A gene regulatory network inference method and system based on ensemble learning

Abstract

The present invention provides a gene regulatory network inference method and system based on ensemble learning, the method comprising: obtaining a gene data set and preprocessing it; constructing a gene regulatory network inference model; training the gene regulatory network inference model using the preprocessed gene data set; establishing a user interaction interface, obtaining gene expression data to be inferred and inputting the trained gene regulatory network inference model to obtain the inferred gene regulatory network; the present invention integrates a plurality of deep learning methods for gene regulatory network inference and integrates the above methods using a stacking ensemble learning strategy, thereby comprehensively combining the advantages of each model, reducing the deviation that may be caused by a single model, and improving the accuracy and robustness of gene regulatory network inference; secondly, the present invention is equipped with an intuitive and friendly user interaction interface, which simplifies the complex data processing and model training configuration process, and greatly reduces the usage threshold.

Description

Technical Field

[0001] The present invention relates to the technical fields of bioinformatics and computational biology, and more specifically, to a gene regulatory network inference method and system based on ensemble learning. Background Art

[0002] The reconstruction of gene regulatory networks (GRNs) is of great significance for studying cellular life activities and treating cancer. GRN inference refers to the inference of the regulatory relationship between genes and transcription factors through the analysis and modeling of gene and transcription factor expression data. Currently, common GRN inference methods include machine learning methods such as linear regression, Bayesian networks, and random forests. However, these methods have many shortcomings compared to deep learning methods, such as weak nonlinear relationship modeling capabilities and feature selection and extraction relying on domain knowledge. Deep learning models can learn more complex nonlinear relationships through deep neural networks and accurately infer the regulatory relationship between genes and transcription factors. Therefore, single-cell RNA sequence gene regulation inference methods based on deep learning have broad prospects.

[0003] In recent years, many deep learning-based methods have been developed for inferring gene regulatory networks. For example, the prior art discloses a gene regulatory network reconstruction method, system, and device. This method predicts and reconstructs gene regulatory networks through AI algorithms, wherein AI algorithms include BP neural networks, deep neural networks, convolutional neural networks, extreme learning machines, recurrent neural networks, and Transformers. However, most solutions, including this prior art, are based on a single neural network model for independent prediction, with poor prediction accuracy and robustness. The inference effect is not good in complex situations, and the inference effect fluctuates greatly when faced with gene expression data with different data distributions. In practical applications, it is often difficult to achieve ideal inference accuracy, and the operation is complicated, which is not conducive to use by non-professional users. Summary of the Invention

[0004] In order to overcome the defects of the above-mentioned existing technologies in inferring gene regulatory networks, such as poor inference effect, complex operation and difficulty in use, the present invention provides a gene regulatory network inference method and system based on ensemble learning. By integrating multiple classic deep learning models and utilizing the strategy of stacking ensemble learning, the inference accuracy of gene regulatory networks is effectively improved. At the same time, an interactive and friendly user interface is provided to make the inference of gene regulatory networks more efficient, accurate and easy to use, thereby promoting the research on gene regulatory networks and their application development in fields such as cancer treatment.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] A gene regulatory network inference method based on ensemble learning includes the following steps:

[0007] S1: Obtaining gene expression data of several groups of known gene regulatory networks to form a gene dataset, and preprocessing the gene dataset;

[0008] S2: Constructing a gene regulatory network inference model, wherein the structure of the gene regulatory network inference model includes: an input layer, a plurality of parallel base learners, a feature fusion layer, a meta-learner, and an output layer;

[0009] The input layer is used to input gene expression data into each base learner; each base learner preliminarily predicts the gene regulatory network corresponding to the gene expression data and inputs its prediction results into the feature fusion layer; the feature fusion layer is used to fuse the prediction results of all base learners, obtain fusion features and input them into the meta learner; the meta learner is used to perform integrated prediction based on the fusion features, obtain integrated prediction results and output them through the output layer to obtain the final gene regulatory network;

[0010] S3: Using the preprocessed gene dataset to train the gene regulatory network inference model to obtain a trained gene regulatory network inference model;

[0011] S4: Establishing a user interaction interface, obtaining the gene expression data to be inferred and using the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred.

