A dimensionality reduction modeling method and system for gene regulatory networks

Abstract

A dimensionality reduction modeling method and system for a gene regulatory network, belonging to the gene regulatory network construction technology, involving the fields of digital medicine and artificial intelligence. The present invention is proposed to effectively solve the contradiction between accuracy and computational complexity in the modeling process, and to achieve an exponential reduction in computational complexity while ensuring the accuracy of modeling. Technical points: First, the genome sequencing data is fused with the upstream and downstream prior knowledge of genes to improve the accuracy of gene clustering, and the genes are clustered using a clustering network. The genome is pre-divided by gene sequence and prior knowledge, and each gene is divided into different small sample branches for gene regulatory network modeling, so as to reduce the amount of computation required for each modeling. Then, a differential equation model of the gene regulatory network of each small sample branch is constructed, and finally, the regulatory networks of each branch are integrated and summarized into the gene regulatory network of the genome. The present invention is applicable to the process of large genome modeling.

Description

Technical Field

[0001] The present invention relates to a dimensionality reduction modeling method and system for a gene regulatory network, and belongs to the field of gene regulatory network construction technology, digital medicine, artificial intelligence, etc. Background Art

[0002] Gene regulatory network is a tool used to discover the interaction relationship between biological information substances such as genes and genes, genes and proteins, or proteins and proteins. The construction of gene regulatory network starts with input data, which usually consists of gene expression data sets. After the input data set is selected, data preprocessing can be performed to improve the modeling effect. Then, the preprocessed data is used as the input of the computational reasoning algorithm to obtain the final network. Among the computational reasoning algorithms, the most effective computational reasoning algorithm is the differential equation model. This method uses continuous variables instead of discrete variables, which can build more accurate models and make dynamic modeling of gene regulation possible. However, the differential equation model cannot construct large-scale gene regulatory networks, and in some cases, due to the complexity of the calculation, it becomes very difficult to estimate the value of the model parameters.

[0003] The prior art with document number CN114155913A discloses a gene regulatory network construction method based on high-order dynamic Bayes, which constructs a high-order dynamic Bayesian gene regulatory network based on time variables by utilizing temporal gene expression data, and searches for parent nodes for target genes under different time delays, so as to improve the accuracy of network structure construction. Since the search space of high-order network structure learning is very large, the time complexity of structure learning is particularly high. In order to speed up the learning speed, a method for screening potential regulatory factors before structure learning is proposed, and genes with a higher degree of association with target genes are first screened out as potential regulatory factor sets of target genes, and genes with a lower degree of association with target genes are deleted, so as to achieve the purpose of reducing the search space, and the speed of network structure learning can be accelerated later. However, how to effectively solve the contradiction between precision and calculation amount in the modeling process, and realize the exponential reduction of calculation amount under the condition of ensuring modeling accuracy, is not mentioned.

[0004] The prior art with document number CN116246713A discloses a single-cell tumor microenvironment data analysis method based on a gene regulatory network, which includes the following steps: obtaining the original data values ​​of several cells, and performing preprocessing and characteristic gene identification; constructing a Degree gene characterization matrix based on the regulatory relationship between the obtained characteristic genes, and identifying cell subpopulations; analyzing the cell entropy, identifying differential Degree genes, and analyzing gene function enrichment for the cell subpopulations. This prior art uses the SCILE algorithm to evaluate the overall stemness entropy value of the cell based on the importance of each gene in the gene regulatory network of each cell. Compared with the traditional evaluation method based on expression, it avoids the influence of high-expression and low-regulated genes on cell stemness. In this application scenario, no technical means are given for how to reduce the dimensionality of the gene regulatory network.

