Nasopharyngeal carcinoma differential expression gene automatic detection system based on genetic algorithm

An automated detection system for differentially expressed genes in nasopharyngeal carcinoma, combining genetic algorithms with multiple machine learning algorithms and SHAP values, solves the problems of low detection accuracy and insufficient interpretability in existing technologies, achieving efficient identification and reliable diagnosis of differentially expressed genes in nasopharyngeal carcinoma.

CN121075441APending Publication Date: 2025-12-05NORTHEASTERN UNIV CHINA
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202511282649.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-09
Publication Date
2025-12-05

AI Technical Summary

Technical Problem

Existing technologies have low accuracy and insufficient interpretability in the detection of differentially expressed genes in nasopharyngeal carcinoma, especially in cases of high noise, high-dimensional sparse data and insufficient sample size, making it difficult to effectively identify differentially expressed miRNA genes related to nasopharyngeal carcinoma.

Method used

An automated detection system for differentially expressed genes in nasopharyngeal carcinoma based on genetic algorithms was developed. This system integrates gene sequencing data acquisition, grouping, graph generation, data preprocessing, and differentially expressed gene detection modules with multiple machine learning algorithms, SHAP values, and a dynamic penalty mechanism to achieve stability and interpretability in gene detection.

Benefits of technology

It significantly improves the accuracy and interpretability of differentially expressed gene detection in nasopharyngeal carcinoma, providing a more scientific and intelligent detection tool and offering a reliable reference for early diagnosis and personalized treatment plans for nasopharyngeal carcinoma.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121075441A_ABST
    Figure CN121075441A_ABST
Patent Text Reader

Abstract

The invention discloses a nasopharyngeal carcinoma differential expression gene automatic detection system based on a genetic algorithm, and relates to the technical field of nasopharyngeal carcinoma early diagnosis. According to the method, various machine learning algorithms are integrated under a genetic algorithm framework, and the performance of each algorithm is comprehensively evaluated as a fitness function, so that the stability and generalization ability of the gene detection method are remarkably enhanced. And meanwhile, the SHAP value is innovatively introduced as an intermediate tool for differential expression gene detection, so that the gene detection process gives consideration to statistical significance and biological significance, and the interpretability of the result is improved. A dynamic punishment mechanism effectively controls the number of detected genes, and model conciseness is promoted. An external independent verification process and abundant visual analysis are also matched, so that the reliability and practicability of the result are ensured. The invention provides a more scientific and intelligent new tool for nasopharyngeal carcinoma differential expression gene detection, can provide reliable reference information for clinical diagnosis and treatment, and has important significance for early diagnosis of nasopharyngeal carcinoma.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of early diagnosis technology for nasopharyngeal carcinoma, and in particular to an automated detection system for differentially expressed genes in nasopharyngeal carcinoma based on a genetic algorithm. Background Technology

[0002] Nasopharyngeal carcinoma (NPC) is one of the most prevalent malignant tumors in East and Southeast Asia, particularly prevalent in southern my country, exhibiting distinct geographical and racial characteristics. The etiology of NPC is believed to be related to a complex interplay of genetics, viruses, environment, and dietary habits. High-risk individuals have a higher incidence of NPC compared to those without risk. Due to the mild early symptoms and lack of effective early diagnostic methods, it is difficult to detect localized primary tumors in the lesser curvatures of the nasal cavity. Therefore, detecting differentially expressed genes for early diagnosis and prognostic assessment of NPC can enhance existing diagnostic methods and guide personalized medical treatment plans.

[0003] In recent decades, with the further exploration of biomarkers for nasopharyngeal carcinoma (NPC), the importance of microRNAs in the early diagnosis of NPC has become increasingly prominent. MicroRNAs, also known as miRNAs, are small non-coding RNAs of about 22 nucleotides in length, whose main function is the fine-tuning of gene expression. Abnormal gene expression is closely related to the occurrence and progression of NPC. However, due to the development of sequencing technology, the complexity of NPC types, and the increasingly complex genomic data, the detection of differentially expressed genes has become extremely difficult. Therefore, there is an urgent need for a new automated system for the early diagnosis of NPC. Automated detection of differentially expressed genes related to NPC in miRNA data is of great significance for understanding the carcinogenicity, cancer progression, and treatment response of NPC.

