Data analysis system and method for gene regulatory network based on deep regression algorithm

Abstract

The application is suitable for the technical field of data analysis, and provides a data analysis system and method for gene regulatory network based on deep regression algorithm, the method comprises the following steps: converting gene identity and gene expression value into vector representation; calculating attention score between gene vector representations as a judgment standard for the relationship between genes; and predicting specific gene expression value. The application enhances the anti-noise performance by introducing a Gaussian layer, captures the complex regulatory relationship between genes by using embedding and attention mechanism, and realizes individualized GRN prediction at the single cell level, which can overcome the defects of the existing method in the personalized treatment of cancer patients, and further support the gene expression analysis and prediction in the personalized treatment of cancer patients.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data analysis, and particularly relates to a gene regulatory network data analysis system and method based on a deep regression algorithm. BACKGROUND

[0002] Gene expression in cancer patients plays a crucial role in predicting disease progression and guiding drug selection. By analyzing changes in gene expression, the function and role of specific genes in disease can be revealed. Single-cell RNA sequencing (scRNA-seq) technology provides detection of gene expression in individual cells, reflecting individual differences and intratumoral heterogeneity, providing detailed cell-level information for personalized analysis. Gene regulatory network (GRN) aims to reveal the regulatory relationship between transcription factors and target genes. Through accurate GRN prediction, the mechanism of gene expression changes can be better understood, providing stronger decision support for personalized cancer treatment.

[0003] However, current gene expression analysis and prediction in personalized treatment of cancer patients, especially modeling based on single-cell RNA sequencing (scRNA-seq) data, still faces many technical difficulties. Traditional data processing and analysis methods are easily affected by noise when processing single-cell data, lack accurate capture of complex gene regulatory networks (GRN), and are difficult to support causal inference. For example, most existing analysis methods can only process static data and cannot achieve dynamic modeling of single-cell level gene expression, thus limiting the accuracy of the model in predicting changes in specific gene expression in cancer patients. In addition, in the presence of a large amount of noise, the performance of traditional analysis methods decreases significantly, leading to incorrect identification of gene regulatory relationships.

[0004] Existing single-cell gene expression data analysis methods (such as Gaussian mixture model, differential expression analysis, etc.) can handle a certain degree of noise data, but lack the ability to adapt to cancer patient heterogeneity and individual specificity. These methods often fail to filter non-biological noise in the data when faced with personalized treatment data, resulting in unstable model inference results and difficulty in identifying complex gene regulatory relationships within tumors. In addition, although many GRN inference methods can simulate some relationships between transcription factors and target genes, they often fail to model the non-linear and multi-dimensional relationships between genes, making it difficult to achieve accurate gene prediction and regulatory network reconstruction at the single-cell level.

[0005] While existing single-cell gene expression simulation tools and GRN inference tools take into account gene regulatory relationships to some extent, these methods lack causal structure support and are prone to distortion in noisy data. Especially in cancer research applications, common tumor samples have large individual differences and high noise intensity, making it difficult for existing tools to support accurate GRN modeling for individual cancer patients, resulting in unreliable results. Summary of the Invention

[0006] The purpose of this invention is to provide a data analysis method for gene regulatory networks based on deep regression algorithms, aiming to solve the problems mentioned in the background art.

[0007] The present invention is implemented as follows: a data analysis method for gene regulatory networks based on deep regression algorithms includes the following steps:

[0008] Convert gene identity and gene expression level numerical values ​​into vector representations;

[0009] Calculate the attention score between gene vector representations as a criterion for evaluating the relationship between genes;

[0010] Predicting specific gene expression levels.

[0011] Preferably, the step of converting gene identity and gene expression level numerical values ​​into vector representations specifically includes:

[0012] Encoding genetic identity;

[0013] Gene expression levels were numerically encoded using a Gaussian distribution;

[0014] The two encoded parts are added together to obtain the final vector representation.

[0015] Preferably, the step of encoding gene expression levels using a Gaussian distribution specifically involves: for a single gene expression level data g, encoding the distance using a set of uniformly distributed Gaussian functions G, each Gaussian function having its own mean. Standard deviation :

[0016] ;

[0017] A single numerical value g will be represented as: .

[0018] Preferably, the step of encoding gene expression levels using a Gaussian distribution further includes: using a separate vector to represent additional considerations for encoding missing values.

[0019] Preferably, the step of calculating the attention score between the gene vector representations is used as a criterion for the relationship between genes, specifically: for each gene, the attention score between the gene and other genes is calculated, after all genes are calculated, the attention scores are sorted from high to low, and it is considered that the genes corresponding to the first n attention scores and the genes have a gene regulation relationship.