[0005] The most commonly used gene regulatory network inference algorithm is the Bayesian model, which uses the Bayesian theorem of probability and combines probability and graph theory to qualitatively model the properties of gene regulatory networks. This model does not have the high modeling accuracy of the differential equation model. However, the differential equation model cannot be applied to the construction of gene regulatory networks of large genomes due to the large amount of calculation. Therefore, it is urgent to propose a dimensionality reduction modeling method for gene regulatory networks to effectively solve the contradiction between accuracy and calculation in the modeling process, and achieve an exponential reduction in calculation while ensuring modeling accuracy. Summary of the invention

[0006] The technical problems to be solved by the present invention are:

[0007] In order to effectively resolve the contradiction between accuracy and computational complexity in the modeling process, the present invention achieves an exponential reduction in computational complexity while ensuring modeling accuracy. In order to solve the problem that the computational complexity required in the large genome modeling process is too large and thus modeling cannot be performed, a dimensionality reduction modeling method and system for gene regulatory networks are proposed.

[0008] The technical solution adopted by the present invention to solve the above technical problems is:

[0009] The dimensionality reduction modeling method of a gene regulatory network described in the present invention comprises the following steps:

[0010] S1, fusing genome sequencing data with prior knowledge of upstream and downstream genes to obtain fused data;

[0011] S2. Divide the genome fusion data into small genome samples: cluster the genome fusion data through an unsupervised deep learning network, output the gene sample branch sequence to which each gene belongs, and divide each base into different gene small sample branches;

[0012] S3, extracting the corresponding RNA-seq data set from the gene database according to the elements in the small sample branch of the genome;

[0013] S4, construct differential equation models of gene regulatory networks in each branch;

[0014] S5. Summarize the gene regulatory networks of each branch to obtain a complete gene regulatory network.

[0015] Further:

[0016] S1, genome sequencing data g i Prior knowledge of upstream and downstream genes ij Perform data fusion to obtain fused data c i ;

[0017] S2. Divide the genome fusion data into small samples: Use an unsupervised deep learning network to divide the genome fusion data into small samples i Perform clustering and output the gene sample branch sequence d to which each gene belongs i , each gene is divided into different gene small sample branches D according to formula 2 k ;

[0018] M(c)=d,d=1,2,…,n (1)

[0019] Where M represents the unsupervised deep learning network; n represents the number of genome sample branches;

[0020] D k ={c i |M(c i )=k},k=1,2,…,m (2)

[0021] In the formula, m represents the number of small sample branches of genes;

[0022] S3, branch D based on small genome samples k The elements in extract the corresponding RNA-seq data set R from the gene database k ;

[0023] S4 constructs each gene regulatory network branch G k Differential equation model: Construct the gene regulatory network branch G through formula 3 k , where the coefficient a j Greater than zero means gene g i Gene g j Positive regulation, a j Equal to zero means gene g i Not affected by gene j Regulation, a j Less than zero means gene g iGene g j Negative regulation;

[0024]

[0025] Where r i Indicates gene g i Corresponding RNA-seq data, n k Indicates the genome small sample branch D k The number of genes in a j For gene g j For gene g i The control factor of , b is a constant;

[0026] S5. Summarize the gene regulatory network branches to obtain the gene regulatory network G.

[0027] G={G1,G2,…,G k} (4)

[0028] The data fusion method described in step S1 comprises the following steps:

[0029] S11. Extract genome sequencing data through feature extraction network to obtain feature vector f i :

[0030] F(g i ) = f i (5)

[0031] Where F represents the feature extraction network;

[0032] S12, encoding the prior knowledge of upstream and downstream genes through the encoding network;

[0033] S13, performing data fusion on the genome sequencing feature vector obtained in step S11 and step S12 and the prior knowledge encoding. This data fusion method fuses the prior knowledge into the genome sequencing feature, which can effectively improve the gene association information and improve the clustering accuracy.

[0034] The present invention can use graph neural networks for clustering, and qualified networks include DAEGC, SDCN or AGC. The network used for clustering is DAEGC, SDCN or AGC or a fusion of the three; the genome sequencing data is extracted through the feature extraction network to obtain a feature vector; the upstream and downstream gene prior knowledge is encoded through the encoding network; the obtained genome sequencing feature vector is fused with the prior knowledge encoding, and the data fusion method fuses the prior knowledge into the genome sequencing features to improve the gene association information and clustering accuracy.