[0004] The technical solution described in Chinese patent "CN202111272652.7 A method and application for determining differentially expressed genes using RNA-seq" proceeds as follows: First, the expression variance of each gene in two groups of samples is compared with its expected value to identify signal genes with variances higher than expected, and genes affected by noise are removed. Next, a t-test is used to calculate the p-value of the retained signal genes to screen out preliminary differentially expressed genes. Then, a partial least squares (PLS) model is used to analyze the overall expression difference characteristics between the two groups, and components related to the differences between groups are extracted, irrelevant noise is removed, and processed data is obtained. Finally, by comparing the residual distributions of non-differentially expressed genes in the processed data, if the residual of a certain gene is significantly higher than that of non-differentially expressed genes, it is identified as a true differentially expressed gene. This method combines statistical inference and feature extraction, which helps to improve the accuracy and robustness of differential gene screening.

[0005] The technical solution described in Chinese patent "CN202111581852.0 Differential Expression Gene Prediction System Based on Hierarchical Self-Attention Mechanism" includes the following steps: First, the acquisition module extracts epigenetic data such as DNA methylation and histone modification from the cell pairs to be tested and constructs the corresponding input matrix. Then, the encoding module uses a first-layer self-attention mechanism to model each epigenetic factor, calculates the Query, Key, and Value matrices, and weights and summarizes each site using attention weights to extract the encoding vector for each factor. Next, the feature extraction module concatenates the encoding vectors of all epigenetic factors and inputs them into a second-layer self-attention mechanism. It calculates the attention weights between different factors by scaling the dot product, generating a global epigenetic factor attention weight matrix, and finally obtains the feature embedding vector representing the interaction of multiple factors. Finally, the prediction module inputs this embedding vector into a trained classifier and outputs the prediction result of the differentially expressed genes. This system effectively integrates local site features and global factor weights through a two-layer attention mechanism, improving the accuracy and interpretability of differentially expressed gene identification.

[0006] However, while the Chinese patent "CN202111272652.7 A Method and Application for Determining Differentially Expressed Genes in RNA-seq" combines variance screening, statistical testing, and PLS modeling, demonstrating strong comprehensive discriminative ability, it still has some limitations. First, the method is highly dependent on the accuracy of the initial variance modeling; inaccurate estimation of the expected variance may lead to misclassification of noise genes or missed detection of signal genes. Second, the t-test assumes the data follows a normal distribution; however, single-cell RNA-seq data is typically sparse, discrete, and skewed, potentially violating the premise of statistical testing and affecting significance judgment. Furthermore, although the PLS model can extract differential features, it is easily limited by the sample size, with a high risk of the curse of dimensionality and overfitting, especially when the sample size is much smaller than the number of genes. In addition, the method relies on a multi-step processing flow, and errors may accumulate at each step, affecting the accuracy of the final residual determination. For scenarios with high cellular heterogeneity, the stability of residual construction also presents challenges. Therefore, the method as a whole still needs improvement in handling high-noise, high-dimensional sparse data, especially in terms of data preprocessing and modeling robustness.

[0007] The Chinese patent "CN202111581852.0 Differential Expression Gene Prediction System Based on Hierarchical Self-Attention Mechanism" has the following problems: First, the self-attention mechanism itself has high computational complexity. When the input matrix is ​​large, the computational and memory overhead of the computer increases sharply, making it difficult to scale to large-scale samples or whole genome levels. Second, the model relies on a large number of parameters and a pre-training process. If the training data is insufficient, it is prone to overfitting, affecting generalization ability. In addition, the model's interpretability is also weak. Although the weights of self-attention are visualized, the decision path is not transparent, making it difficult to provide an explanation at the biological mechanism level. Finally, the influence of different epigenetic factors on differential expression may have spatiotemporal heterogeneity, but this method lacks a dynamic modeling mechanism, limiting its predictive ability for nasopharyngeal carcinoma. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an automated detection system for differentially expressed genes in nasopharyngeal carcinoma based on genetic algorithms. This system solves the problems of low accuracy and insufficient interpretability in existing technologies for nasopharyngeal carcinoma detection, and meets the current demand for automated detection of differentially expressed miRNA genes related to nasopharyngeal carcinoma.

[0009] The technical solution of this invention is as follows:

[0010] On the one hand, the present invention provides an automatic detection system for differentially expressed genes in nasopharyngeal carcinoma based on genetic algorithms, including a gene sequencing data acquisition module, a data grouping module, a graph generation module, a data preprocessing module, and a differentially expressed gene detection module;

[0011] The gene sequencing data acquisition module is used to acquire gene sequencing data as input to the data preprocessing module; the gene sequencing data uses miRNA data, which can be a CPM file or a raw data file.

[0012] The data grouping module is used to group the gene sequencing data to obtain gene sequencing data of normal sample group and nasopharyngeal carcinoma patient sample group as input to the graph generation module;

[0013] The graph generation module is used to generate distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and grouping statistical plots based on gene sequencing data from the normal sample group and the nasopharyngeal carcinoma patient sample group, respectively, as inputs to the data preprocessing module.

