A microbe-disease association prediction method and device based on a multi-level perception aggregation and deep hierarchical graph hybrid learner

Abstract

This invention discloses a method and apparatus for predicting microbial-disease associations based on a multi-level perception aggregation and deep hierarchical graph hybrid learner, comprising the following specific steps: S1: Obtain known microbial-disease association information and construct a microbial-disease association matrix A. 1 Simultaneously, a comprehensive microbial similarity matrix (CM), a microbial characterization mapping matrix (MM), a comprehensive disease similarity matrix (CD), and a disease characterization mapping matrix (DD) are constructed. S2: Based on CM, MM, CD, and DD, a multi-level perceptual aggregation strategy (MLPA) is applied to obtain the microbial hierarchical linkage feature matrix (LM) and the disease hierarchical linkage feature matrix (LD). This invention employs a multi-level perceptual aggregation strategy to enhance feature representation, combines a dual-drive association enhancement strategy to mine potential associations, and utilizes a deep hierarchical graph hybrid learner to achieve efficient learning and dynamic feature aggregation of complex heterogeneous network structures, significantly improving the accuracy and reliability of association prediction.

Description

A method and apparatus for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs. Technical Field

[0001] This invention relates to the field of biomedical engineering technology, specifically to a method and apparatus for predicting microbial-disease associations based on a multi-level perception aggregation and deep hierarchical graph hybrid learner. Background Technology

[0002] With the development of artificial intelligence technology, the study of microbe-disease associations has become a hot topic in the interdisciplinary research of bioinformatics and artificial intelligence. Research shows that imbalances in ecosystems such as the human gut microbiota and skin microbiome are closely related to a variety of complex diseases, including metabolic diseases, autoimmune diseases, neurological diseases, and even cancer. Against this backdrop, developing efficient and accurate methods and devices for predicting microbe-disease associations is of great significance for understanding disease mechanisms and clinical diagnosis and treatment.

[0003] In existing methods for predicting microbe-disease associations, network propagation methods construct microbe and disease similarity networks and use algorithms such as random walks and label propagation to spread known association information within the network, thereby predicting potential associations. These methods effectively utilize network topology, but are highly dependent on the quality of network construction and struggle to capture nonlinear association patterns. Matrix factorization methods, on the other hand, decompose the microbe-disease association matrix into low-dimensional feature matrices and reconstruct the association matrix through linear combinations to predict unknown associations. These methods are computationally efficient but have limited ability to handle sparse data and cannot effectively integrate multi-source heterogeneous information.

[0004] In recent years, Graph Neural Networks (GNNs) have demonstrated great potential in the field of microbe-disease association prediction, with Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs) being the two most representative methods. GCNs, through an efficient neighborhood information aggregation mechanism, can automatically learn low-dimensional feature representations of microorganisms and diseases. Their advantages lie in low computational complexity and ease of implementation, making them particularly suitable for processing large-scale biological network data. This method can capture network structural features at different scales through multi-layer convolutional operations, providing an end-to-end solution for microbe-disease association prediction. GATs, on the other hand, introduce an attention mechanism based on GCNs. By calculating dynamic weights between nodes, they can more finely characterize the differences in the strength of associations between microorganisms and diseases. The core advantage of GATs is their ability to automatically identify important associations in the network and assign them higher attention weights, thereby improving the model's prediction accuracy. These two methods, through deep learning technology, overcome the limitations of traditional methods in automatic feature extraction, providing a new technical path for microbe-disease association research.

[0005] Despite some progress in existing methods for predicting microbial-disease associations, several key issues remain: First, similarity information is not fully utilized, and there is a lack of multi-level feature fusion mechanisms for similarity information, resulting in incomplete and inaccurate prediction results. Second, the microbial-disease association matrix suffers from data sparsity, but existing methods for handling this problem have limitations. Third, existing graph neural network architectures in microbial-disease association prediction suffer from insufficient high-order interaction modeling, a lack of dynamic feature adjustment mechanisms, and excessive feature smoothing, which severely restricts the predictive performance of the models.

[0006] Based on the above, a method and apparatus for predicting microbial-disease associations based on a hybrid learner of multi-level perception aggregation and deep hierarchical graphs are invented. Summary of the Invention

[0007] To address the aforementioned technical problems, according to one aspect of the present invention, the present invention provides the following technical solution:

[0008] A microbiome-disease association prediction method based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs includes the following specific steps:

[0009] S1: Obtain known microorganism-disease association information and construct the microorganism-disease association matrix A1; at the same time, construct the comprehensive microorganism similarity matrix CM, the microorganism characterization mapping matrix MM, the comprehensive disease similarity matrix CD, and the disease characterization mapping matrix DD;

[0010] S2: Based on CM, MM, CD and DD, the multi-level perception aggregation strategy MLPA is applied to obtain the microbial hierarchical linkage feature matrix LM and the disease hierarchical linkage feature matrix LD.

[0011] S3: Based on A1, the dual-drive association enhancement strategy DDAES is applied to alleviate the sparsity problem of the association matrix data, resulting in an enhanced microbiome-disease association matrix A2;

[0012] S4: Using A2, CM, and CD, construct the adjacency matrix B and the initial feature matrix X0 of the microbial-disease heterogeneous graph. Then, use the deep hierarchical graph hybrid learner DHGHL to perform deep learning and dynamic aggregation of the features of microorganisms and diseases to obtain the microbial aggregation linkage feature matrix HM and the disease aggregation linkage feature matrix HD.

[0013] S5: Based on LM, HM, LD, and HD, reconstruct and predict the association between microorganisms and diseases, obtain the reconstruction prediction matrix A′, and output the final association probability between microorganisms and diseases.

[0014] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S1 are as follows:

[0015] S11: Obtain known microbial-disease association information, delete redundant and ambiguous entries in the obtained association information, and ensure the accuracy and consistency of the association information;

[0016] S12: Based on the collected microbial-disease association information, let nm and nd represent the number of microorganisms and diseases, respectively, to obtain the microbial-disease association matrix.