[0035] The application of the dimensionality reduction modeling method of the gene regulatory network is to use the dimensionality reduction modeling method of the gene regulatory network for large genome modeling. In a broad sense, large genomes (Genome Size) usually refer to genome sizes exceeding 5Gb, mostly concentrated in gymnosperms, amphibians, reptiles, etc. They often have ultra-high repetitive sequences and a large number of heterozygous regions. For example, the genome size of angiosperms (flowering plants) varies by 2400 times (1C = 0.063-148.8Gb), and the average genome size is 1C = 5.7Gb.

[0036] The present invention has the following beneficial technical effects:

[0037] A dimensionality reduction modeling method for a gene regulatory network proposed in the present invention first fuses genome sequencing data with prior knowledge of upstream and downstream genes to improve the accuracy of gene clustering. In order to solve the problem that the amount of calculation required for the large genome modeling process is too large to lead to the inability to model, the genes are clustered using a clustering network, the genome is pre-divided by gene sequence and prior knowledge, and each gene is divided into different small sample branches for gene regulatory network modeling to reduce the amount of calculation required for each modeling. Then, a differential equation model of the gene regulatory network of each small sample branch is constructed, and finally, each branch regulatory network is integrated to form the genome gene regulatory network.

[0038] The present invention proposes a dimensionality reduction modeling method for gene regulatory networks. The main idea is to divide the genome into small sample branches through a clustering network, and to model different branches separately to reduce the modeling complexity, reduce the amount of calculation in the modeling process, and improve the robustness of the modeling. Experiments show that our gene regulatory network dimensionality reduction modeling method can effectively solve the contradiction between accuracy and calculation in the modeling process, and achieve an exponential reduction in the amount of calculation while ensuring the accuracy of modeling. It effectively solves the problem that the amount of calculation required in the large genome modeling process is too large, resulting in the inability to model. It is fully suitable for the construction of gene regulatory networks for large genomes, and greatly improves the application scope of gene regulatory networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flowchart of the dimensionality reduction modeling method for the gene regulatory network described in the present invention. DETAILED DESCRIPTION

[0040] S1, fusing genome sequencing data with prior knowledge of upstream and downstream genes to obtain fused data;

[0041] S2. Divide the genome fusion data into small genome samples: cluster the genome fusion data through an unsupervised deep learning network, output the gene sample branch sequence to which each gene belongs, and divide each base into different gene small sample branches;

[0042] S3, extracting the corresponding RNA-seq data set from the gene database according to the elements in the small sample branch of the genome;

[0043] S4. Construct differential equation models of gene regulatory networks in each branch

[0044] S5. Summarize the gene regulatory networks of each branch to obtain a complete gene regulatory network.

[0045] The specific implementation process of the method is:

[0046] S1, genome sequencing data g i Prior knowledge of upstream and downstream genes ij Perform data fusion to obtain fused data c i ;

[0047] S2. Divide the genome fusion data into small samples: Use an unsupervised deep learning network to divide the genome fusion data into small samples i Perform clustering and output the gene sample branch sequence d to which each gene belongs i , each gene is divided into different gene small sample branches D according to formula (2) k ;

[0048] M(c i )=d,d=1,2,…,n (1)

[0049] Where M represents the unsupervised deep learning network; n represents the number of genome sample branches;

[0050] D k ={c i |M(c i )=k},k=1,2,…,m (2)

[0051] In the formula, m represents the number of small sample branches of genes;

[0052] S3, branch D based on small genome samples k The elements in extract the corresponding RNA-seq data set R from the gene database k ;

[0053] S4. Construct each gene regulatory network branch G k Differential equation model: Gene regulation network branch G is constructed by formula (3) k , where the coefficient a j Greater than zero means gene g i Gene g j Positive regulation, a j Equal to zero means gene g i Not affected by gene jRegulation, a j Less than zero means gene g i Gene g j Negative regulation;

[0054]

[0055] Where r i Indicates gene g i Corresponding RNA-seq data, n k Indicates the genome small sample branch D k The number of genes in a j For gene g j For gene g i The control factor of , b is a constant;

[0056] S5. Summarize the gene regulatory network branches to obtain the gene regulatory network G:

[0057] G={G1,G2,…,G k} (4).