[0014] The data preprocessing module preprocesses the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistical plot input by the graph generation module. The preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are obtained and stored as input to the differentially expressed gene detection module.

[0015] The differentially expressed gene detection module is used to identify differentially expressed genes in nasopharyngeal carcinoma from the input preprocessed gene sequencing data.

[0016] Furthermore, the nasopharyngeal carcinoma differentially expressed gene automatic detection system based on genetic algorithm also includes a visualization module for visualizing the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and previewing the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group.

[0017] Furthermore, the CPM file is used to represent the standardized expression level of the gene in each sample. In the CPM file, the first column "ID" is the gene identifier column, which contains the gene name of the miRNA; the column after "ID" is the sample identifier column; the matrix cells after the first row contain the CPM value of the gene in the corresponding sample; the original data file refers to the probe-level original fluorescence information file of gene sequencing.

[0018] Furthermore, the grouping statistics chart is used to represent the number of samples and genes included in the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0019] Furthermore, the method for grouping and preprocessing CPM files is as follows:

[0020] A1: Load and read the CPM file obtained by the gene sequencing data acquisition module, and convert it into DataFrame format;

[0021] A2: Based on the sample categories in the CPM file, divide all samples into a normal sample group and a nasopharyngeal carcinoma patient sample group, and count the number of samples in each group; the sample categories include the nasopharyngeal carcinoma patient category and the normal sample category.

[0022] A3: Determine whether there are missing CPM values ​​for genes in each sample. For missing values, fill them with the average CPM values ​​of genes in the corresponding group of the sample. Determine whether there are genes whose CPM values ​​are less than the set minimum CPM threshold. For genes whose CPM values ​​are less than the set minimum CPM threshold, fill them with the average CPM values ​​of genes in the corresponding group of the sample.

[0023] A4: Set the minimum sample size for the normal sample group and the nasopharyngeal carcinoma patient sample group respectively, and screen the samples in the filled normal sample group and the nasopharyngeal carcinoma patient sample group according to the minimum sample size of each group, and retain samples with at least the minimum sample size.

[0024] A5: Data standardization was performed on the screened normal sample group and the nasopharyngeal carcinoma patient sample group respectively;

[0025] A6: Determine whether to perform feature selection based on the requirements. If so, perform feature selection on the filled normal sample group and nasopharyngeal carcinoma patient sample group respectively based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics plot to screen genes in the samples. Otherwise, proceed directly to A7.

[0026] A7: Save all genes and their CPM values ​​as a single gene sample, and then save each gene as a separate file according to its sample.

[0027] Furthermore, the methods for grouping and preprocessing the original data files are as follows:

[0028] B1: Read the original data file and retain only the key column data, including: gene name, gene ID, signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity. Treat the gene name, gene ID, signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity of each gene as a gene sample.

[0029] B2: Sort the screened gene samples by gene ID, and then divide the sorted gene samples into normal sample group and nasopharyngeal carcinoma patient sample group and save them to a new preprocessing directory;

[0030] B3: Merge all gene samples from the normal sample group and all gene samples from the nasopharyngeal carcinoma patient sample group to obtain normal sample data files and nasopharyngeal carcinoma patient data files respectively.

[0031] B4: Confirm whether there are missing values ​​in the normal sample data file and the nasopharyngeal carcinoma patient data file. For missing values, fill them with the average value of other gene samples in the same group in that column. At the same time, output the descriptive statistics of the normal sample data file and the nasopharyngeal carcinoma patient data file respectively, including the standard deviation, mean and average correlation between groups of the key column data.

[0032] B5: Perform Z-score standardization on the median, mean, background, signal-to-noise ratio and total signal intensity of the normal sample data file and the nasopharyngeal carcinoma patient data file respectively to obtain the standardized normal sample data file and the nasopharyngeal carcinoma patient data file and save them.

[0033] B6: Apply the 3-Sigma rule to handle outliers in the median, mean, background, signal-to-noise ratio, and total signal intensity of the standardized normal sample data file and the nasopharyngeal carcinoma patient data file, respectively;

[0034] B7: Calculate the score data of each gene sample in the normal sample data file and the nasopharyngeal carcinoma patient data file. For the case of duplicate genes, take the average of their score data and then merge the gene samples. Use the score data to replace the signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity. Finally, each gene sample includes the gene name, gene ID and score data.

[0035] The method for calculating the score data is as follows:

[0036] First, after removing background values, the signal intensity value of each gene is calculated:

[0037] ;

[0038] in, Represents the signal strength value. Represents the signal mean. Represents the median of the signal. Represents the background signal;

[0039] Then, the signal intensity value of each gene is normalized:

[0040] ;

[0041] in, It is the normalized signal strength value. Represents the signal strength value. Represents the total signal strength;

[0042] Finally, signal-to-noise ratio correction is performed to obtain the gene score data: ;

[0043] in, It is the gene score data, and SNR is the signal-to-noise ratio;

[0044] B8: Save the normal sample data file and the nasopharyngeal carcinoma patient data file to the output directory according to the unified naming rules.