[0017] S13: Constructing a comprehensive microbial similarity matrix based on A1 and microbial characterization mapping matrix

[0018] S14: Constructing a comprehensive disease similarity matrix based on A1 and disease characterization mapping matrix

[0019] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S13 are as follows:

[0020] S131: Based on A1, for any two microorganisms m i and m j Calculate m i and m j Gaussian kernel similarity between SM Gip (m i ,m j );

[0021] S132: For any two microorganisms m i and m j Calculate m i With m j Cosine similarity between SM Cos (m i ,m j );

[0022] S133: For any two microorganisms m i and m j Calculate m i With m j Jaccard similarity between SM Jac (m i ,m j );

[0023] S134: For any two microorganisms m i and m j Calculate m i With m j Sigmoid kernel similarity between SM Sig (m i ,m j );

[0024] S135: Based on protein function association, the method of calculating the microbial-microbial functional association index is used to obtain the m for any given two microorganisms. i and m j Functional similarity between SM Fun (m i ,m j );

[0025] S136: Based on any two microorganisms m i and m j A comprehensive microbial similarity matrix was constructed by using Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity among the microorganisms. and microbial characterization mapping matrix

[0026] The specific steps of S14 are as follows:

[0027] S141: Based on A1, for any two diseases d i and d j Calculate d i With d j Gaussian kernel similarity between SD Gip (d i ,d j );

[0028] S142: For any two diseases d i and d j Calculate d i With d j Cosine similarity between SD Cos (d i ,d j );

[0029] S143: For any two diseases d i and d j Calculate d i With d j Jaccard similarity between SD Jac (d i ,d j );

[0030] S144: For any two diseases d i and d j Calculate d i With d j Sigmoid kernel similarity between SD Sig (d i ,d j );

[0031] S145: Based on gene interaction information, calculate d for any given two diseases. i and d j Functional similarity between SD Fun (d i ,d j );

[0032] S146: Based on any two diseases d i and d j A comprehensive disease similarity matrix is ​​constructed by using Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity among the samples. and disease characterization mapping matrix

[0033] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S2 are as follows:

[0034] S21: Based on CM and MM, the multi-level perception aggregation strategy MLPA is applied to obtain the microbial hierarchical linkage feature matrix LM;

[0035] S22: Based on CD and DD, the multi-level perception aggregation strategy MLPA is applied to obtain the disease hierarchical linkage feature matrix LD.

[0036] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S21 are as follows:

[0037] S211: Based on CM and MM, construct a three-level sensory feature matrix: microbial single-layer sensory feature matrix Microbial dual-layer sensing feature matrix and the three-layer sensory feature matrix of microorganisms

[0038] S212: Based on and Constructing a microbial composite sensing feature matrix and microbial equilibrium residual characteristic matrix

[0039] S213: Based on and A microbial hierarchical linkage feature matrix is ​​constructed through matrix concatenation operations.

[0040] The specific steps of S22 are as follows:

[0041] S221: Based on CD and DD, construct a three-level perceptual feature matrix: a single-layer perceptual feature matrix for diseases. Disease dual-layer perception feature matrix Disease three-layer perception feature matrix

[0042] S222: Based on and Constructing a disease composite perception feature matrix Disease equilibrium residual feature matrix

[0043] S223: Based on and A disease-linked feature matrix is ​​constructed through matrix concatenation.

[0044] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S3 are as follows:

[0045] S31: Constructing a microbial partner matrix based on A1

[0046] S32: Constructing a Disease Partner Matrix Based on A1

[0047] S33: Normalize M1 and D1 to obtain the normalized microbial partner matrix. and normalized disease partner matrix

[0048] S34: Based on A1, M norm D norm Construct a correlation potential score matrix

[0049] S35: Based on A pot For each disease d i Calculate the average potential score of its associated microorganisms.i ;

[0050] S36: Based on A1, A pot and score i Construct an enhanced microbe-disease association matrix.

[0051] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S4 are as follows:

[0052] S41: Construct the adjacency matrix of the microbiome-disease heterogeneity graph based on A2, CM, and CD. and the initial feature matrix

[0053] S42: Based on B and X0, the deep hierarchical graph hybrid learner DHGHL is applied to obtain the microbial aggregation linkage feature matrix HM and the disease aggregation linkage feature matrix HD.

[0054] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S42 are as follows:

[0055] S421: Based on B and X0, the first-layer collaborative representation matrix is ​​obtained using a graph convolutional network (GCN) in the first layer of DHGHL.

[0056] S422: Based on B and H (1) In the second layer of DHGHL, a graph convolutional network (GCN) is used to obtain the second-layer collaborative representation matrix.

[0057] S423: Based on H (1) and H (2) By applying the two-layer feature dynamic fusion strategy DL-FDFS to the third layer of DHGHL, the third-layer collaborative representation matrix is ​​obtained.

[0058] S424: Based on B and H (3) In the fourth layer of DHGHL, the enhanced dual-path graph attention network E-DPGAT is applied to further preserve the original features and explore higher-order structural information, resulting in the microbial aggregation linkage feature matrix HM and the disease aggregation linkage feature matrix HD.

[0059] The specific steps of S424 are as follows:

[0060] S4241: Based on B, construct the set of direct neighbor nodes for each node i. and indirect neighbor node set

[0061] S4242: Based on sets and H (3) Calculate the attention coefficient between each node i and its direct neighbor node j. Then normalization is performed to obtain...

[0062] S4243: Attention coefficients obtained from normalization We weight and aggregate the features of direct neighbors to obtain the aggregated feature vector of direct neighbors.

[0063] S4244: Based on the node's own feature vector h i Aggregate feature vectors with direct neighbors Construct the first path feature vector of the node

[0064] S4245: Based on sets and H (3) Calculate the attention coefficient between each node i and its indirect neighbor node j. Then normalization is performed to obtain...

[0065] S4246: Attention coefficients obtained from normalization We weight and aggregate the indirect neighbor features to obtain the indirect neighbor aggregated feature vector.

[0066] S4247: Based on the node's own feature vector h i Aggregate feature vectors with indirect neighbors Constructing the second path feature vector of the node

[0067] S4248: Based on and Construct the fourth-layer collaborative representation feature vector for any node i.

