Method for reconstructing gene network regulation relationship based on deep directed graph convolution

Abstract

The application discloses a kind of based on deep directed graph convolution gene network regulation relationship reconstruction method, comprising:1, using Node2Vec method obtains the topological information of Saccharomyces cerevisiae gene regulation network;2, for Saccharomyces cerevisiae factor regulation network using graph data enhancement means, obtain more efficient information expression;3, for the regulation direction problem of Saccharomyces cerevisiae gene regulation network, using directed graph neural network to predict the regulation relationship direction information;4, construct deep directed graph convolution model, obtain the neighbor information of gene high order;5, the expression characteristics of Saccharomyces cerevisiae gene and the characteristics obtained in steps 1, 2 are spliced and then input into model to obtain Saccharomyces cerevisiae regulation network prediction score matrix, according to the relationship between prediction score matrix and threshold value, judge whether there is regulation relationship between genes.The application can accurately predict the regulation relationship between Saccharomyces cerevisiae genes, and help to more efficiently study the nature of biochemical reaction of organism.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent information processing and bioinformatics, and particularly to a Saccharomyces cerevisiae gene regulation network reconstruction method based on a deep directed graph convolutional network. BACKGROUND

[0002] Genes are common attributes of all life, and the activity of genes is related to the expression level of genes, and there is a complex relationship between the expression products of genes (such as proteins and transcription factors) and their gene targets. Gene regulation networks (GRNs) explain the complex life phenomena of organisms at the genome level, and are important mechanisms for maintaining life processes, controlling biochemical reactions, and regulating the levels of chemical compounds. For cells, the control of biochemical levels by gene expression products allows cells to adjust to external environmental changes in a timely manner. Therefore, GRNs play an important role in predicting gene function, identifying cancer biomarkers, assisting in drug research and development, and discovering potential drug targets. Network reconstruction of Saccharomyces cerevisiae genes can better understand the regulation of Saccharomyces cerevisiae genes in environmental changes, and can better carry out industrial production.

[0003] Some existing methods for predicting gene regulation networks study the regulation mechanism between genes. For example, a method based on correlation theory: for a completely connected network composed of genes, the correlation between genes and genes is directly calculated as a reference for importance, and the results of the correlation are sorted, and some threshold values are set to select and construct the gene regulation network. Feature selection technology: is a kind of unsupervised learning method. The feature selection scheme often converts the p gene network construction problem into p independent feature selection problems, and uses feature selection technology to obtain a subset of regulatory factors that are strongly correlated with target factors. The GENIE3 method is based on regression trees, and these trees can be learned using the random forest method and the EXTREA-TREE method, which is a hybrid method for GRN inference, based on the formal dynamics model of each gene expression of the GRN, and uses a greedy, non-parametric method to reconstruct the topological structure of the GRN. In recent years, with the development of deep learning and neural networks, more and more research methods based on neural networks have appeared. They are based on supervised learning schemes: CNNC, DGRNS, GRGNN, TDL, DeepDRIM, etc.

[0004] The methods described above are based on traditional approaches such as information theory or traditional machine learning, or utilize basic convolutional neural networks or a combination of convolutional neural networks and recurrent neural networks. For naturally directed graph data like gene regulatory networks, using directed graph neural networks can be more efficient and convenient. However, due to the cumbersome data construction process and the difficulty in obtaining graph data information, directed graph neural networks are rarely used. Secondly, methods using only gene expression data tend to have poor prediction results because they fail to uncover deeper information, neglecting the natural topological structure of gene regulatory networks. Furthermore, the large number of low-density gene nodes in gene regulatory networks prevents previous models from being fully trained, resulting in performance degradation. Additionally, the sparsity and long regulatory chains in gene regulatory networks prevent these models from fully capturing information, leading to poor performance. Summary of the Invention

[0005] The present invention addresses the shortcomings of the existing technology by proposing a method for reconstructing gene network regulatory relationships based on deep directed graph convolution. This method aims to predict the regulatory relationships between genes more efficiently and accurately, thereby facilitating the study of gene regulatory relationships and promoting industrial production.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0007] The present invention provides a method for reconstructing gene network regulatory relationships based on deep directed graph convolution, characterized by the following steps:

[0008] Step 1: Obtain the gene network of Saccharomyces cerevisiae Where V is the set of gene nodes in *Saccharomyces cerevisiae*, and , This represents the nth Saccharomyces cerevisiae gene node; Let E represent the total number of gene nodes in *Saccharomyces cerevisiae*, and let E represent the set of directed edges formed by the regulatory relationships between these gene nodes. ; Indicates the first A directed edge, Let E represent the total number of directed edges, and let E be the adjacency matrix of dimension N×N. ;

[0009] Get The M-dimensional gene expression profile is thus obtained, and an expression profile matrix of dimension N×M is formed by the M-dimensional gene expression profiles of N Saccharomyces cerevisiae gene nodes. ;

[0010] Step 2: Construct a neighborhood enhancement network and use it for... Processing with A yields enhanced features. ;

[0011] Step 3, input into the Node2Vec model, generate the network neighborhood of by the neighborhood sampling strategy, and then traverse the network neighborhood of by the random walk method to obtain the 1xT-dimensional topological information embedding of ; thereby obtaining the topological information embedding of N Saccharomyces cerevisiae gene nodes and forming an NxT-dimensional network topological information embedding matrix F; wherein T represents the final representation dimension of each gene;

[0012] Step 4, after splicing and , a feature matrix with a dimension of is obtained, and is input into a feature fusion module together with for processing, so as to obtain a fusion feature matrix B by formula (1):

[0013] (1)

[0014] In formula (1), represents the message propagation form, represents the regularization, represents the splicing, represents the information learning weight matrix;

[0015] Step 5: construct a directed graph neural network, and process A, B and to obtain a regulatory relationship prediction score matrix ;

[0016] Step 6: use the stochastic gradient descent method to train the directed graph convolutional neural network, and calculate the cross-entropy loss function to update the network parameters until the cross-entropy loss function converges, thereby obtaining a trained Saccharomyces cerevisiae gene regulatory network inference model for outputting the prediction score of the regulatory relationship between genes, and finally determining whether there is a directed regulatory relationship by setting a threshold.

[0017] The method also has the characteristics that the neighborhood enhanced network in step 2 comprises an encoder network and a decoder network :

[0018] Step 2.1, sample the neighborhood of to obtain H neighborhood nodes of , so as to form a neighborhood node feature matrix with a dimension of HxM composed of the M-dimensional gene expression profiles of the H neighborhood nodes of .​ Then, the input is encoded in the encoder network to obtain a neighborhood node feature encoding vector with a dimension of Hx1;

[0019] Step 2.2, input the neighborhood node feature encoding matrix with a dimension of Hx1 into the decoder network to process and generate a neighborhood node feature decoding vector with a dimension of 1xM ; ;

[0020] Step 2.3, group the neighborhood node feature decoding vectors of the N Saccharomyces cerevisiae gene nodes into an enhanced feature matrix .

[0021] Further, the step 5 comprises:

[0022] Step 5.1, obtain the adjacency information between the first Saccharomyces cerevisiae gene node and the mth Saccharomyces cerevisiae gene node using formula (2) , so as to form an adjacency matrix from the adjacency information between the N Saccharomyces cerevisiae gene nodes;

[0023] (2)

[0024] In formula (2), represents the regulatory relationship between and , represents the regulatory relationship between and ;

[0025] Step 5.2, obtain the degree matrix of the first Saccharomyces cerevisiae gene node using formula (3) , so as to obtain the degree matrix of the N Saccharomyces cerevisiae gene nodes;

[0026] (3)

[0027] Step 5.3, construct the phase difference between the first Saccharomyces cerevisiae gene node and the mth Saccharomyces cerevisiae gene node under the condition of parameter q using formula (4) , so as to obtain the phase matrix under the condition of parameter q;

[0028] (4)

[0029] In formula (4), is the constant q, and q is a constant;

[0030] Construct the complex Hermitian matrix under parameter q using equations (5) and (6). :

[0031] (5)

[0032] (6)

[0033] In equations (5) and (6), Represents component multiplication. The imaginary unit is represented by equation (6), which is the expansion of the Euler equation.