[0020] Another object of the embodiment of the present application is to provide a data analysis system for gene regulatory network based on deep regression algorithm, which is used to implement the data analysis method for gene regulatory network based on deep regression algorithm.

[0021] The encoding layer module is used to convert the gene identity and gene expression value into vector representation.

[0022] The self-attention layer module is used to calculate the attention score between the gene vector representations as a criterion for the relationship between genes.

[0023] The prediction layer module is used to predict the specific gene expression.

[0024] Preferably, the encoding layer module comprises:

[0025] The gene identity encoding layer unit is used to encode the gene identity.

[0026] The expression value Gaussian encoding layer unit is used to encode the gene expression value using Gaussian distribution.

[0027] The adding unit is used to add the two parts of the encoding to obtain the final vector representation.

[0028] Preferably, the parameters of the self-attention layer module are: the number of self-attention heads is 8, the number of self-attention layers is 1, and the vector encoding length of the self-attention layer is 64.

[0029] Preferably, it further comprises a model training module, which is used to randomly divide the samples into a training set, a validation set and a test set, the training set is responsible for training the model and learning the potential gene regulatory network GRN, the validation set is responsible for preventing the model from overfitting and is used to select the optimal model, and the test set is responsible for testing the performance of the model in gene expression prediction.

[0030] The data analysis method for gene regulatory network based on deep regression algorithm provided by the embodiment of the present application enhances the anti-noise performance by introducing a Gaussian layer, captures the complex regulation relationship between genes by using embedding and attention mechanism, and realizes individualized GRN prediction at the single cell level, which can overcome the defects of the existing method in personalized treatment of cancer patients, and further support the gene expression analysis and prediction in personalized treatment of cancer patients. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 A flow chart of a data analysis method of a gene regulatory network based on a deep regression algorithm is provided for an embodiment of the present application.

[0032] Figure 2 A block diagram of a data analysis system of a gene regulatory network based on a deep regression algorithm is provided for an embodiment of the present application.

[0033] Figure 3 A block diagram of an encoding layer module and an encoding process diagram are provided for an embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0035] The specific implementation of the present application is described in detail below in combination with specific embodiments.

[0036] Embodiment 1, a data analysis method of a gene regulatory network based on a deep regression algorithm, as shown in Figure 1 , a block diagram of its analysis system is shown in Figure 1 , including the following steps:

[0037] First, the data will pass through the encoding layer module of the system to convert the identity of the gene and the numerical value of the gene expression into a vector representation;

[0038] The vector representation is input into the self-attention layer module, the model implementation of the self-attention layer is the same as the implementation of BERT (Bidirectional Encoder Representations from Transformers), which will consider the context information, set the self-attention head in the self-attention layer module parameters to 8, the number of self-attention layers to 1, and the vector encoding length of the self-attention layer to 64;

[0039] The output of the self-attention layer module is input into the prediction layer module, and the specific gene expression can be predicted.

[0040] The encoding layer module, as shown in Figure 2 , its encoding process is as follows:

[0041] Encode the gene identity;

[0042] A plurality of Gaussian distributions are used to encode the gene expression numerical value, and specifically for a single gene expression numerical value g, a set of uniformly distributed Gaussian basis functions G are used to encode the distance, each Gaussian basis function has its own mean , standard deviation : , the last single value g, will be represented as , while using a separate vector representation for the encoding of missing values, which is a learnable parameter;

[0043] After encoding the gene identity and the gene expression, the two parts are added together to get the final vector representation as the input of the self-attention layer.

[0044] In addition, the data analysis system also includes a model training module. Specifically, in the training process, a single sample is randomly set as a missing value with a probability of 50%, so that the model can predict the missing value during the training process, and the model can reduce the error of predicting the missing value.

[0045] The original analysis data N*M gene expression matrix can be randomly divided into 80%, 10%, and 10% as the training set, the validation set and the test set, respectively. The training set is responsible for training the model and learning the underlying gene regulatory network GRN. The validation set is responsible for preventing the model from overfitting and is used to select the optimal model. The test set is responsible for testing the performance of the model in gene expression prediction.

[0046] The parameters of this module are: set the training to 100 rounds, the number of samples in each iteration in each round is 32, the learning rate is 0.001, and the warm-up iteration number of the learning rate is set to 5000.