[0012] Base learners are independent deep learning models that are responsible for capturing patterns or regularities in the data. The output of the base learners is used as input for subsequent steps to build higher-level prediction models.

[0013] The meta learner is a higher-level model in stacking ensemble learning, which is used to further integrate the output of the base learners and predict the final results. Its goal is to make full use of the distribution and characteristics of the data learned by each base learner and obtain better prediction performance under different data distributions.

[0014] Preferably, in step S1, the preprocessing includes:

[0015] Converting the gene expression data into a matrix form and standardizing the data, wherein rows in the matrix represent genes, columns represent cells, and elements in the matrix represent standardized gene expression values;

[0016] The gene regulatory network corresponding to the gene expression data is converted into an adjacency matrix, in which a known gene regulatory relationship is labeled as "1" and an unknown gene regulatory relationship is labeled as "0".

[0017] Preferably, in step S1, the gene dataset is an open source dataset in the Gene Expression Omnibus database, specifically any one of a mouse erythroid hematopoietic stem cell dataset mHSC-E, a mouse granulocyte hematopoietic stem cell dataset mHSC-GM, a mouse lymphoid hematopoietic stem cell dataset mHSC-L, a mouse dendritic cell dataset mDC, a human embryonic stem cell dataset hESC, and a human hepatocyte dataset hHEP.

[0018] Preferably, the step S2 includes at least the following five models: DeepDRIM model, CNNC model, STGRNS model, GENELink model and GNNLink model.

[0019] Preferably, in step S2, the feature fusion layer performs a splicing operation on the prediction results of all base learners, regards the spliced ​​vector as a new feature and performs a normalization operation to obtain the fused feature.

[0020] Preferably, in step S2, the meta-learner is specifically a multi-layer perceptron MLP;

[0021] The structure of the multi-layer perceptron MLP includes: a sub-input layer, a first hidden layer, a second hidden layer, a third hidden layer and a sub-output layer connected in sequence; the first hidden layer and the second hidden layer are used to sequentially perform feature dimensionality upgrade on the fusion features input by the sub-input layer, the third hidden layer is used for feature dimensionality reduction, and the sub-output layer is used to convert the fusion features after dimensionality reduction into prediction scores to obtain the integrated prediction results.

[0022] The advantage of using a multilayer perceptron as a meta-learner is that it can adaptively learn the importance of each base learner and assign different weights to the base learners, thereby effectively alleviating the prediction bias caused by different data distributions and improving the overall prediction performance.

[0023] Preferably, in step S2, the output layer performs binarization processing on the integrated prediction result through a preset threshold to obtain the final gene regulatory network.

[0024] Preferably, in step S3, the preprocessed gene data set is used to train the gene regulatory network inference model. During the training process, the prediction performance of the gene regulatory network inference model is improved by optimizing a preset loss function, and an early stopping method or a learning rate adjustment strategy is used to prevent the model from overfitting.

[0025] Preferably, the preset loss function is specifically a binary cross entropy loss function.

[0026] The present invention also provides a gene regulatory network inference system based on ensemble learning, which applies the above-mentioned gene regulatory network inference method based on ensemble learning, including:

[0027] Preprocessing unit: used to obtain gene expression data of several groups of known gene regulatory networks to form a gene data set, and preprocess the gene data set;

[0028] Model building unit: used to build gene regulatory network inference models;

[0029] Model training unit: used to train the gene regulatory network inference model using the preprocessed gene data set to obtain a trained gene regulatory network inference model;

[0030] Human-computer interaction unit: used to establish a user interaction interface, obtain the gene expression data to be inferred and use the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred.

[0031] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0032] The present invention provides a gene regulatory network inference method and system based on ensemble learning. First, gene expression data of several groups of known gene regulatory networks are obtained to form a gene data set, and the gene data set is preprocessed. Then, a gene regulatory network inference model is constructed. The structure of the gene regulatory network inference model includes: an input layer, several parallel base learners, a feature fusion layer, a meta-learner, and an output layer. Then, the gene regulatory network inference model is trained using the preprocessed gene data set to obtain a trained gene regulatory network inference model. Finally, a user interaction interface is established to obtain the gene expression data to be inferred and the trained gene regulatory network inference model is input into the user interaction interface to obtain the gene regulatory network corresponding to the gene expression data to be inferred.

