hierarchical feature data is input into the trained disease prediction model to obtain the classification results and confidence.

9. A disease prediction system based on adaptive fusion of hierarchical features of microbiome, characterized in that: include: The preprocessing module preprocesses the multi-source microbiome data samples to obtain hierarchical feature data samples at five levels: phylum, class, order, family, and genus; The training module builds a disease prediction model and uses stratified feature data samples for training. During the training process, the category balance mechanism and multi-dimensional uncertainty assessment strategy are combined to dynamically adjust the sample sampling weights. The prediction module uses the trained disease prediction model to predict diseases based on microbiome data; In the disease prediction model, the hierarchical feature extraction module performs multi-level feature extraction on the hierarchical feature data of the five classification levels, and integrates them to obtain multi-level fusion features containing hierarchical information; the multi-scale feature fusion module processes the multi-level fusion features through three parallel pathways, and outputs features containing structural information, environmental dependency information, and individual feature information respectively, and then integrates the features output by the three pathways through a dynamic weighted fusion mechanism to obtain multi-dimensional fusion features; The classifier predicts diseases based on multi-scale fusion features.

Images