[0034] Step 5.4: Construct the Laplace operator under parameter q using equation (7). :

[0035] (7)

[0036] Step 5.5: Use equation (8) to obtain the final direction information transmission paradigm. :

[0037] (8)

[0038] In equation (8), This means equivalent to; express In addition to the form of an identity matrix, and , It is the identity matrix. express Add the form of an identity matrix;

[0039] Step 5.6: Use equation (9) to obtain the first... Feature information matrix after layer learning :

[0040] (9)

[0041] In equation (9), Indicates weight, This represents the feature information matrix learned in the (l-1)th layer, when When =1, let =B;

[0042] Step 5.7: Obtain using equation (10) and The predictive score of the regulatory relationship between them Thus, a predictive score matrix of regulatory relationships for N Saccharomyces cerevisiae gene nodes was obtained. :

[0043] (10)

[0044] In formula (10), ( ) is a classification function, ( ) is a concatenation function, is the last layer of learned feature information the first row, the first row feature information vector.

[0045] The electronic device comprises a memory and a processor, and the memory is used for storing a program supporting the processor to execute the reconstruction method, and the processor is configured to execute the program stored in the memory.

[0046] The computer readable storage medium stores a computer program, and when the computer program is run by a processor, the steps of the reconstruction method are executed.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1. The gene regulatory network is a natural network data with network topology information. The present application uses a graph representation learning method to obtain the topology information of the graph, solves the problem that the neural network model only uses the expression value of the cell gene as the feature matrix of the neural network model input, and cannot mine more useful information from the expression data of the gene, and provides new features and ideas for the prediction of the gene regulatory relationship of the Saccharomyces cerevisiae.

[0049] 2. The present application uses a conditional variational autoencoder CVAE to enhance the local graph neural network, solves the problem that there are a large number of low-degree nodes in the GRN, and the directed graph neural network cannot effectively obtain the surrounding neighbor information, so that the model cannot efficiently predict the regulatory relationship between genes. The present application enhances the local features of the nodes in the graph neural network, obtains efficient feature representation, and enhances the expression ability of the graph neural network, thereby solving the problem that it is difficult to mine the correlation information of the low-degree gene nodes in the gene regulatory network.

[0050] 3. The present application uses a deep graph neural network to learn the knowledge carried by the high-order neighbors by traversing the deeper neighbor nodes. Therefore, the feature values of the genes can be extracted from the global perspective of the graph, and efficient feature representation can be learned, thereby improving the prediction performance of the Saccharomyces cerevisiae gene regulatory model.

[0051] 4. The application uses a directed graph convolutional neural network, solves the problem that the traditional convolutional neural network or the combination with the recurrent neural network does not pay attention to the directionality of the regulatory relationship in the process of message passing, so that the application can more accurately predict the gene regulatory relationship while efficiently processing the directed graph data. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is a flowchart of the method of the application;

[0053] Figure 2 is a structure diagram of the neighborhood enhanced network of the application;

[0054] Figure 3 is a schematic diagram of the Node2Vec model of the application;

[0055] Figure 4 is a schematic diagram of the feature fusion module of the application;

[0056] Figure 5 is a comparison result of the area under the receiver operating characteristic curve and the area under the precision recall curve of each method in five different networks in the DREAM4 simulation data under ten five-fold cross-validations, and other existing methods;

[0057] Figure 6 is a comparison result of the area under the receiver operating characteristic curve and the area under the precision recall curve of each method in three different networks in the E. coli real data under ten five-fold cross-validations, and other existing methods;

[0058] Figure 7 is a comparison result of the area under the receiver operating characteristic curve and the area under the precision recall curve of the application in the gene expression profile of S. cerevisiae, and other existing methods. DETAILED DESCRIPTION

[0059] In this embodiment, a gene regulatory relationship prediction method based on multi-layer directed graph convolution is as shown in the figure, which is performed according to the following steps: Figure 1

[0060] Step 1, obtaining the gene network of S. cerevisiae , wherein V is a set of S. cerevisiae gene nodes, and , represents the nth S. cerevisiae gene node; represents the total number of S. cerevisiae gene nodes, E represents a set of directed edges formed by the regulatory relationship between each S. cerevisiae gene node, and ; represents the th directed edge, ​Let E represent the total number of directed edges, and let E be the adjacency matrix of dimension N×N. ;

[0061] Get The M-dimensional gene expression profile is thus obtained, and an expression profile matrix of dimension N×M is formed by the M-dimensional gene expression profiles of N Saccharomyces cerevisiae gene nodes. ;

[0062] Step 2, Constructing the Neighborhood Enhancement Network, includes: Encoder Network and decoder network and used for Processing with A yields enhanced features. ,like Figure 2 As shown;

[0063] Step 2.1, for Sampling is performed on the neighborhood of [the target] to obtain [the desired result]. H neighboring nodes, thus by The M-dimensional gene expression profiles of H neighboring nodes constitute a feature matrix of neighboring nodes with dimension H×M. Then, it is input into the encoder network. Encode the feature vector of the neighborhood node with dimension H×1.