[0033] By integrating multiple classic neural networks and adopting an ensemble learning strategy, the present invention can combine the advantages of each model, effectively reduce the deviation that may be caused by a single model, significantly improve the accuracy and robustness of gene regulatory network inference, and at the same time adapt to different types of gene expression data and research needs; secondly, the present invention is also equipped with an intuitive and friendly user interaction interface, which simplifies the complex data processing and model configuration process, allowing users to easily operate without having a deep programming or bioinformatics background, greatly lowering the usage threshold; in addition, accurate gene regulatory network inference is crucial for understanding the mechanism of cell life activities, the occurrence and development of diseases, and the search for therapeutic targets; the promotion and use of the present invention will accelerate the research progress in these fields, especially in the development of cancer treatment strategies, and is expected to provide strong support for precision medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of a gene regulatory network inference method based on ensemble learning provided in Example 1.

[0035] Figure 2 This is an architecture diagram of a gene regulatory network inference method based on ensemble learning provided in Example 2.

[0036] Figure 3 This is a structural diagram of a gene regulatory network inference system based on ensemble learning provided in Example 3.

[0037] Figure 4 This is a schematic diagram of a gene regulatory network inference system based on ensemble learning provided in Example 4.

[0038] Figure 5 This is an architecture diagram of a gene regulatory network inference system based on ensemble learning provided in Example 4.

[0039] Figure 6 This is a schematic diagram of the system login interface provided in Example 4.

[0040] Figure 7 This is a schematic diagram of the system main interface provided in Example 4.

[0041] Figure 8 This is a schematic diagram of the system introduction page provided in Example 4.

[0042] Figure 9 This is a schematic diagram of the model evaluation interface provided in Example 4.

[0043] Figure 10 This is a schematic diagram of the gene regulatory network inference interface provided in Example 4. DETAILED DESCRIPTION

[0044] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0045] In order to better illustrate this embodiment, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product size;

[0046] It is understandable to those skilled in the art that some well-known structures and descriptions thereof may be omitted in the drawings.

[0047] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0048] Example 1

[0049] like Figure 1 As shown, this embodiment provides a gene regulatory network inference method based on ensemble learning, comprising the following steps:

[0050] S1: Obtaining gene expression data of several groups of known gene regulatory networks to form a gene dataset, and preprocessing the gene dataset;

[0051] S2: Constructing a gene regulatory network inference model, wherein the structure of the gene regulatory network inference model includes: an input layer, a plurality of parallel base learners, a feature fusion layer, a meta-learner, and an output layer;

[0052] The input layer is used to input gene expression data into each base learner; each base learner preliminarily predicts the gene regulatory network corresponding to the gene expression data and inputs its prediction results into the feature fusion layer; the feature fusion layer is used to fuse the prediction results of all base learners, obtain fusion features and input them into the meta learner; the meta learner is used to perform integrated prediction based on the fusion features, obtain integrated prediction results and output them through the output layer to obtain the final gene regulatory network;

[0053] S3: Using the preprocessed gene dataset to train the gene regulatory network inference model to obtain a trained gene regulatory network inference model;

[0054] S4: Establishing a user interaction interface, obtaining the gene expression data to be inferred and using the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred.

[0055] In the specific implementation process, firstly, gene expression data of several groups of known gene regulatory networks are obtained to form a gene data set, and the gene data set is preprocessed;

[0056] Next, we build a gene regulatory network inference model. The structure of the gene regulatory network inference model includes: an input layer, several parallel base learners, a feature fusion layer, a meta-learner, and an output layer.

[0057] Then, the preprocessed gene dataset is used to train the gene regulatory network inference model to obtain the trained gene regulatory network inference model;

[0058] Finally, a user interaction interface is established to obtain the gene expression data to be inferred and the trained gene regulatory network inference model is input into the user interaction interface to obtain the gene regulatory network corresponding to the gene expression data to be inferred;

[0059] This method integrates multiple classic neural network models and uses an ensemble learning strategy to effectively improve the inference accuracy of gene regulatory networks. At the same time, it is equipped with an interactive and friendly user interface to make the inference of gene regulatory networks more efficient, accurate and easy to use, thereby promoting the research on gene regulatory networks and their application in fields such as cancer treatment.