[0058] The data fusion method described in step S1 comprises the following steps:

[0059] S11. Extract genome sequencing data through feature extraction network to obtain feature vector f i :

[0060] F(g i ) = f i (5)

[0061] Where F(g i ) represents the feature extraction network;

[0062] S12, encoding the prior knowledge of upstream and downstream genes through the encoding network;

[0063] S13, performing data fusion on the genome sequencing feature vector obtained in step S11 and step S12 and the prior knowledge encoding, wherein the data fusion integrates the prior knowledge into the genome sequencing feature, so as to effectively improve the gene association information and improve the clustering accuracy.

[0064] The constant b is used to represent the bias, which is a parameter that the constructed network needs to learn.

[0065] The feature extraction network is: using a Resnet network for feature extraction, and the Resnet network is a Resnet50 network.

[0066] An application of a dimensionality reduction modeling method for a gene regulatory network is to use the dimensionality reduction modeling method for a gene regulatory network for large genome modeling. Large genomes usually refer to genomes with a size of more than 5Gb, mostly concentrated in gymnosperms, amphibians, and reptiles; with ultra-high repetitive sequences and a large number of heterozygous regions.

[0067] The genome sequencing data is extracted through the feature extraction network to obtain the feature vector; the prior knowledge of upstream and downstream genes is encoded through the encoding network.

[0068] The obtained genome sequencing feature vector is fused with the prior knowledge encoding. This data fusion method integrates prior knowledge into genome sequencing features, which can effectively improve gene association information and clustering accuracy.

[0069] The example of the present invention first obtains gene sequencing data and upstream and downstream gene prior knowledge from the gene database, and after data preprocessing through the feature extraction network and the encoder, the obtained genome feature vector and the upstream and downstream gene prior knowledge one-hot encoding are fused. The fused data is clustered through the clustering network to cluster the genome to divide the genome into small samples and obtain gene small sample branches. After that, RNA-seq data is obtained from the gene database, RNA-seq is divided according to the gene small sample division results, and the gene regulatory network differential equation model of the gene branch is constructed respectively through different RNA-seq branch data, and finally the gene regulatory network of each branch is summarized to form a complete gene regulatory network of the genome.

[0070] The present invention has been repeatedly verified to effectively solve the contradiction between accuracy and computational complexity in the modeling process, and achieves an exponential reduction in computational complexity while ensuring modeling accuracy. Compared under the same conditions, the modeling accuracy of the method of the present invention is improved by at least 5%.

[0071] The dimensionality reduction modeling method (algorithm) of a gene regulatory network proposed in the present invention is the underlying technical core of the present invention, and various products can be derived based on the algorithm.

[0072] Based on the method (algorithm) proposed in the present invention, a dimensionality reduction modeling system for a gene regulatory network is developed using a programming language. The system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned dimensionality reduction modeling method for a gene regulatory network during operation.

[0073] The computer program of the developed system (software) is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-mentioned method for dimensionality reduction modeling of a gene regulatory network when called by a processor, that is, the present invention is materialized on a carrier to become a computer program product.

[0074] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0075] The computer programs (also referred to as programs, software, software applications, or codes) of the present invention include machine instructions for programmable processors, and these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or device (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0076] To sum up, the above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A dimensionality reduction modeling method for a gene regulatory network, characterized in that: The method comprises the following steps: S1, fusing genome sequencing data with prior knowledge of upstream and downstream genes to obtain fused data; S2. Divide the genome fusion data into small genome samples: cluster the genome fusion data through an unsupervised deep learning network, output the gene sample branch sequence to which each gene belongs, and divide each gene into different gene small sample branches; S3, extracting the corresponding RNA-seq data set from the gene database according to the elements in the small sample branch of the genome; S4, construct differential equation models of gene regulatory networks in each branch; S5. Summarize the gene regulatory networks of each branch to obtain a complete gene regulatory network.