[0068] S4249: Based on Constructing a microbial aggregation and linkage feature matrix Disease aggregation and linkage feature matrix

[0069] As a preferred embodiment of the microbial-disease association prediction method based on multi-level perceptual aggregation and deep hierarchical graph hybrid learner described in this invention, the specific steps of S5 are as follows:

[0070] S51: Based on LM, HM, LD, and HD, multimodal feature splicing is performed to construct a reconstructed microbial feature matrix. and reconstructing the disease feature matrix

[0071] S52: Based on FM and FD, the association between microorganisms and diseases is reconstructed to obtain a reconstructed prediction matrix.

[0072] S53: Define a loss function to measure the difference between the reconstructed prediction matrix A′ and the true correlation matrix A1, and train the model parameters by optimizing the loss function;

[0073] S54: Based on A′, output the final probability of association between microorganisms and diseases.

[0074] A microbiome-disease association prediction device based on a hybrid learner of multi-level perception aggregation and deep hierarchical graphs, comprising:

[0075] The data acquisition and preprocessing module is used to automatically collect microbial-disease association information from authoritative biomedical databases and remove redundant and ambiguous entries from the acquired association information; then, it constructs a microbial-disease association matrix A1 to provide a standardized data foundation for subsequent feature calculations.

[0076] The feature calculation engine is used to extract various similarity information and enhance the association information. First, based on the microbiome-disease association matrix A1, a comprehensive similarity matrix and representation mapping matrix between microorganisms and diseases are constructed by integrating Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity. Then, an enhanced microbiome-disease association matrix A2 is generated through a dual-drive association enhancement strategy, providing high-information-density feature input for subsequent intelligent analysis and learning modules.

[0077] The intelligent analysis and learning module is used to apply a multi-level perception aggregation strategy and a deep hierarchical graph hybrid learner to achieve deep modeling of microbial and disease features, and to reconstruct and predict the association between microorganisms and diseases to obtain the association probability between microorganisms and diseases.

[0078] The interaction and output module is responsible for transforming complex prediction results into intuitive and usable information.

[0079] Compared with existing technologies:

[0080] 1. The Multi-Level Perceptual Aggregation (MLPA) strategy, based on the systematic integration of multiple similarities, uses progressive matrix multiplication to extract feature information layer by layer. It also designs two novel feature representation forms: composite perceptual feature matrix and balanced residual feature matrix. Through a nonlinear fusion mechanism, it realizes the dynamic optimization and fusion of multi-level features, effectively overcoming the limitations of traditional methods in the utilization of similarity features. It achieves in-depth mining and full utilization of similarity information of microorganisms and diseases, and significantly improves the comprehensiveness of feature representation and model robustness.

[0081] 2. The Dual-Drive Association Enhancement Strategy (DDAES) addresses the problem of data sparsity in the microbiome-disease association matrix. It effectively solves the problem of large fluctuations and poor stability in prediction results caused by data sparsity in existing sparse matrix processing methods, and significantly improves the model's ability to identify potential microbiome-disease associations.

[0082] 3. The Deep Hierarchical Graph Hybrid Learner (DHGHL) employs a unique four-layer architecture, systematically addressing key issues in existing graph neural networks for microbe-disease association prediction. The first and second layers of DHGHL utilize Graph Convolutional Networks (GCNs) to capture direct and indirect neighborhood features in heterogeneous microbe-disease graphs, respectively. The third layer employs a dual-layer dynamic feature fusion strategy (DL-FDFS) to adaptively weight and fuse direct and indirect neighborhood features, balancing the contributions of local and indirect neighborhood information. The fourth layer innovatively introduces an Enhanced Dual-Path Graph Attention Network (E-DPGAT) to accurately model the interaction relationships between one-hop direct neighbors and two-hop indirect neighbors. It also incorporates residual connections and layer normalization techniques to achieve efficient learning and dynamic feature aggregation of complex heterogeneous network structures. This learner effectively solves the long-range dependency modeling problem of traditional graph neural networks while avoiding excessive smoothing of node features, ensuring the complete preservation of key feature information. Attached Figure Description

[0083] Figure 1 is a schematic diagram of the overall process of the present invention;

[0084] Figure 2 is a diagram of the overall architecture of the DHGHL of the present invention;

[0085] Figure 3 is the overall design diagram of the E-DPGAT of the present invention. Detailed Implementation

[0086] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0087] This invention provides a method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs, as shown in Figures 1-3, and includes the following specific steps:

[0088] S1: Obtain known microorganism-disease association information and construct the microorganism-disease association matrix A1; at the same time, construct the comprehensive microorganism similarity matrix CM, the microorganism characterization mapping matrix MM, the comprehensive disease similarity matrix CD, and the disease characterization mapping matrix DD.

[0089] The specific steps of S1 are as follows:

[0090] S11: Obtain known microbial-disease association information, delete redundant and ambiguous entries in the obtained association information, and ensure the accuracy and consistency of the association information;

[0091] Example: Download known microbe-disease association information from the Disbiome database (https: / / disbiome.ugent). This database provides validated microbe-disease association information. The database initially contained 5573 associations, and the experiment retained 4351 non-redundant and non-fuzzy associations, involving 218 diseases and 1052 microbes.

[0092] S12: Based on the collected microbial-disease association information, let nm and nd represent the number of microorganisms and diseases, respectively, to obtain the microbial-disease association matrix. The specific construction method of A1 is as follows: For any given microorganism m i and disease d j If there is a known relationship between them, then A1(i,j) = 1; otherwise, A1(i,j) = 0. Here, A1(i,j) represents the value of the element in the i-th row and j-th column of A1.

[0093] S13: Constructing a comprehensive microbial similarity matrix based on A1 and microbial characterization mapping matrix

[0094] The specific steps of S13 are as follows:

[0095] S131: Based on A1, for any two microorganisms m i and m j Calculate m i and m j Gaussian kernel similarity between SM Gip (m i ,m j The specific calculation process is as follows:

[0096] SM Gip (m i ,m j )=exp(-λ1||IP(m i)-IP(m j )|| 2 )

[0097]

[0098] Where λ1 represents the normalized bandwidth parameter; ||.|| represents the Euclidean distance, ||.|| 2 IP(m) represents the square of the Euclidean distance. i ) represents a microorganism m constructed based on A1 i A binary associative vector used to record m i Information on associations with all diseases; IP(m) j ) represents a microorganism m constructed based on A1 j A binary associative vector used to record m j Information on associations with all diseases.