[0045] Furthermore, the processing procedure of the differentially expressed gene detection module is as follows:

[0046] C1: Read the gene sequencing data of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, randomly select a certain proportion of gene samples as a gene subset, encode each gene subset to generate an individual, and then generate a certain number of individuals to obtain the initial population.

[0047] C2: Set the current iteration number t=1;

[0048] C3: Evaluate the initial population using multiple pre-trained machine learning models, and calculate the classification accuracy and individual SHAP value for each individual under different machine learning models; the machine learning models include XGBoost, LightGBM, CatBoost, ElasticNet, Ngboost, KNN, and SVM; the individual SHAP value includes the SHAP value of each gene in the individual;

[0049] The specific steps for calculating the classification accuracy and individual SHAP value for each individual under different machine learning models are as follows: The gene sequences of one individual and several patients are input into the machine learning model. The machine learning model performs binary classification on the patient's gene sequence based on the genes included in the individual, determining whether the patient's gene sequence is a normal sample or a nasopharyngeal carcinoma patient sample. The classification accuracy represents the binary classification accuracy of the machine learning model when performing binary classification on the patient's gene sequence using the genes included in the input individual.

[0050] C4: The Robust Rank Aggregation method is used to aggregate the classification accuracy and individual SHAP values ​​obtained from multiple machine learning models to obtain the individual's fitness score;

[0051] The formula for calculating fitness score is:

[0052] ;

[0053] in, An individual's fitness score This represents the average classification accuracy of each machine learning model. This is the preset generation number, where Current_Generation is the current iteration number of the operation. It represents the number of genes in the current individual. It is the average SHAP value of all genes in the current individual. and These are the preset parameter values ​​during static initialization. This is a dynamic penalty term used to control the number of genes;

[0054] C5: Using the tournament selection method, a set number of individuals are randomly selected from the current population to compare their fitness scores, and the individual with the highest fitness score is selected as the parent individual;

[0055] C6: Merge the genes in the selected parent individuals into a gene set, calculate the absolute SHAP value of each gene in the gene set as the selection probability of the crossover operation, randomly select genes in the gene set according to their selection probabilities, and the selected genes become the genes of the new generation of offspring individuals. If the number of genes in the final offspring individuals is less than the expected number of differentially expressed genes, then randomly supplement from all remaining genes to finally obtain the offspring individuals.

[0056] C7: Perform gene position flipping on offspring individuals according to the preset mutation probability of static initialization;

[0057] C8: Repeat C4-C7 to obtain several newly generated offspring individuals, and select the set number of individuals with the highest fitness scores from the previous generation individuals to form a new population. If the number of individuals in the new population is greater than the number of individuals in the initial population, the previous generation individuals with low fitness scores will be removed one by one until the number of individuals in the new population is equal to the number in the initial population.

[0058] C9: Record the fitness score, number of genes, classification accuracy, and SHAP value of genes in each individual in the new population as evolutionary history.

[0059] C10: Determine whether the current iteration count has reached the preset number of generations. If yes, output the individuals in the current population and execute C11; otherwise, increment the iteration count by 1 and return to C3.

[0060] C11: Sort the genes of all individuals in the population according to their SHAP values, and select the genes with the largest SHAP values ​​as differentially expressed genes in nasopharyngeal carcinoma.

[0061] Secondly, this application proposes an electronic device, comprising: one or more processors, and a memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the following steps:

[0062] Acquire gene sequencing data; the gene sequencing data uses miRNA data, which can be a CPM file or a raw data file.

[0063] Gene sequencing data were grouped to obtain gene sequencing data from normal sample group and nasopharyngeal carcinoma patient sample group;

[0064] Based on the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group, distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and grouping statistical plots were generated respectively.

[0065] Based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics input from the graph generation module, the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are preprocessed to obtain and store the preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0066] Visualize the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and preview the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group;

[0067] Differentially expressed genes in nasopharyngeal carcinoma were identified from preprocessed gene sequencing data.

[0068] Thirdly, this application proposes a computer-readable storage medium storing executable instructions that, when executed, cause a processor to perform the following steps:

[0069] Acquire gene sequencing data; the gene sequencing data uses miRNA data, which can be a CPM file or a raw data file.

[0070] Gene sequencing data were grouped to obtain gene sequencing data from normal sample group and nasopharyngeal carcinoma patient sample group;

[0071] Based on the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group, distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and grouping statistical plots were generated respectively.