[0060] Example 2

[0061] like Figure 2 As shown, this embodiment provides a gene regulatory network inference method based on ensemble learning, comprising the following steps:

[0062] S1: Obtaining gene expression data of several groups of known gene regulatory networks to form a gene dataset, and preprocessing the gene dataset;

[0063] S2: Constructing a gene regulatory network inference model, wherein the structure of the gene regulatory network inference model includes: an input layer, five parallel base learners, a feature fusion layer, a meta-learner, and an output layer;

[0064] The input layer is used to input gene expression data into each base learner; each base learner preliminarily predicts the gene regulatory network corresponding to the gene expression data and inputs its prediction results into the feature fusion layer; the feature fusion layer is used to fuse the prediction results of all base learners, obtain fusion features and input them into the meta learner; the meta learner is used to perform integrated prediction based on the fusion features, obtain integrated prediction results and output them through the output layer to obtain the final gene regulatory network;

[0065] S3: Using the preprocessed gene dataset to train the gene regulatory network inference model to obtain a trained gene regulatory network inference model;

[0066] S4: establishing a user interaction interface, obtaining the gene expression data to be inferred, and using the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred;

[0067] In step S1, the pre-processing includes:

[0068] Converting the gene expression data into a matrix form and standardizing the data, wherein rows in the matrix represent genes, columns represent cells, and elements in the matrix represent standardized gene expression values;

[0069] Converting the gene regulatory network corresponding to the gene expression data into an adjacency matrix, wherein known gene regulatory relationships are labeled as "1" and unknown gene regulatory relationships are labeled as "0";

[0070] In step S1, the gene dataset is an open source dataset in the Gene Expression Omnibus database, specifically any one of the mouse erythroid hematopoietic stem cell dataset mHSC-E, the mouse granulocyte hematopoietic stem cell dataset mHSC-GM, the mouse lymphoid hematopoietic stem cell dataset mHSC-L, the mouse dendritic cell dataset mDC, the human embryonic stem cell dataset hESC, and the human hepatocyte dataset hHEP;

[0071] In step S2, the five parallel base learners include the following five types: DeepDRIM model, CNNC model, STGRNS model, GENELink model and GNNLink model;

[0072] In step S2, the feature fusion layer performs a splicing operation on the prediction results of all base learners, regards the spliced ​​vector as a new feature and performs a normalization operation to obtain the fused feature;

[0073] In step S2, the meta-learner is specifically a multi-layer perceptron MLP;

[0074] The structure of the multi-layer perceptron MLP includes: a sub-input layer, a first hidden layer, a second hidden layer, a third hidden layer, and a sub-output layer connected in sequence; the first hidden layer and the second hidden layer are used to sequentially perform feature dimension increase on the fusion features input by the sub-input layer, the third hidden layer is used for feature dimension reduction, and the sub-output layer is used to convert the fusion features after dimension reduction into prediction scores to obtain the integrated prediction result;

[0075] In step S2, the output layer performs binarization processing on the integrated prediction results using a preset threshold to obtain the final gene regulatory network;

[0076] In step S3, the preprocessed gene dataset is used to train the gene regulatory network inference model. During the training process, the prediction performance of the gene regulatory network inference model is improved by optimizing the preset loss function, and an early stopping method or a learning rate adjustment strategy is used to prevent the model from overfitting. In this embodiment, the preset loss function is specifically a binary cross entropy loss function.