[0064] Step 2.2: Input the neighborhood node feature encoding matrix of dimension H×1 into the decoder network. The process is performed to generate a 1×M dimension. Neighborhood node feature decoding vector As shown in equation (11):

[0065] (11)

[0066] In equation (11), It is a decoder network activation function, It is an encoder network activation function, and Representing the decoder network and encoder network The weight matrix, It is a decoder network Middle offset term;

[0067] The encoder and decoder structures of the neighborhood enhancement network both consist of two-layer multilayer perceptrons. The encoder is composed of gene nodes. The input conditional information, and in the form of gene nodes The characteristics of the composition of neighboring gene nodes (The data information input to the encoder is processed by the encoder network to generate gene nodes.) The mean and standard deviation of the distribution are calculated. Then, a neighborhood node feature encoding matrix of dimension H×1 is sampled from the distribution obtained from the encoder and used as the input data for the decoder, ultimately yielding a 1×M matrix. Neighborhood node feature decoding vector .

[0068] Step 2.3: Combine the feature decoding vectors of the neighboring nodes of the N Saccharomyces cerevisiae gene nodes into an enhanced feature matrix. ;

[0069] Step 3, Input into the Node2Vec model, such as Figure 3 As shown, it is generated through a neighborhood sampling strategy. The network neighborhood, and then use a random walk to... Traverse the network neighborhood to obtain a 1×T dimensional [structure / database]. The topological information embedding is obtained; thus, the topological information embeddings of N Saccharomyces cerevisiae gene nodes are formed into an N×T dimensional network topological information embedding matrix F; where T represents the possible final representation dimension of each gene. In the experiment, T is set to 4 as the final output dimension.

[0070] Step 4, and After concatenation, the resulting dimension is Feature matrix and with They are input together into the feature fusion module for processing, such as Figure 4 As shown, the fusion feature matrix B is obtained using equation (1):

[0071] (1)

[0072] In equation (1), Indicates regularization, Indicates splicing, The information learning weight matrix represents the learning weight matrix. In this embodiment, the message propagation format is indicated. It is a message aggregation function used to first process... Filter by network neighborhood, using Calculate the embeddings of all nodes, then aggregate the embeddings of the selected neighboring nodes to obtain the node. The final embedding, thereby utilizing By processing N gene nodes in this way, we obtain the final embedding of N gene nodes.

[0073] Step 5: Constructing a directed graph neural network and processing A, B and to obtain the regulatory relationship prediction score matrix ;

[0074] Step 5.1, obtaining the adjacency information between the first, second and mth S. cerevisiae gene nodes by using (2), so as to construct an adjacency matrix composed of the adjacency information between the N S. cerevisiae gene nodes ; ;

[0075] (2)

[0076] In formula (2), represents the regulatory relationship between and , represents the regulatory relationship between and ;

[0077] Step 5.2, obtaining the degree matrix of by using formula (4), so as to obtain the degree matrix of the N S. cerevisiae gene nodes ;

[0078] (3)

[0079] In formula (3), represents the out-degree of the mth S. cerevisiae gene node in all of ;

[0080] Step 5.3, constructing the phase difference between and under the condition of parameter q by using formula (4), so as to obtain the phase matrix under the condition of parameter q ;

[0081] (4)

[0082] In formula (4), is the constant q, and

[0083] The complex Hermitian matrix under the condition of parameter q is constructed by using formula (5) and formula (6):

[0084] (5)

[0085] ​​​​​(6)

[0086] In formula (5) and formula (6), represents component multiplication, represents an imaginary unit, and formula (7) is an expansion form of Euler equation, when q is not equal to 0, can be used to encode the direction of the edge.

[0087] Step 5.4, constructing the Laplacian operator under the condition of the parameter q by using formula (7) :

[0088] (7)

[0089] Step 5.5, obtaining the final direction information transmission paradigm by using formula (8) :

[0090] (8)

[0091] In formula (8), represents equivalent to; represents plus the form of the unit matrix, and , is the unit matrix, represents plus the form of the unit matrix, represents out-degree information of, and , represents and adjacent information between plus the unit matrix.