[0072] Based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics input from the graph generation module, the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are preprocessed to obtain and store the preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0073] Visualize the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and preview the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group;

[0074] Differentially expressed genes in nasopharyngeal carcinoma were identified from preprocessed gene sequencing data.

[0075] Fourthly, this application proposes a computer program product, including a computer program or instructions, which, when executed by a processor, perform the following steps:

[0076] Acquire gene sequencing data; the gene sequencing data uses miRNA data, which can be a CPM file or a raw data file.

[0077] Gene sequencing data were grouped to obtain gene sequencing data from normal sample group and nasopharyngeal carcinoma patient sample group;

[0078] Based on the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group, distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and grouping statistical plots were generated respectively.

[0079] Based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics input from the graph generation module, the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are preprocessed to obtain and store the preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0080] Visualize the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and preview the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group;

[0081] Differentially expressed genes in nasopharyngeal carcinoma were identified from preprocessed gene sequencing data.

[0082] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0083] This invention integrates multiple machine learning algorithms within a genetic algorithm framework, comprehensively evaluating their performance as a fitness function, significantly enhancing the stability and generalization ability of gene detection methods. Simultaneously, it innovatively introduces the SHAP value as an intermediate tool for differentially expressed gene detection, ensuring both statistical significance and biological meaning in the gene detection process, thus improving the interpretability of the results. A dynamic penalty mechanism effectively controls the number of detected genes, promoting model simplicity. This invention also includes an external independent validation process and rich visualization analysis to ensure the reliability and practicality of the results. This invention provides a more scientific and intelligent new tool for differentially expressed gene detection in nasopharyngeal carcinoma, offering reliable reference information for clinical diagnosis and treatment, and is of great significance for the early diagnosis of nasopharyngeal carcinoma. Attached Figure Description

[0084] Figure 1 This is an execution flowchart of an automatic detection system for differentially expressed genes in nasopharyngeal carcinoma based on a genetic algorithm, as described in an embodiment of the present invention.

[0085] Figure 2 This is a structural diagram of the differentially expressed gene detection module in an embodiment of the present invention;

[0086] Figure 3 This is a graph showing the average accuracy of differentially expressed gene classification in nasopharyngeal carcinoma in this embodiment of the invention. Detailed Implementation

[0087] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0088] Example 1:

[0089] An automated detection system for differentially expressed genes in nasopharyngeal carcinoma based on genetic algorithms, such as... Figure 1 As shown, it includes a gene sequencing data acquisition module, a data grouping module, a graph generation module, a data preprocessing module, a visualization module, and a differentially expressed gene detection module;

[0090] The gene sequencing data acquisition module is used to acquire gene sequencing data as input to the data preprocessing module. The gene sequencing data uses miRNA data, which is a CPM file in CSV format or a raw data file in GPR format. The CPM file is an expression matrix file with normalized values ​​per million counts, used to represent the normalized expression level of genes in each sample. In the CPM file, the first column "ID" is the gene identifier column, containing the gene name of the miRNA. From "ID" onwards are the sample identifier columns, such as NPC1, NPC2, NNM1, NNM2, etc., where NPC represents the nasopharyngeal carcinoma patient category and NNM represents the normal sample category. The matrix cells after the first row contain the CPM value of the corresponding gene in the corresponding sample. The raw data file refers to the probe-level raw fluorescence information file of gene sequencing, including columns such as gene name, gene ID, median signal, mean signal, background signal, signal-to-noise ratio, and total signal intensity. All raw data files (GPR format) are organized into the specified input directory.

[0091] The data grouping module is used to group the gene sequencing data to obtain gene sequencing data of normal sample group and nasopharyngeal carcinoma patient sample group as input to the graph generation module;

[0092] The graph generation module is used to generate distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and group statistical plots based on gene sequencing data from the normal sample group and the nasopharyngeal carcinoma patient sample group, respectively. These aid in understanding the data structure and group differences. The distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and group statistical plots are saved to a pre-defined computer directory as input to the data preprocessing module for subsequent feature selection and modeling. The group statistical plots represent the number of samples and genes included in the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0093] The data preprocessing module preprocesses the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistical plot input by the graph generation module. The preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are obtained and stored as input to the differentially expressed gene detection module.

[0094] In this embodiment, relevant settings are configured according to the set gene sequencing data type, including the data preprocessing method and the computer directory for storing and reading related files; then initialization is performed, and the data is automatically processed according to the data preprocessing method and the computer directory for storing and reading related files;

[0095] Specifically, the process for grouping and preprocessing CPM files is as follows:

[0096] A1: Load and read the CPM file (csv format) obtained by the gene sequencing data acquisition module, and convert it into DataFrame format;

[0097] In this embodiment, the data shape and column names of the CPM file are checked to ensure that the data in the CPM file is successfully loaded.