[0077] In the specific implementation process, first, gene expression data of several groups of known gene regulatory networks are obtained to form a gene dataset, and the gene dataset is preprocessed; the gene dataset in this embodiment includes mHSC-E, mHSC-GM, mHSC-L, mDC, hESC and hHEP datasets;

[0078] The model's input data consists of two parts: gene expression data and some known gene regulatory networks. Gene expression data is a gene × cell matrix, characterized by high dimensionality, high noise, and nonlinearity. After preprocessing, the gene expression data exists in the form of a matrix, where rows represent genes and columns represent cells. Gene regulatory networks are usually in the form of an adjacency matrix, where known gene regulatory relationship labels are considered "1" and unknown ones are considered "0". Table 1 shows an example of gene expression data:

[0079] Table 1 Gene expression data examples

[0080]

[0081]

[0082] Next, we build a gene regulatory network inference model. The structure of the gene regulatory network inference model includes: an input layer, five parallel base learners, a feature fusion layer, a meta-learner, and an output layer.

[0083] In this embodiment, the five parallel base learners are: DeepDRIM model, CNNC model, STGRNS model, GENELink model and GNNLink model. The brief introduction of each model is as follows:

[0084] 1) CNNC: An algorithm based on convolutional neural networks (CNNs) that requires less computing resources and takes less time to train, and is suitable for simple single-cell RNA sequencing data;

[0085] 2) DeepDRIM: An algorithm based on convolutional neural networks (CNNs), which requires more computing resources and takes longer to train. It is suitable for complex single-cell RNA sequencing data.

[0086] 3) GENELink: An algorithm based on graph neural networks (GNNs) that requires less computational resources and takes less training time. It is suitable for single-cell RNA sequencing data with a certain scale of prior networks.

[0087] 4) GNNLink: An algorithm based on graph neural networks (GNNs) that requires less computational resources and takes less training time. It is suitable for single-cell RNA sequencing data with a certain scale of prior networks.

[0088] 5) STGRNS: A Transformer-based algorithm that requires fewer computational resources and takes less training time. It is suitable for time series data and most single-cell RNA sequencing data.

[0089] The preprocessed gene dataset is then used to train the gene regulatory network inference model. The input data is fed into multiple base learners, namely DeepDRIM, CNNC, STGRNS, GENELink, and GNNLink. The base learners are trained and their prediction results are obtained. The prediction results of the base learners are expressed as a series of scores that can reflect the regulatory relationships between genes.

[0090] The prediction results of the base learners are then treated as a series of features, which are further fused to form a unified feature vector. The feature vector is normalized to improve the stability and efficiency of model training. The fusion method is to splice the prediction results of each base learner and normalize them in the feature dimension to prevent large discrepancies between the prediction results of each base learner.

[0091] The meta-learner (denoted as DEEPGENE in this embodiment) is composed of a multi-layer perceptron (MLP), which contains five layers of neural networks: a sub-input layer, a first hidden layer, a second hidden layer, a third hidden layer, and a sub-output layer. The fused features are input into the MLP, where the first and second hidden layers increase the feature dimension to better capture deep features. The third hidden layer then performs dimensionality reduction, and the sub-output layer fully converts the features into prediction scores.

[0092] After the above steps, the integrated prediction results can be obtained. Finally, the output layer binarizes the prediction results through a specific threshold, that is, the final reconstructed gene regulatory network is obtained, realizing the inference of the gene regulatory network;

[0093] This example improves the prediction performance of the model by optimizing the loss function (binary cross entropy loss). During the training process, strategies such as early stopping and learning rate adjustment are used to prevent overfitting of the model.

[0094] After obtaining the trained gene regulatory network inference model, a user interaction interface is established to obtain the gene expression data to be inferred and input the trained gene regulatory network inference model into the user interaction interface to obtain the gene regulatory network corresponding to the gene expression data to be inferred;

[0095] Table 2 shows the comparison of AUROC scores of six methods on six different datasets. The bold indicates the best, the underline indicates the second best, and the numbers in brackets represent the specific rankings.