[0099] in:

[0100] The core idea of ​​the Gaussian Interaction Profile (GIP) is to use a Gaussian kernel function to transform the distance between the association vectors of entities into a similarity score. The advantages of GIP are its simplicity and efficiency; by mapping discrete interaction information to a continuous similarity space, it effectively captures latent patterns.

[0101] Euclidean distance is used to calculate the straight-line distance between two points in a multidimensional space.

[0102] S132: For any two microorganisms m i and m j Calculate m i With m j Cosine similarity between SM Cos (m i ,m j The specific calculation formula is as follows:

[0103]

[0104] Where · represents the dot product operation of vectors.

[0105] S133: For any two microorganisms m i and m j Calculate m i With m j Jaccard similarity between SM Jac (m i ,m j The specific calculation formula is as follows:

[0106]

[0107] Among them, |IP(m i )∩IP(m j )| represents m i and m j The number of diseases that are commonly associated; |IP(m i )∪IP(m j )| represents m i and m j The total number of all related, deduplicated diseases.

[0108] Jaccard similarity is an index used to measure the similarity between two sets. It is defined as the ratio of the size of the intersection to the size of the union of the two sets. It can intuitively reflect the proportion of common features and ignore the interference of unique features.

[0109] S134: For any two microorganisms m i and m j Calculate m i With m j Sigmoid kernel similarity between SM Sig (m i ,m j The specific calculation formula is as follows:

[0110]

[0111] Where tanh represents the hyperbolic tangent function; θ m Indicates the bias term;

[0112] Among them, Sigmoid kernel similarity is a nonlinear similarity measurement method based on the hyperbolic tangent function (tanh), which evaluates the correlation strength by calculating the nonlinear transformation of the dot product of two vectors.

[0113] S135: Based on protein function association, the method of calculating the microbial-microbial functional association index is used to obtain the m for any given two microorganisms. i and m j Functional similarity between SM Fun (m i ,m j );

[0114] in:

[0115] The calculation of microbial functional similarity is based on the proposed Microbial Functional Index (MFI) method, which quantifies the functional similarity between different microorganisms by analyzing protein-protein functional association networks in microbial genomes. Specifically, this method uses a database to retrieve protein-protein functional association networks and statistically analyzes information about microbial pairs within these networks, including the presence of network nodes (gene families) in at least one genome and the connections linking these gene families (i.e., edges in the network). Gene families are labeled to indicate the presence or absence of proteins in genomes A, B, or both, resulting in three types of gene families: those belonging to genome A, those belonging to genome B, and shared gene families. Such a network contains three types of undirected edges, connecting three different types of start and end nodes (A to A, A and B interconnected, B to B). The microbial functional association index between two microorganisms is defined as the total number of edges that must cross a gene family specifically encoded in one genome to connect to a gene family in another genome. Microbial functional similarity can be expressed by the following formula:

[0116]

[0117] Wherein, MFI represents the functional association index between microorganisms, i.e., the functional similarity between microorganisms, AB represents the number of edges connecting A and B, AA represents the number of edges connecting A to A, and BB represents the number of edges connecting B to B.

[0118] S136: Based on any two microorganisms m i and m j A comprehensive microbial similarity matrix was constructed by using Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity among the microorganisms. and microbial characterization mapping matrix

[0119] The specific construction process of CM is as follows:

[0120] First, calculate m. i and m j Comprehensive similarity between SM Com (m i ,m j The calculation formula is as follows:

[0121]

[0122] Then, the comprehensive similarities between all pairs of microorganisms are integrated to construct a CM. The integration process is as follows:

[0123]

[0124] In addition, the specific construction process of MM is as follows:

[0125] First, calculate m. i and m j Similarity of representation mappings between SM Map (m i ,m j The calculation formula is as follows:

[0126]

[0127] Then, the similarity of the representation maps between all pairs of microorganisms is integrated to construct a Model (MM). The integration process is as follows:

[0128]

[0129] Among them, comprehensive similarity is a concept used to measure the degree of similarity between two or more objects. It does not rely on a single similarity to make a judgment, but rather uses a combination of multiple similarity calculation methods to more comprehensively and accurately assess the similarity relationship between objects.

[0130] S14: Constructing a comprehensive disease similarity matrix based on A1 and disease characterization mapping matrix

[0131] The specific steps of S14 are as follows:

[0132] S141: Based on A1, for any two diseases d i and d j Calculate d i With d j Gaussian kernel similarity between SD Gip (d i ,d j The specific calculation process is as follows:

[0133] SD Gip (d i ,d j )=exp(-λ2||IP(d i )-IP(d j )|| 2 )

[0134]

[0135] Where λ2 represents the normalized bandwidth parameter; IP(d i ) represents the disease d constructed based on A1. i The binary associative vector, recording d i Information on associations with all microorganisms; IP(d) j ) represents the disease d constructed based on A1. j The binary associative vector, recording d j Information on associations with all microorganisms.

[0136] S142: For any two diseases d i and d j Calculate d i With d j Cosine similarity between SD Cos (d i ,d j The specific calculation formula is as follows:

[0137]

[0138] S143: For any two diseases d i and d j Calculate d i With d j Jaccard similarity between SD Jac (d i ,d j The specific calculation formula is as follows:

[0139]

[0140] Among them, |IP(d i )∩IP(d j )| represents d i and d j The number of co-related microorganisms; |IP(d i )∪IP(d j )| represents d i and d j The total number of all related, deduplicated microorganisms.

[0141] S144: For any two diseases d i and d j Calculate d i With d j Sigmoid kernel similarity between SD Sig (d i ,d j The specific calculation formula is as follows:

[0142]

[0143] Where, θ d This indicates the bias term.

[0144] S145: Based on gene interaction information, calculate d for any given two diseases. i and d j Functional similarity between SD Fun (d i ,d j );

[0145] One method for calculating functional similarity between diseases based on gene interaction information involves: first, deriving the relevant gene set for each disease; then, measuring the similarity between diseases based on the functional similarity between genes (assessed through log-likelihood scores). Specifically, the similarity between disease pairs is obtained by calculating the maximum functional similarity between relevant gene sets and comprehensively considering the functional associations of all genes.