[0098] A2: Based on the sample categories in the CPM file, all samples are divided into a normal sample group and a nasopharyngeal carcinoma patient sample group, and the number of samples in each group is counted to prepare for subsequent group analysis; the sample categories include the nasopharyngeal carcinoma patient category and the normal sample category.

[0099] A3: Determine whether there are missing CPM values ​​for genes in each sample. For missing values, fill them with the average CPM values ​​of genes in the corresponding group of the sample. Determine whether there are genes whose CPM values ​​are less than the set minimum CPM threshold. For genes whose CPM values ​​are less than the set minimum CPM threshold, fill them with the average CPM values ​​of genes in the corresponding group of the sample.

[0100] A4: Set the minimum sample size for the normal sample group and the nasopharyngeal carcinoma patient sample group respectively, and screen the samples in the filled normal sample group and the nasopharyngeal carcinoma patient sample group according to the minimum sample size of each group, retaining at least the minimum sample size, and removing low-quality or non-informative genes in the above way;

[0101] A5: Data standardization was performed on the screened normal sample group and the nasopharyngeal carcinoma patient sample group respectively;

[0102] In this embodiment, the default method is log2 transformation, but z-score, robust, TMM, quantile and other normalization methods can also be selected to eliminate the differences in sequencing depth and distribution between samples;

[0103] A6: Determine whether to perform feature selection based on the requirements. If so, perform feature selection on the filled normal sample group and nasopharyngeal carcinoma patient sample group respectively based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics plot. Screen genes with large expression differences or related to grouping in the sample to reduce the feature dimensions of subsequent analysis. Otherwise, proceed directly to A7.

[0104] In this embodiment, feature selection can be performed using methods such as variance-based, ANOVA-based, or correlation-based methods.

[0105] A7: Save all genes and their CPM values ​​as a single gene sample, according to the sample to which they belong, into separate files for subsequent algorithm and model analysis;

[0106] The grouping and preprocessing process for the raw data files is as follows:

[0107] B1: Read the original data file and retain only the key column data, including: gene name, gene ID, signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity. Treat the gene name, gene ID, signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity of each gene as a gene sample.

[0108] B2: Sort the screened gene samples by gene ID, and then divide the sorted gene samples into normal sample group and nasopharyngeal carcinoma patient sample group and save them to a new preprocessing directory to ensure consistency in subsequent analysis;

[0109] B3: Merge all gene samples from the normal sample group and all gene samples from the nasopharyngeal carcinoma patient sample group to obtain normal sample data files and nasopharyngeal carcinoma patient data files respectively.

[0110] B4: Confirm whether there are missing values ​​in the normal sample data file and the nasopharyngeal carcinoma patient data file. For missing values, fill them with the average value of other gene samples in the same group in that column. At the same time, output the descriptive statistics of the normal sample data file and the nasopharyngeal carcinoma patient data file respectively, including the standard deviation, mean and inter-group average correlation of the key column data, to facilitate subsequent standardization processing.

[0111] B5: Perform Z-score standardization on the median, mean, background, signal-to-noise ratio and total signal intensity of the normal sample data file and the nasopharyngeal carcinoma patient data file respectively to eliminate the influence of different feature dimensions, and obtain the standardized normal sample data file and nasopharyngeal carcinoma patient data file and save them.

[0112] B6: Apply the 3-Sigma rule to outlier handling for the median, mean, background signal, signal-to-noise ratio, and total signal intensity in the standardized normal sample data file and the nasopharyngeal carcinoma patient data file, respectively, to ensure data robustness;

[0113] B7: Calculate the score data of each gene sample in the normal sample data file and the nasopharyngeal carcinoma patient data file. For the case of duplicate genes, take the average of their score data and then merge the gene samples to ensure that each gene sample retains only one record. Use the score data to replace the signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity. Finally, each gene sample includes the gene name, gene ID and score data.

[0114] The method for calculating the score data is as follows:

[0115] First, after removing background values, the signal intensity value of each gene is calculated:

[0116]

[0117] in, Represents the signal strength value. Represents the signal mean. Represents the median of the signal. The background signal is represented by a geometric average. By combining the background values, the actual signal strength can be more accurately reflected, thus improving the robustness of the performance indicators.

[0118] Then, the signal intensity value of each gene is normalized:

[0119] in, It is the normalized signal strength value. Represents the signal strength value. Represents the total signal strength;

[0120] Finally, signal-to-noise ratio correction is performed to obtain the gene score data: ;

[0121] in, It is the gene score data, and SNR is the signal-to-noise ratio;

[0122] B8: Save the normal sample data files and nasopharyngeal carcinoma patient data files to the output directory according to a unified naming rule to complete the data preprocessing process;

[0123] The visualization module is used to visualize the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and to preview the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group.