[0096] Table 2 Comparison of AUROC scores of six methods on six different datasets

[0097] mHSC-E mHSC-GM mHSC-L mDC hESC hHEP CNNC 0.890(5) 0.885(5) 0.762(6) 0.500(6) 0.727(6) 0.518(6) GNNLink 0.821(6) 0.861(6) 0.830(5) 0.690(5) 0.847(4) 0.817(4) GENELink 0.893(4) 0.907(4) 0.839(4) 0.752(4) 0.840(5) 0.823(3) STGRNS 0.928(2) 0.924(3) 0.842(3) 0.811(2) 0.863(3) 0.881(1) DeepDRIM 0.926(3) 0.937(1) 0.863(2) 0.814(1) 0.874(2) 0.545(5) DEEPGENE 0.935(1) 0.937(1) 0.874(1) 0.804(3) 0.877(1) 0.880(2)

[0098] Table 3 shows the comparison of AUPRC scores of six methods on six different datasets. The bold indicates the best, the underline indicates the second best, and the numbers in brackets represent the specific rankings.

[0099] Table 3 Comparison of AUPRC scores of six methods on six different datasets

[0100] mHSC-E mHSC-GM mHSC-L mDC hESC hHEP CNNC 0.896(5) 0.885(5) 0.722(6) 0.044(6) 0.391(6) 0.362(6) GNNLink 0.794(6) 0.852(6) 0.782(5) 0.137(4) 0.497(5) 0.664(4) GENELink 0.905(4) 0.916(4) 0.814(4) 0.123(5) 0.513(4) 0.684(2) STGRNS 0.942(2) 0.937(3) 0.824(3) 0.166(2) 0.551(3) 0.805(1) DeepDRIM 0.936(3) 0.949(1) 0.856(2) 0.166(2) 0.588(1) 0.397(5) DEEPGENE 0.947(1) 0.946(2) 0.864(1) 0.179(1) 0.586(2) 0.805(1)

[0101] According to the experimental results in Table 2 and Table 3, the integrated prediction method DEEPGENE provided in this embodiment has the best performance, and its prediction accuracy and robustness are the best;

[0102] This method integrates multiple classic neural network models and uses an ensemble learning strategy to effectively improve the inference accuracy of gene regulatory networks. At the same time, it is equipped with an interactive and friendly user interface to make the inference of gene regulatory networks more efficient, accurate and easy to use, thereby promoting the research on gene regulatory networks and their application in fields such as cancer treatment.

[0103] Example 3

[0104] like Figure 3 As shown, this embodiment provides a gene regulatory network inference system based on ensemble learning, applying a gene regulatory network inference method based on ensemble learning described in Example 1 or 2, including:

[0105] Preprocessing unit 301: used for acquiring gene expression data of several groups of known gene regulatory networks to form a gene dataset, and preprocessing the gene dataset;

[0106] Model building unit 302: used to build a gene regulatory network inference model;

[0107] Model training unit 303: used to train the gene regulatory network inference model using the preprocessed gene dataset to obtain a trained gene regulatory network inference model;

[0108] The human-computer interaction unit 304 is used to establish a user interaction interface, obtain the gene expression data to be inferred, and use the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred.

[0109] In a specific implementation process, first, the pre-processing unit 301 obtains gene expression data of several groups of known gene regulatory networks to form a gene data set, and pre-processes the gene data set;

[0110] Then the model construction unit 302 constructs a gene regulatory network inference model. The structure of the gene regulatory network inference model includes: an input layer, a plurality of parallel base learners, a feature fusion layer, a meta-learner and an output layer;

[0111] Then, the model training unit 303 uses the preprocessed gene data set to train the gene regulatory network inference model to obtain a trained gene regulatory network inference model;

[0112] Finally, the human-computer interaction unit 304 establishes a user interaction interface, obtains the gene expression data to be inferred, and uses the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred;

[0113] This system integrates multiple classic neural network models and uses ensemble learning strategies to effectively improve the inference accuracy of gene regulatory networks. At the same time, it is equipped with an interactive and friendly user interface to make the inference of gene regulatory networks more efficient, accurate and easy to use, thereby promoting the research on gene regulatory networks and their application in fields such as cancer treatment.

[0114] Example 4

[0115] like Figure 4 As shown, this embodiment provides a gene regulatory network inference system developed based on Python language, which includes a user graphical interface and core algorithm components.

[0116] In the specific implementation process, Figure 5 As shown in the architecture diagram, this system includes core algorithm components such as CNNC, DeepDRIM, STGRNS, GENELink, GNNLink and DEEPGENE integration, which together form the algorithmic basis for inferring gene regulatory networks. At the same time, it adopts a graphical user interface (GUI) design, including a login interface, a system introduction interface, a data import module, a parameter setting module and a result display module. Users can complete data import, model selection and parameter setting, and view inference results through simple click and drag operations.