2. A dimensionality reduction modeling method for a gene regulatory network according to claim 1, characterized in that: The specific implementation process of the method is: S1, genome sequencing data g i Prior knowledge of upstream and downstream genes ij Perform data fusion to obtain fused data c i ; S2. Divide the genome fusion data into small samples: Use an unsupervised deep learning network to divide the genome fusion data into small samples i Perform clustering and output the gene sample branch sequence d to which each gene belongs i , each gene is divided into different gene small sample branches D according to formula (2) k ; M(c i )=d,d=1,2,…,n (1) Where M represents the unsupervised deep learning network; n represents the number of genome sample branches; D k ={c i |M(c i )=k},k=1,2,…,m (2) In the formula, m represents the number of small sample branches of genes; S3, branch D based on small genome samples k The elements in extract the corresponding RNA-seq data set R from the gene database k ; S4. Construct each gene regulatory network branch G k Differential equation model: Gene regulatory network branch G is constructed by formula (3) k , where the coefficient a j Greater than zero means gene g i Gene g j Positive regulation, a j Equal to zero means gene g i Not affected by gene j Regulation, a j Less than zero means gene g i Gene g j Negative regulation; Where r i Indicates gene g i Corresponding RNA-seq data, n k Indicates the genome small sample branch D k The number of genes in a j For gene g j For gene g i The control factor of , b is a constant; S5. Summarize the gene regulatory network branches to obtain the gene regulatory network G: G={G1,G2,…,G k } (4)。 3. A dimensionality reduction modeling method for a gene regulatory network as claimed in claim 1 or 2, characterized in that: The data fusion method described in step S1 comprises the following steps: S11. Extract genome sequencing data through feature extraction network to obtain feature vector f i : F(g i )=f i (5) Where F(g i ) represents the feature extraction network; S12, encoding the prior knowledge of upstream and downstream genes through the encoding network; S13, performing data fusion on the genome sequencing feature vector obtained in step S11 and step S12 and the prior knowledge encoding, wherein the data fusion integrates the prior knowledge into the genome sequencing feature, so as to effectively improve the gene association information and improve the clustering accuracy.

4. The dimensionality reduction modeling method of a gene regulatory network according to claim 2, characterized in that: The constant b is used to represent the bias, which is a parameter that the constructed network needs to learn.

5. The dimensionality reduction modeling method of a gene regulatory network according to claim 3, characterized in that: The feature extraction network is: using a Resnet network for feature extraction, and the Resnet network is a Resnet50 network.

6. A dimensionality reduction modeling method for a gene regulatory network according to claim 5, characterized in that: The network used for clustering is DAEGC, SDCN or AGC or a fusion of the three; the genome sequencing data is extracted through the feature extraction network to obtain a feature vector; the upstream and downstream gene prior knowledge is encoded through the encoding network; the obtained genome sequencing feature vector is fused with the prior knowledge encoding. This data fusion method integrates prior knowledge into genome sequencing features to improve gene association information and clustering accuracy.

7. A dimensionality reduction modeling system for gene regulatory networks, characterized in that: The system has a program module corresponding to the steps of any one of claims 1 to 6, and executes the steps in the above-mentioned dimensionality reduction modeling method for a gene regulatory network when running.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of a dimensionality reduction modeling method for a gene regulatory network according to any one of claims 1 to 6 when called by a processor.

9. An application of the dimensionality reduction modeling method for a gene regulatory network according to claim 1, characterized in that: The dimensionality reduction modeling method of the gene regulatory network is used for large genome modeling.

10. The application of the dimensionality reduction modeling method of a gene regulatory network according to claim 9, characterized in that: Large genomes usually refer to genomes larger than 5Gb, which are mostly found in gymnosperms, amphibians, and reptiles. They have extremely high repetitive sequences and a large number of heterozygous regions.

Images