[0124] The differentially expressed gene detection module is used to identify differentially expressed genes in nasopharyngeal carcinoma from the input preprocessed gene sequencing data. The differentially expressed gene detection module proposed in this invention mainly includes the following aspects:

[0125] I. Multi-algorithm integration and RRA aggregation: Multiple machine learning algorithms such as XGBoost, LightGBM, CatBoost, and ElasticNet are used to evaluate feature subsets. The accuracy of multiple models is aggregated through the RRA method to improve the robustness and stability of feature selection and reduce the bias caused by a single model.

[0126] II. Smart crossover based on SHAP values: In the crossover operation of the genetic algorithm, SHAP values ​​are introduced to intelligently allocate parent genes to offspring based on the importance probability of genes, giving priority to retaining genes that contribute most to classification, while taking into account both statistical significance and biological interpretability.

[0127] III. Dynamic Penalty Mechanism: A dynamic penalty term related to the number of evolutionary generations and the expected number of features is introduced into the fitness function. As evolution progresses, the penalty on the number of features is gradually increased, prompting the algorithm to converge to fewer and better features in the later stages, thus preventing feature redundancy.

[0128] IV. Visual Analysis: Detailed evolutionary history and enrichment analysis provide interpretability and scientific validity of the results, enhancing the generalization ability and reliability of the algorithm results.

[0129] Figure 2 The differentially expressed gene detection module used in this invention combines genetic algorithms and ensemble learning methods. Traditional genetic algorithms include four steps: selection, crossover, mutation, and elimination. This invention uses multiple machine learning algorithms as fitness calculation functions to control the selection step, and uses the ranking score of SHAP values ​​as the crossover step. During model operation, model parameters are dynamically adjusted to regulate the mutation and elimination steps. A grid optimization control function is provided at the entrance of the differentially expressed gene detection module for static fine-tuning of the relevant parameters of each machine learning model. This establishes a differentially expressed gene detection module that combines dynamic modeling mechanism and static initialization.

[0130] The main process of the model of this invention is as follows:

[0131] C1: Read the gene sequencing data of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, randomly select a certain proportion of gene samples as a gene subset, encode each gene subset to generate an individual, and then generate a certain number of individuals to obtain the initial population.

[0132] C2: Set the current iteration number t=1;

[0133] C3: Evaluate the initial population using multiple pre-trained machine learning models, and calculate the classification accuracy and individual SHAP value for each individual under different machine learning models; the machine learning models include XGBoost, LightGBM, CatBoost, ElasticNet, Ngboost, KNN, and SVM; the individual SHAP value includes the SHAP value of each gene in the individual;

[0134] The specific steps for calculating the classification accuracy and individual SHAP value for each individual under different machine learning models are as follows: The gene sequences of one individual and several patients are input into the machine learning model. The machine learning model performs binary classification on the patient's gene sequence based on the genes included in the individual, determining whether the patient's gene sequence is a normal sample or a nasopharyngeal carcinoma patient sample. The classification accuracy represents the binary classification accuracy of the machine learning model when using the genes included in the input individual to perform binary classification on the patient's gene sequence, i.e., the proportion of gene sequences correctly classified by the machine learning model. The calculation formula is "(number of true positive gene sequences + number of true negative gene sequences) ÷ total sample gene sequences × 100%".

[0135] C4: The Robust Rank Aggregation (RRA) method is used to aggregate the classification accuracy and individual SHAP values ​​obtained from multiple machine learning models to obtain the individual's fitness score. The specific formula for calculating the fitness score is as follows:

[0136] ;

[0137] in, An individual's fitness score This represents the average classification accuracy of each machine learning model. This is the preset generation number, where Current_Generation is the current iteration number of the operation. It represents the number of genes in the current individual. It is the average SHAP value of all genes in the current individual. and These are the preset parameter values ​​during static initialization. This is a dynamic penalty term used to control the number of genes;

[0138] C5: Using the tournament selection method, a set number of individuals are randomly selected from the current population to compare their fitness scores. The individual with the highest fitness score is selected as the parent individual to prepare for subsequent crossovers.

[0139] C6: Merge the genes in the selected parent individuals into a gene set, calculate the absolute SHAP value of each gene in the gene set as the selection probability of the crossover operation, randomly select genes in the gene set according to their selection probabilities, and the selected genes become the genes of the new generation of offspring individuals. If the number of genes in the final offspring individuals is less than the expected number of differentially expressed genes, then randomly supplement from all remaining genes to finally obtain the offspring individuals.