[0117] The use of this system interface includes the following aspects:

[0118] 1) Data collation and import into the system: First, users need to collate and import gene expression data into the system; this step ensures that the data quality and format meet the requirements of subsequent analysis;

[0119] The example dataset used in this example was downloaded from the public dataset provided by BEELINE. The gene expression data was in .csv format, and the real gene regulatory network data used was in .csv format.

[0120] like Figure 6 The following is the login interface, where users can choose to log in as "algorithm researcher" or "medical staff mode";

[0121] 2) System mode selection and parameter setting: Figure 7 As shown, users can select functions on the main interface. Users can select the appropriate inference mode and set relevant parameters through the graphical user interface. This step allows users to customize the inference process according to specific research needs.

[0122] like Figure 8 As shown, users can view the introduction of tasks, models, and parameters on the "System Introduction" page;

[0123] 3) Model evaluation: The system evaluates the selected model to determine its performance on the current dataset; this step helps users select the best-performing model for research;

[0124] like Figure 9 As shown in the figure, the user first needs to select the path of the local file and enter the case and model to be evaluated; on the right side, you can customize the relevant parameters and tune the model yourself; click [generate&train] to start evaluating the selected model; the result display module will show the scores of each model to help users choose the optimal model for practical problems; Figure 9 The results shown show that the performance of DEEPGENE ensemble is on par with the best algorithm, with the best robustness and accuracy;

[0125] 4) Inferring regulatory networks: Based on the selected model and parameters, the system infers the gene regulatory network. This step is the core of the entire inference process and is directly related to the accuracy of the inference results.

[0126] like Figure 10 As shown, the user first needs to select the local file path and enter the case and model to be evaluated; the relevant parameters can be customized on the right, and different numbers of gene pairs can be displayed in the central display module; click [predict] to start inferring the regulatory network; the result display module will display the inference results and store the complete inferred CSV file in the user-selected path to help users make judgments in practical problems;

[0127] By integrating five classic gene regulatory network inference methods and utilizing an ensemble learning strategy, the toolkit of this embodiment can significantly improve the accuracy of inference. Users can select an appropriate inference method based on their specific research needs and use the model evaluation module to select the model with the best performance for research. In addition, the user-friendly graphical user interface design makes it easy for non-professional users to use the toolkit of this embodiment. Users can simply click and drag to complete data import, model selection and parameter setting, and view inference results, greatly simplifying the operation process.

[0128] This system has the following beneficial effects:

[0129] 1) Improved research efficiency: The toolkit of this embodiment greatly improves the efficiency of gene regulatory network inference through automated processing and ensemble learning methods. Users can complete data import, model selection, and parameter setting in a relatively short time and obtain accurate inference results, thereby saving research time and costs.

[0130] 2) Improving research quality: By integrating multiple classic inference methods and optimizing the model training process, the toolkit of this embodiment can provide more accurate and reliable gene regulatory network inference results; this helps researchers gain a deeper understanding of gene regulatory mechanisms and provides strong support for disease diagnosis and treatment;

[0131] 3) Promoting scientific collaboration: The toolkit of this embodiment is easy to use and expand, enabling convenient collaboration and communication with other scientific research teams. By sharing data and models, researchers can jointly promote the development of gene regulatory network research and make greater contributions to the field of life sciences.

[0132] The system of this embodiment can be widely used in hospitals and biological research laboratories. In hospitals, it can be used to predict pathogenic genes, prevent diseases, and design cancer treatment plans; in biological research laboratories, it can be used to make effective predictions about the regulatory relationships of genes, thereby deepening the understanding of life processes such as cell behavior and cell differentiation.