[0146] S146: Based on any two diseases d i and d j A comprehensive disease similarity matrix is ​​constructed by using Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity among the samples. and disease characterization mapping matrix

[0147] The specific construction process of CD is as follows:

[0148] First, calculate d. i and d j Comprehensive similarity between SD Com (d i ,d j The calculation formula is as follows:

[0149]

[0150] Then, the comprehensive similarity between all pairs of diseases is integrated to construct a CD. The integration process is as follows:

[0151]

[0152] In addition, the specific construction process of DD is as follows:

[0153] First, calculate d. i and d jRepresentation mapping similarity between SD Map (d i ,d j The calculation formula is as follows:

[0154]

[0155] Then, the similarity of the representation maps between all pairs of diseases is integrated to construct a Data Type Map (DD). The integration process is as follows:

[0156]

[0157] S2: Based on CM, MM, CD and DD, the Multi-Level Perceptive Aggregation (MLPA) strategy is applied to obtain the microbial hierarchical linkage feature matrix LM and the disease hierarchical linkage feature matrix LD.

[0158] The specific steps of S2 are as follows:

[0159] S21: Based on CM and MM, the multi-level perception aggregation strategy MLPA is applied to obtain the microbial hierarchical linkage feature matrix LM;

[0160] The specific steps of S21 are as follows:

[0161] S211: Based on CM and MM, construct a three-level sensory feature matrix: microbial single-layer sensory feature matrix Microbial dual-layer sensing feature matrix and the three-layer sensory feature matrix of microorganisms and The construction process is as follows:

[0162]

[0163] Here, * denotes matrix multiplication.

[0164] S212: Based on and Constructing a microbial composite sensing feature matrix and microbial equilibrium residual characteristic matrix and The construction process is as follows:

[0165]

[0166] Sigmoid() is the activation function.

[0167] S213: Based on and A microbial hierarchical linkage feature matrix is ​​constructed through matrix concatenation operations. The specific construction process of LM is as follows:

[0168]

[0169] Here, concat() represents the horizontal concatenation operation of the matrix, which completes the expansion of the feature dimensions.

[0170] S22: Based on CD and DD, the multi-level perception aggregation strategy MLPA is applied to obtain the disease hierarchical linkage feature matrix LD;

[0171] The specific steps of S22 are as follows:

[0172] S221: Based on CD and DD, construct a three-level perceptual feature matrix: a single-layer perceptual feature matrix for diseases. Disease dual-layer perception feature matrix Disease three-layer perception feature matrix and The construction process is as follows:

[0173]

[0174] S222: Based on and Constructing a disease composite perception feature matrix Disease equilibrium residual feature matrix and The construction process is as follows:

[0175]

[0176] S223: Based on and A disease-linked feature matrix is ​​constructed through matrix concatenation. The specific construction process of LD is as follows:

[0177]

[0178] S3: Based on A1, the dual-driven association enhancement strategy (DDAES) is applied to alleviate the sparsity problem of the association matrix data, resulting in an enhanced microbiome-disease association matrix A2.

[0179] The specific steps of S3 are as follows:

[0180] S31: Constructing a microbial partner matrix based on A1 The specific construction process of M1 is as follows: For any given microorganism m i and microorganisms m j If there is at least one common associated disease among them, that is, if there is at least one disease d k If A1(i,k) = 1 and A1(j,k) = 1, then M1(i,j) = 1; otherwise, M1(i,j) = 0.

[0181] S32: Constructing a Disease Partner Matrix Based on A1 The specific construction process of D1 is as follows: For any given disease d i and disease d j If there is at least one common associated microorganism among them, that is, if there is at least one microorganism m k If A1(k,i)=1 and A1(k,j)=1, then D1(i,j)=1; otherwise, D1(i,j)=0.

[0182] S33: Normalize M1 and D1 to obtain the normalized microbial partner matrix. and normalized disease partner matrix

[0183] Among them, M norm The construction process is as follows:

[0184] For M norm For any element in M1: if M1(i,j)=0, then M norm (i,j)=0; if M1(i,j)=1, then M norm (i,j) is calculated according to the following formula:

[0185]

[0186] Where |S(m i )| indicates that in A1, there is a relationship with microorganism m i The number of associated diseases represents the number of elements in row i of A1 with a value of 1; |S(m j)| indicates that in A1, there is a relationship with microorganism m j Number of associated diseases; |S(m i )∩S(m j )| indicates that it is simultaneously with m i and m j The number of associated diseases.

[0187] In addition, D norm The construction process is as follows:

[0188] For D norm Any element in the array: If D1(i,j)=0, then D norm (i,j)=0; if D1(i,j)=1, then D norm (i,j) is calculated according to the following formula:

[0189]

[0190] Where |S(d i )| indicates that in A1, it is related to disease d i The number of associated microorganisms represents the number of elements in column i of A1 with a value of 1; |S(d j )| indicates that in A1, it is related to disease d j The number of associated microorganisms; |S(d i )∩S(d j )| indicates that it is simultaneously with d i and d j The number of associated microorganisms.

[0191] S34: Based on A1, M norm D norm Construct a correlation potential score matrix The specific construction process is as follows:

[0192] A pot =M norm *A1+A1*(D norm ) T

[0193] S35: Based on A pot For each disease d i Calculate the average potential score of its associated microorganisms. i score i The calculation formula is as follows:

[0194]

[0195] Wherein, N(d) i ) indicates in Apot In and d i A set of microorganisms with non-zero association potential scores; |N(d i )| indicates that in A pot In and d i The number of microorganisms with a non-zero association potential score; A pot (k,i) represents A pot The value of the element in the k-th row and i-th column is recorded in m. k and d i The potential score of the association between them.

[0196] S36: Based on A1, A pot and score i Construct an enhanced microbe-disease association matrix. The specific construction process is as follows: If A1(k,i) = 1, or the microorganism m k With disease d i In A pot The correlation potential score A in pot (k,i)>score i So, we assume m k With d i There is a correlation, so let A2(k,i) = 1. The constructed formula is as follows:

[0197]

[0198] S4: Using A2, CM, and CD, construct the adjacency matrix B and the initial feature matrix X0 of the microbial-disease heterogeneous graph. Then, use the deep hierarchical graph hybrid learner DHGHL to perform deep learning and dynamic aggregation of the features of microorganisms and diseases to obtain the microbial aggregation linkage feature matrix HM and the disease aggregation linkage feature matrix HD.