[0047] Embodiment 2, analysis process of the data analysis system of the gene regulatory network based on the deep regression algorithm:

[0048] For the gene x of interest, 50 samples are randomly extracted from the training data;

[0049] Using the trained optimal model, the gene expression values in the 50 samples are predicted. The self-attention scores of the gene x to other genes in the self-attention layer of the model during the prediction process represent the regulatory relationship of other genes to the gene x. The larger the value, the stronger the regulatory relationship.

[0050] After sorting the self-attention scores, the top 100 genes are selected as the regulatory genes of the gene, that is, the gene regulatory network of the sample is constructed.

[0051] Example 3, analysis is carried out in three single cell data GSD, HSC, VSC constructed in the public data set BEELINE, the three single cell data GSD, HSC, VSC have 20000 sample data respectively, 18000 sample data are randomly divided as training data, 1000 data are as verification data, 1000 data are as test data, respectively have 79 genes, 30 genes, 15 genes, all provide verified gene regulation relationship, AUROC, AUPR can be used to judge the accuracy of gene regulation relationship prediction, MSE is used to judge the prediction error of gene expression, the data analysis system provided in example 1 and GENIE3, GRNBoost2, scGeneRAI are compared in the construction of gene relationship, and scGeneRAI is compared in the prediction of gene expression, and the results are shown in table 1 and table 2, as can be seen from the following table, the model can capture the complex regulation relationship between genes, and realize individualized gene expression prediction at single cell level:

[0052]

[0053]

[0054] In summary, the data analysis method of gene regulation network based on deep regression algorithm provided by the embodiment of the application introduces a Gaussian layer to enhance the noise processing capability of the system, and through the combination of gene embedding layer and attention mechanism, the complex interaction relationship between genes is captured, and more accurate GRN inference effect is achieved, the individual-specific GRN can be inferred based on single cell data, personalized treatment is supported, and the method has high universality, can adapt to different types of cancer samples, and thus has wide application potential in precision medicine.

[0055] The above only describes the preferred embodiments of the application and is not intended to limit the application, and any modification, equivalent replacement and improvement made within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A data analysis method for gene regulatory networks based on deep regression algorithm, characterized in that, Includes the following steps: Convert gene identity and gene expression level numerical values ​​into vector representations; Calculate the attention score between gene vector representations as a criterion for evaluating the relationship between genes; Predicting specific gene expression levels; The step of converting gene identity and gene expression level numerical values ​​into vector representations specifically includes: Encoding genetic identity; Gene expression levels were encoded using a Gaussian distribution; The two encoded parts are added together to obtain the final vector representation; The step of encoding gene expression levels using a Gaussian distribution specifically involves: for a single gene expression level data g, encoding the distance using a set of uniformly distributed Gaussian functions G, each Gaussian function having its own mean. Standard deviation : ； A single numerical value g will be represented as: ; The step of encoding gene expression levels using a Gaussian distribution further includes: using a separate vector to represent additional considerations for encoding missing values; The step of calculating the attention score between gene vector representations, as a criterion for evaluating the relationship between genes, specifically includes: For each gene, the attention score between that gene and other genes is calculated. After all genes have been calculated, these attention scores are sorted from high to low, and the genes corresponding to the top n attention scores are considered to have a gene regulatory relationship.

2. A data analysis system for gene regulatory networks based on deep regression algorithm, used to implement the data analysis method for gene regulatory networks based on deep regression algorithm as described in claim 1, characterized in that, include: The coding layer module is used to convert gene identity and gene expression level numerical values ​​into vector representations; The self-attention layer module is used to calculate the attention score between gene vector representations, which serves as a criterion for evaluating the relationship between genes. The prediction layer module is used to predict the expression levels of specific genes.

3. The data analysis system for gene regulatory networks based on deep regression algorithm according to claim 2, characterized in that, The coding layer module includes: Genetic identity coding layer unit, used to encode genetic identity; Expression level numerical Gaussian coding layer unit, used to encode gene expression level values ​​using a Gaussian distribution; The addition unit is used to add the two encoded parts together to obtain the final vector representation.

4. The data analysis system for gene regulatory networks based on deep regression algorithm according to claim 2, characterized in that, The parameters of the self-attention layer module are: 8 self-attention heads, 1 self-attention layer, and 64 vector encoding length of the self-attention layer.

5. The data analysis system for gene regulatory networks based on deep regression algorithm according to claim 3, characterized in that, Also includes: The model training module is used to randomly divide the samples into training set, validation set and test set. The training set is responsible for training the model and learning the underlying gene regulatory network GRN. The validation set is responsible for preventing the model from overfitting and is used to select the optimal model. The test set is responsible for testing the model's performance in gene expression prediction.

Images