[0133] The same or similar reference numerals correspond to the same or similar components;

[0134] The terms used in the drawings to describe positional relationships are for illustrative purposes only and should not be construed as limiting this patent;

[0135] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A gene regulatory network inference method based on ensemble learning, characterized in that: The following steps are involved: S1: Obtaining gene expression data of several groups of known gene regulatory networks to form a gene dataset, and preprocessing the gene dataset; the preprocessing includes: Converting the gene expression data into a matrix form and standardizing the data, wherein rows in the matrix represent genes, columns represent cells, and elements in the matrix represent standardized gene expression values; Converting the gene regulatory network corresponding to the gene expression data into an adjacency matrix, wherein known gene regulatory relationships are labeled as "1" and unknown gene regulatory relationships are labeled as "0"; S2: Constructing a gene regulatory network inference model, wherein the structure of the gene regulatory network inference model includes: an input layer, a plurality of parallel base learners, a feature fusion layer, a meta-learner, and an output layer; The input layer is used to input gene expression data into each base learner; each base learner preliminarily predicts the gene regulatory network corresponding to the gene expression data and inputs its prediction results into the feature fusion layer; the feature fusion layer is used to fuse the prediction results of all base learners, obtain fusion features and input them into the meta learner; the meta learner is used to perform integrated prediction based on the fusion features, obtain integrated prediction results and output them through the output layer to obtain the final gene regulatory network; The plurality of parallel base learners include at least the following five: DeepDRIM model, CNNC model, STGRNS model, GENELink model and GNNLink model; The meta-learner is specifically a multi-layer perceptron MLP; The structure of the multi-layer perceptron MLP includes: a sub-input layer, a first hidden layer, a second hidden layer, a third hidden layer, and a sub-output layer connected in sequence; the first hidden layer and the second hidden layer are used to sequentially perform feature dimension increase on the fusion features input by the sub-input layer, the third hidden layer is used for feature dimension reduction, and the sub-output layer is used to convert the fusion features after dimension reduction into prediction scores to obtain the integrated prediction result; S3: Using the preprocessed gene dataset to train the gene regulatory network inference model to obtain a trained gene regulatory network inference model; S4: Establishing a user interaction interface, obtaining the gene expression data to be inferred and using the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred.

2. A gene regulatory network inference method based on ensemble learning according to claim 1, characterized in that: In step S1, the gene dataset is an open source dataset in the Gene Expression Omnibus database, specifically any one of the mouse erythroid hematopoietic stem cell dataset mHSC-E, the mouse granulocyte hematopoietic stem cell dataset mHSC-GM, the mouse lymphoid hematopoietic stem cell dataset mHSC-L, the mouse dendritic cell dataset mDC, the human embryonic stem cell dataset hESC, and the human hepatocyte dataset hHEP.

3. A gene regulatory network inference method based on ensemble learning according to claim 1, characterized in that: In step S2, the feature fusion layer concatenates the prediction results of all base learners, regards the concatenated vector as a new feature and performs a normalization operation to obtain the fused feature.

4. A gene regulatory network inference method based on ensemble learning according to claim 1, characterized in that: In step S2, the output layer performs binarization processing on the integrated prediction result through a preset threshold to obtain the final gene regulatory network.

5. The gene regulatory network inference method based on ensemble learning according to claim 1, characterized in that: In step S3, the preprocessed gene dataset is used to train the gene regulatory network inference model. During the training process, the prediction performance of the gene regulatory network inference model is improved by optimizing a preset loss function, and an early stopping method or a learning rate adjustment strategy is used to prevent the model from overfitting.

6. A gene regulatory network inference method based on ensemble learning according to claim 5, characterized in that: The preset loss function is specifically a binary cross entropy loss function.

7. A gene regulatory network inference system based on ensemble learning, applying a gene regulatory network inference method based on ensemble learning according to any one of claims 1 to 6, characterized in that: include: Preprocessing unit: used to obtain gene expression data of several groups of known gene regulatory networks to form a gene data set, and preprocess the gene data set; Model building unit: used to build gene regulatory network inference models; Model training unit: used to train the gene regulatory network inference model using the preprocessed gene data set to obtain a trained gene regulatory network inference model; Human-computer interaction unit: used to establish a user interaction interface, obtain the gene expression data to be inferred and use the user interaction interface to input the trained gene regulatory network inference model to obtain the gene regulatory network corresponding to the gene expression data to be inferred.

Images