[0199] The specific steps of S4 are as follows:

[0200] S41: Construct the adjacency matrix of the microbiome-disease heterogeneity graph based on A2, CM, and CD. and the initial feature matrix The specific construction process is as follows:

[0201]

[0202] S42: Based on B and X0, the Deep Hierarchical Graph Hybrid Learner (DHGHL) is applied to obtain the microbial aggregation linkage feature matrix HM and the disease aggregation linkage feature matrix HD; the overall architecture diagram of DHGHL is shown in Figure 2.

[0203] The specific steps of S42 are as follows:

[0204] S421: Based on B and X0, the first-layer collaborative representation matrix is ​​obtained using a graph convolutional network (GCN) in the first layer of DHGHL. Get H (1) The specific process is as follows:

[0205]

[0206] in, It is the adjacency matrix normalized based on B, and the construction process is as follows:

[0207]

[0208] D is the degree matrix of B. It is the inverse of taking the square root of each element of D.

[0209] This is the first-layer weight matrix obtained through training, where d represents the feature map dimension; ELU is the activation function, and the complete formula is as follows:

[0210]

[0211] Among them, Graph Convolutional Network (GCN) is a deep learning model specifically designed for processing graph-structured data. By extending traditional convolution operations to graph data in non-Euclidean space, it enables efficient learning and propagation of node features.

[0212] S422: Based on B and H (1) In the second layer of DHGHL, a graph convolutional network (GCN) is used to obtain the second-layer collaborative representation matrix. Get H (2) The specific process is as follows:

[0213]

[0214] in, It is the weight matrix of the second layer obtained through training.

[0215] S423: Based on H (1) and H (2) The Double-Layer Feature Dynamic Fusion Strategy (DL-FDFS) is applied to the third layer of DHGHL to obtain the third-layer collaborative representation matrix. Get H(3) The specific process is as follows:

[0216] H (3) =α1H (1) +α2H (2)

[0217] Where α1 and α2 represent the weight contribution values ​​obtained during training.

[0218] In DHGHL: Direct neighborhood information is obtained through the first-layer GCN, effectively modeling the direct neighborhood features in the heterogeneous graph, and obtaining H... (1) The second-layer GCN is used to obtain deeper and more complex indirect neighborhood information, effectively modeling the indirect neighborhood features in heterogeneous graphs, and obtaining H. (2) The third layer, in order to integrate the direct and indirect neighborhood features obtained from the two-layer GCN, employs a DL-FDFS strategy to dynamically balance the contributions of direct and indirect neighborhood features, ultimately yielding H. (3) .

[0219] S424: Based on B and H (3) In the fourth layer of DHGHL, the Enhanced Dual-Path Graph Attention Network (E-DPGAT) is applied to further preserve the original features and explore higher-order structural information, resulting in the microbial aggregation linkage feature matrix HM and the disease aggregation linkage feature matrix HD; the overall design diagram of E-DPGAT is shown in Figure 3.

[0220] The specific steps of S424 are as follows:

[0221] S4241: Based on B, construct the set of direct neighbor nodes for each node i. and indirect neighbor node set That is, the neighboring nodes that node i can reach directly in one hop. The construction process is as follows: Based on B, if there exists a node j that satisfies B(i,j)=1, then node j belongs to... That is, the neighboring nodes that node i can reach in two hops. The construction process is as follows: based on B 2 If there exists a node j satisfying B 2 If (i,j) = 1, then node j belongs to

[0222] In the microbiome-disease heterogeneous graph, each node represents any microorganism or disease. In the following description, nodes include both microbiome nodes and disease nodes.

[0223] S4242: Based on sets and H (3) Calculate the attention coefficient between each node i and its direct neighbor node j. Then normalization is performed to obtain... The specific calculation process is as follows:

[0224]

[0225] LeakyReLU is an activation function that mitigates the vanishing gradient in the negative region. This is the attention parameter vector; The learned weight matrix; For from H (3) The feature vector of node i obtained from learning, i.e., H (3) The column vector consisting of all elements in the i-th row; For from H (3) The feature vector of node j obtained from learning, i.e., H (3) The column vector formed by all elements in the j-th row; || represents the vector concatenation operation, which aims to concatenate two feature vectors into a long vector.

[0226] S4243: Attention coefficients obtained from normalization We weight and aggregate the features of direct neighbors to obtain the aggregated feature vector of direct neighbors. The specific aggregation process is as follows:

[0227]

[0228] ReLU is the activation function.

[0229] S4244: Based on the node's own feature vector h i Aggregate feature vectors with direct neighbors Construct the first path feature vector of the node The specific construction formula is as follows:

[0230]

[0231] Here, LayerNorm represents the layer normalization operation; MLP is a multilayer perceptron used to process data and learn complex relationships between inputs and outputs. The purpose of using MLP here is to analyze W1h. i and The concatenation result is then dimensionality reduced so that the size of the reduced vector is the same as h. i The same applies, which facilitates subsequent operations and calculations.

[0232] S4245: Based on sets and H (3) Calculate the attention coefficient between each node i and its indirect neighbor node j. Then normalization is performed to obtain... The specific calculation process is as follows:

[0233]

[0234] in, This is the attention parameter vector; h is the learned weight matrix. i For from H (3) The feature vector of node i obtained from learning; h j For from H (3) The feature vector of node j obtained from learning.

[0235] S4246: Attention coefficients obtained from normalization We weight and aggregate the indirect neighbor features to obtain the indirect neighbor aggregated feature vector. The specific aggregation process is as follows:

[0236]

[0237] S4247: Based on the node's own feature vector h i Aggregate feature vectors with indirect neighbors Constructing the second path feature vector of the node The specific construction formula is as follows:

[0238]

[0239] The purpose of using MLP here is to analyze W2h. i and The concatenation result is then dimensionality reduced so that the size of the reduced vector is the same as h. i The same applies, which facilitates subsequent operations and calculations.