[0092] Step 5.6: obtaining the feature information matrix of the first layer after learning by using formula (9) :

[0093] (9)

[0094] In formula (9), represents a weight, represents the feature information matrix of the l-1 layer after learning, when =1, let =B;

[0095] Step 5.7: obtaining the regulatory relationship prediction score between Figure 1 and by using formula (10) as shown, so as to obtain the regulatory relationship prediction score matrix of N Saccharomyces cerevisiae gene nodes :​​

[0096] (10)

[0097] In formula (10), is a classification function, is a concatenation function, is the last layer of learned feature information the first row, the first row feature information vector;

[0098] a prediction score matrix represents the prediction score between any two gene nodes, if the prediction score is greater than the set threshold, the corresponding two genes have a regulatory relationship, and the regulatory direction is from the regulatory factor to the target gene, otherwise, it means that there is no regulatory relationship between the two genes;

[0099] Step 6: training the directed graph convolutional neural network by using the stochastic gradient descent method, and calculating the cross-entropy loss function by using formula (12) to update the network parameters until the cross-entropy loss function converges, thereby obtaining a trained Saccharomyces cerevisiae gene regulatory network inference model for outputting the prediction score of the regulatory relationship between genes, and finally determining whether there is a directed regulatory relationship by setting a threshold.

[0100] (12)

[0101] In formula (12), represents the true label of A regulatory relationship with other genes, represents the prediction result of R regulatory relationship with other genes.

[0102] In this embodiment, an electronic device includes a memory for storing a program supporting a processor to execute the above method, and a processor configured to execute the program stored in the memory.

[0103] In this embodiment, a computer readable storage medium has a computer program stored thereon, and the computer program is executed by a processor to perform the steps of the above method.

[0104] To evaluate the performance of the model, the results are compared with single classifiers and other methods, and the evaluation index is the area under the receiver operating characteristic curve (AUROC). AUROC is the area under the receiver operating characteristic curve, and the value of this area is between 0 and 1, which can intuitively evaluate the performance of the classifier, and the larger the value of AUC, the better the performance of the classifier. Ten five-fold cross-validation experiments are performed, and the average value is taken as the model result.

[0105] Figure 5 The results of the model of the method of the application in the DREAM4 simulation data set are compared with other prediction models Jump3, dynamical GEne Network Inference with Ensemble of trees (dynGENIE3), Bidirectional eXtreme Gradient Boosting (BiXGBoost) and Multi-source Multi-model Fusion method (MMFGRN). The experimental results show that the results of the model of the application in the five network data of network 1 to network 5 of DREAM4 are better than those of other prediction models, verifying the effectiveness and superiority of the model.

[0106] Figure 6 The results of the model of the method of the application in the E. coli real data set are compared with other prediction models Jump3, dynGENIE3 and BiXGBoost. In the three network data of cold stress (Cold), heat stress (Heat) and oxidative stress (Oxidative), the prediction results of the model of the application in the real data set are also better than those of other prediction models, which shows that the model has superiority and can more accurately predict the regulatory relationship between genes.

[0107] Figure 7The implementation case results of the method model in the Saccharomyces cerevisiae data are shown, and the CNNGRN (a Convolutional Neural Network-Based Method for Gene Regulatory Network), XGBnet (XGBoost-Based Gene Network), iRafNet (Integrative Random Forest for Gene Regulatory Network) and PLSNET (Gene regulatory network inference using PLS-based methods) methods are compared. The experimental results show that the prediction results of the model in the actual case of the Saccharomyces cerevisiae data are better than those of other prediction models, and the feasibility and effectiveness of the model in the actual application are verified.