[0240] S4248: Based on and Construct the fourth-layer collaborative representation feature vector for any node i. The construction process is as follows:

[0241]

[0242] S4249: Based on Constructing a microbial aggregation and linkage feature matrix Disease aggregation and linkage feature matrix Here, HM is composed of the fourth-layer collaborative representation feature vectors of all microbial nodes; HD is composed of the fourth-layer collaborative representation feature vectors of all disease nodes. The specific construction formulas for HM and HD are as follows:

[0243]

[0244] in, Represents microbial node m k The corresponding fourth-layer collaborative representation feature vector; Represents disease node d k The corresponding fourth-layer collaborative representation feature vector.

[0245] S5: Based on LM, HM, LD, and HD, reconstruct and predict the association between microorganisms and diseases, obtain the reconstruction prediction matrix A′, and output the final association probability between microorganisms and diseases;

[0246] The specific steps of S5 are as follows:

[0247] S51: Based on LM, HM, LD, and HD, multimodal feature splicing is performed to construct a reconstructed microbial feature matrix. and reconstructing the disease feature matrix The specific process is as follows:

[0248] FM = concat(LM, HM)

[0249] FD = concat(LD, HD)

[0250] S52: Based on FM and FD, the association between microorganisms and diseases is reconstructed to obtain a reconstructed prediction matrix. The specific process design for the refactoring is as follows:

[0251] A′=g(FM*W′*FD T )

[0252] Here, g() represents the activation function, and the Sigmoid function is used as the activation function. It is the training matrix obtained through learning.

[0253] S53: Define a loss function to measure the difference between the reconstructed prediction matrix A′ and the true correlation matrix A1, and train the model parameters by optimizing the loss function; the formula for the loss function is as follows:

[0254]

[0255] in, This represents the Frobenius norm, which is the mean square error of the matrix elements.

[0256] S54: Based on A′, output the final association probability between the microorganism and the disease; the element A′(i,j) in A′ represents the microorganism m. i With disease d j The probability of association between them.

[0257] A microbiome-disease association prediction device based on a hybrid learner of multi-level perception aggregation and deep hierarchical graphs, comprising:

[0258] The data acquisition and preprocessing module is used to automatically collect microbial-disease association information from authoritative biomedical databases and remove redundant and ambiguous entries from the acquired association information; then, it constructs a microbial-disease association matrix A1 to provide a standardized data foundation for subsequent feature calculations.

[0259] The feature calculation engine is used to extract various similarity information and enhance the association information. First, based on the microbiome-disease association matrix A1, a comprehensive similarity matrix and representation mapping matrix between microorganisms and diseases are constructed by integrating Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity. Then, an enhanced microbiome-disease association matrix A2 is generated through the dual-drive association enhancement strategy (DDAES), providing high-information-density feature input for subsequent intelligent analysis and learning modules.

[0260] The intelligent analysis and learning module is used to apply the multi-level perceptual aggregation strategy (MLPA) and the deep hierarchical graph hybrid learner (DHGHL) to achieve deep modeling of microbial and disease features, and to reconstruct and predict the association between microorganisms and diseases to obtain the association probability between microorganisms and diseases.

[0261] The interaction and output module is responsible for transforming complex prediction results into intuitive and usable information. As the user interface layer of the microbiome-disease association prediction device, this module provides a visual interactive interface. The system automatically matches information from the database and calls up the association between microorganisms and diseases to reconstruct the prediction results, ultimately outputting the microbiome-disease association prediction results in a visual form, facilitating rapid interpretation and decision-making by medical professionals.

[0262] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs, characterized in that, The specific steps are as follows: S1: Obtain known microorganism-disease association information and construct a microorganism-disease association matrix. Simultaneously, a comprehensive microbial similarity matrix was constructed. Microbial characterization mapping matrix Disease comprehensive similarity matrix Disease characterization mapping matrix S2: Based on 、 、 and By applying the multi-level perception aggregation strategy (MLPA), the microbial hierarchical linkage feature matrix was obtained. Disease-related hierarchical linkage feature matrix S3: Based on The dual-drive association enhancement strategy (DDAES) was applied to alleviate the sparsity problem of the association matrix data, resulting in an enhanced microorganism-disease association matrix. S4: Utilize 、 、 Construct the adjacency matrix of the microbiome-disease heterogeneity graph. and the initial feature matrix By using a deep hierarchical graph hybrid learner (DHGHL) to perform deep learning and dynamic aggregation of features of microorganisms and diseases, a microbial aggregated linkage feature matrix is ​​obtained. Disease aggregation and linkage feature matrix S5: Based on 、 、 、 The association between microorganisms and diseases is reconstructed and predicted to obtain a reconstructed prediction matrix. And output the final association probability between microorganisms and diseases; the specific steps of S4 are as follows: S41: Based on 、 、 Construct the adjacency matrix of the microbiome-disease heterogeneity graph. and the initial feature matrix S42: Based on and The deep hierarchical graph hybrid learner (DHGHL) was applied to obtain the microbial aggregation and linkage feature matrix. Disease aggregation and linkage feature matrix The specific steps of S42 are as follows: S421: Based on and In the first layer of DHGHL, a graph convolutional network (GCN) is used to obtain the first-layer collaborative representation matrix. ； S422: Based on and In the second layer of DHGHL, a graph convolutional network (GCN) is used to obtain the second-layer collaborative representation matrix. S423: Based on and By applying the two-layer feature dynamic fusion strategy DL-FDFS to the third layer of DHGHL, the third-layer collaborative representation matrix is ​​obtained. ； S424: Based on and In the fourth layer of DHGHL, the enhanced dual-path graph attention network E-DPGAT is applied to further preserve the original features and explore higher-order structural information, resulting in a microbial aggregation and linkage feature matrix. Disease aggregation and linkage feature matrix The specific steps of S424 are as follows: S4241: Based on Build each node The set of direct neighbor nodes and indirect neighbor node set S4242: Based on sets and Calculate each node and its direct neighbor nodes Attention coefficient between Then normalize to obtain S4243: Attention coefficients obtained from normalization We weight and aggregate the features of direct neighbors to obtain the aggregated feature vector of direct neighbors. S4244: Based on the node's own feature vector Aggregate feature vectors with direct neighbors Construct the first path feature vector of the node S4245: Based on sets and Calculate each node and its indirect neighbor nodes Attention coefficient between Then normalize to obtain S4246: Attention coefficients obtained from normalization We weight and aggregate the indirect neighbor features to obtain the indirect neighbor aggregated feature vector. S4247: Based on the node's own feature vector Aggregate feature vectors with indirect neighbors Construct the second path feature vector of the node S4248: Based on and Construct for any node The fourth layer collaborative representation feature vector S4249: Based on Construct a microbial aggregation and linkage feature matrix Disease aggregation and linkage feature matrix 。 2. The method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs according to claim 1, characterized in that, The specific steps of S1 are as follows: S11: Obtain known microorganism-disease association information, delete redundant and ambiguous entries in the obtained association information, and ensure the accuracy and consistency of the association information; S12: Based on the collected microorganism-disease association information, let nm and nd represent the number of microorganisms and diseases, respectively, to obtain the microorganism-disease association matrix. S13: Based on Constructing a comprehensive microbial similarity matrix and microbial characterization mapping matrix S14: Based on Constructing a comprehensive disease similarity matrix Disease characterization mapping matrix 。 3. The method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs according to claim 2, characterized in that, The specific steps of S13 are as follows: S131: Based on For any two microorganisms and ,calculate and Gaussian kernel similarity between S132: For any two microorganisms and ,calculate and Cosine similarity between S133: For any two microorganisms and ,calculate and Jaccard similarity between S134: For any two microorganisms and ,calculate and Sigmoid kernel similarity between S135: Based on protein-microbe functional association, the method of calculating the microbial-microbe functional association index is used to obtain the functional association index for any given two microorganisms. and Functional similarity between S136: Based on any two microorganisms and A comprehensive microbial similarity matrix was constructed by using Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity among the microorganisms. and microbial characterization mapping matrix The specific steps of S14 are as follows: S141: Based on For any two diseases and ,calculate and Gaussian kernel similarity between S142: For any two diseases and ,calculate and Cosine similarity between S143: For any two diseases and ,calculate and Jaccard similarity between S144: For any two diseases and ,calculate and Sigmoid kernel similarity between S145: Based on gene interaction information, calculate the interaction between any two given diseases. and Functional similarity between S146: Based on any two diseases and A comprehensive disease similarity matrix is ​​constructed by using Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity among the samples. Disease characterization mapping matrix 。 4. The method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs according to claim 1, characterized in that, The specific steps of S2 are as follows: S21: Based on and By applying the multi-level perception aggregation strategy (MLPA), the microbial hierarchical linkage feature matrix was obtained. S22: Based on and By applying the multi-level perception aggregation strategy MLPA, a disease-related hierarchical linkage feature matrix is ​​obtained. 。 5. The method for predicting microbial-disease associations based on a multi-level perceptual aggregation and deep hierarchical graph hybrid learner according to claim 4, characterized in that, The specific steps of S21 are as follows: S211: Based on and A three-tiered sensory feature matrix was constructed: a microbial single-layer sensory feature matrix. Microbial dual-layer sensing feature matrix and the three-layer sensory feature matrix of microorganisms S212: Based on 、 and Constructing a microbial composite sensing feature matrix and microbial equilibrium residual characteristic matrix ； S213: Based on 、 、 、 and By performing matrix concatenation operations, a microbial hierarchical linkage feature matrix is ​​constructed. The specific steps of S22 are as follows: S221: Based on and Construct a three-layer perception feature matrix: a single-layer disease perception feature matrix. Disease dual-layer perception feature matrix Disease three-layer perception feature matrix S222: Based on 、 and Construct a disease composite perception feature matrix Disease equilibrium residual feature matrix S223: Based on 、 、 、 and By performing matrix concatenation operations, a disease-linked feature matrix is ​​constructed. 。 6. The method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs according to claim 1, characterized in that, The specific steps of S3 are as follows: S31: Based on Constructing a microbial partner matrix S32: Based on Building a disease partner matrix S33: To 、 Normalization was performed to obtain the normalized microbial partner matrix. and normalized disease partner matrix S34: Based on 、 、 Construct a correlation potential score matrix S35: Based on For each disease Calculate the average potential score of its associated microorganisms. S36: Based on 、 and Construct an enhanced microbe-disease association matrix. 。 7. The method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs according to claim 1, characterized in that, The specific steps of S5 are as follows: S51: Based on 、 、 、 Multimodal feature splicing is performed to construct a reconstructed microbial feature matrix. and reconstructing the disease feature matrix S52: Based on and The association between microorganisms and diseases is reconstructed to obtain a reconstructed prediction matrix. S53: Define a loss function to measure the reconstructed prediction matrix. Correlation Matrix The differences between them are analyzed, and the model parameters are trained by optimizing the loss function; S54: Based on The output is the probability of the association between the microorganism and the disease.

8. A method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs according to claim 1, comprising a prediction apparatus for running the method for predicting microbial-disease associations based on a hybrid learner of multi-level perceptual aggregation and deep hierarchical graphs, characterized in that, The prediction device includes: a data acquisition and preprocessing module, used to automatically acquire microbial-disease association information from authoritative biomedical databases, and delete redundant and ambiguous entries in the acquired association information; subsequently, a microbial-disease association matrix is ​​constructed. This provides a standardized data foundation for subsequent feature calculations; the feature calculation engine is used to extract various similarity information and enhance the association information; firstly, based on the microbiome-disease association matrix... By integrating Gaussian kernel similarity, cosine similarity, Jaccard similarity, Sigmoid kernel similarity, and functional similarity, a comprehensive similarity matrix and representation mapping matrix between microorganisms and diseases are constructed. Subsequently, an enhanced microorganism-disease association matrix is ​​generated through a dual-drive association enhancement strategy. The first module provides high-information-density feature inputs for subsequent intelligent analysis and learning modules. The second module applies a multi-level perception aggregation strategy and a deep hierarchical graph hybrid learner to achieve deep modeling of microbial and disease features, and reconstructs and predicts the association between microorganisms and diseases to obtain the association probability between microorganisms and diseases. The third module is responsible for transforming complex prediction results into intuitive and usable information.

Images