Claims

1. A method for reconstructing gene network regulatory relationships based on deep directed graph convolution, characterized in that, Includes the following steps: Step 1: Obtain the gene network of Saccharomyces cerevisiae Where V is the set of gene nodes in *Saccharomyces cerevisiae*, and , Indicates the first n One Saccharomyces cerevisiae gene node; Let E represent the total number of gene nodes in *Saccharomyces cerevisiae*, and let E represent the set of directed edges formed by the regulatory relationships between these gene nodes. ; Indicates the first A directed edge, Let E represent the total number of directed edges, and let E be the adjacency matrix of dimension N×N. ; Get The M-dimensional gene expression profile is thus obtained, and an expression profile matrix of dimension N×M is formed by the M-dimensional gene expression profiles of N Saccharomyces cerevisiae gene nodes. ; Step 2: Construct a neighborhood enhancement network and use it for... and Processing is performed to obtain enhanced features. ; Step 3, The data is input into the Node2Vec model and generated using a neighborhood sampling strategy. The network neighborhood, and then use a random walk to... Traverse the network neighborhood to obtain a 1×T dimensional [structure / database]. The topological information embedding is obtained; thus, the topological information embedding of N Saccharomyces cerevisiae gene nodes is obtained and formed into an N×T dimensional network topological information embedding matrix F; where T represents the final representation dimension of each gene; Step 4, and After concatenation, the resulting dimension is Feature matrix and with The inputs are fed into the feature fusion module for processing, and the fused feature matrix B is obtained using equation (1): (1) In equation (1), Indicates the form of message dissemination. Indicates regularization, Indicates splicing, Represents the information learning weight matrix; Step 5: Construct a directed graph convolutional neural network, and apply it to A, B, and... The process is performed to obtain the regulation relationship prediction score matrix. ; Step 6: Train the directed graph convolutional neural network using stochastic gradient descent and calculate the cross-entropy loss function to update the network parameters until the cross-entropy loss function converges, thereby obtaining the trained Saccharomyces cerevisiae gene regulation network inference model, which is used to output the predicted score of the regulatory relationship between genes, and finally determine whether there is a directed regulatory relationship by using a set threshold.

2. The method for reconstructing gene network regulatory relationships based on deep directed graph convolution according to claim 1, characterized in that, The neighborhood enhancement network in step 2 includes: an encoder network. and decoder network : Step 2.1, for Sampling is performed on the neighborhood of [the target] to obtain [the desired result]. H neighboring nodes, thus by The M-dimensional gene expression profiles of H neighboring nodes constitute a feature matrix of neighboring nodes with dimension H×M. Then, it is input into the encoder network. Encode the feature vector of the neighborhood node with dimension H×1. Step 2.2: Input the neighborhood node feature encoding matrix of dimension H×1 into the decoder network. The process is performed to generate a 1×M dimension. Neighborhood node feature decoding vector ; Step 2.3: Combine the feature decoding vectors of the neighboring nodes of the N Saccharomyces cerevisiae gene nodes into an enhanced feature matrix. .

3. The method for reconstructing gene network regulatory relationships based on deep directed graph convolution according to claim 2, characterized in that, Step 5 includes: Step 5.1: Obtain using equation (2) and the m A gene node in Saccharomyces cerevisiae Adjacency information between Thus, the adjacency matrix is ​​formed by the adjacency information between N Saccharomyces cerevisiae gene nodes. ; (2) In equation (2), express and The regulatory relationship between them express and The regulatory relationship between them; Step 5.2, using equation (3) to obtain degree matrix Thus, the degree matrix of N Saccharomyces cerevisiae gene nodes is obtained. ; (3) Step 5.3: Construct the system under parameter q using equation (4). and phase difference between Thus, the phase matrix under parameter q is obtained. ; (4) In equation (4), Pi is the mathematical constant, and q is a constant. Construct the complex Hermitian matrix under parameter q using equations (5) and (6). : (5) (6) In equations (5) and (6), Represents component multiplication. The imaginary unit is represented by equation (6), which is the expansion of the Euler equation. Step 5.4: Construct the Laplace operator under parameter q using equation (7). : (7) Step 5.5: Use equation (8) to obtain the final direction information transmission paradigm. : (8) In equation (8), This means equivalent to; express In addition to the form of an identity matrix, and , It is the identity matrix. express Add the form of an identity matrix; Step 5.6: Use equation (9) to obtain the first... Feature information matrix after layer learning : (9) In equation (9), Indicates weight, Indicates the first l- The feature information matrix after 1 layer of learning, when When =1, let =B; Step 5.7: Obtain using equation (10) and The predictive score of the regulatory relationship between them Thus, a predictive score matrix of regulatory relationships for N Saccharomyces cerevisiae gene nodes was obtained. : (10) In equation (10), ( ) is the classification function. ( ) is a concatenation function. It is the feature information learned in the last layer. The Middle line, number Row feature information vector.

4. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing any of the reconstruction methods described in claims 1-3, the processor being configured to execute the program stored in the memory.

5. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the steps of any of the reconstruction methods described in claims 1-3.

Images