[0140] C7: Perform gene position flipping on offspring individuals according to the preset mutation probability of static initialization, that is, increase population diversity and prevent getting trapped in local optima;

[0141] C8: Repeat C4-C7 to obtain several newly generated offspring individuals, and select the set number of individuals with the highest fitness scores from the previous generation individuals to form a new population. If the number of individuals in the new population is greater than the number of individuals in the initial population, the previous generation individuals with low fitness scores will be removed one by one until the number of individuals in the new population is equal to the number in the initial population.

[0142] C9: Record the fitness score, number of genes, classification accuracy, and SHAP value of genes in each individual in the new population as evolutionary history for subsequent analysis and visualization.

[0143] C10: Determine whether the current iteration count has reached the preset number of generations. If yes, output the individuals in the current population and execute C11; otherwise, increment the iteration count by 1 and return to C3.

[0144] C11: Sort the genes of all individuals in the population according to their SHAP values, and select the genes with the largest SHAP values ​​as differentially expressed genes in nasopharyngeal carcinoma.

[0145] In this embodiment, KEGG enrichment analysis and GO enrichment analysis can be performed on differentially expressed genes in nasopharyngeal carcinoma (optional). This can be performed after the algorithm is completed.

[0146] KEGG and GO enrichment analysis can transform a large number of gene screening results into specific biological meanings, helping researchers to grasp the core mechanisms of diseases or biological processes from a macroscopic level. They are important tools for gene function annotation and mechanism research.

[0147] The GO enrichment analysis workflow automatically categorizes differentially expressed genes in nasopharyngeal carcinoma into the GO database's annotation systems for "Bioprocesses (BP)," "Molecular Functions (MF)," and "Cellular Components (CC)," statistically analyzing the enrichment of functional entries and generating corresponding tables and visualizations. The KEGG enrichment analysis workflow maps the selected genes to known biological pathways in the KEGG database, analyzing their distribution in signal transduction, metabolism, and disease pathways. The output is a set of significantly enriched biological pathways, with each pathway providing information such as the number of enriched genes, statistical significance (p-value, FDR), pathway name, and pathway number. The resulting graphs for these pathways are also saved to the corresponding directory.

[0148] The fine-tuning process for the differentially expressed gene detection module in this embodiment is as follows:

[0149] Save the differentially expressed gene detection module to the configured computer directory; load the trained differentially expressed gene detection module; select whether to fine-tune the parameters of the differentially expressed gene detection module. If so, fine-tune them; otherwise, directly use the pre-trained differentially expressed gene detection module to identify differentially expressed genes in nasopharyngeal carcinoma, output the results, and save the differentially expressed genes in nasopharyngeal carcinoma to the configured computer directory.

[0150] The nasopharyngeal carcinoma differentially expressed gene automatic detection system proposed in this invention realizes a complete process for detecting differentially expressed genes in nasopharyngeal carcinoma, provides automatic data preprocessing function, and establishes a differentially expressed gene detection algorithm that combines dynamic modeling mechanism and static initialization, which can efficiently and accurately detect differentially expressed genes in nasopharyngeal carcinoma.

[0151] The automatic detection system for differentially expressed genes in nasopharyngeal carcinoma (NPC) described in this invention utilizes a differential gene detection algorithm to automatically detect and analyze differentially expressed genes in NPC. This invention was tested on gene data from multiple NPC patients and healthy individuals. Results show that this method can accurately screen for differentially expressed genes in NPC, exhibiting high detection accuracy and significant biological value, providing reliable technical support for the molecular diagnosis of NPC. Figure 3 As shown, the average accuracy rate is 99.7%.

[0152] Example 2:

[0153] This embodiment provides an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors perform the following steps:

[0154] Acquire gene sequencing data; the gene sequencing data uses miRNA data, which can be a CPM file or a raw data file.

[0155] Gene sequencing data were grouped to obtain gene sequencing data from normal sample group and nasopharyngeal carcinoma patient sample group;

[0156] Based on the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group, distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and grouping statistical plots were generated respectively.

[0157] Based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics input from the graph generation module, the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are preprocessed to obtain and store the preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0158] Visualize the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and preview the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group;

[0159] Differentially expressed genes in nasopharyngeal carcinoma were identified from preprocessed gene sequencing data.

[0160] The electronic device may be a mobile phone, computer, or tablet computer, and includes a memory and a processor, on which computer programs are stored. It is understood that the electronic device may also include input / output (I / O) interfaces and communication components.

[0161] The processor is used to execute all or part of the steps described in the above embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in the electronic device, as well as application-related data.

[0162] The processor can be implemented as an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a programmable logic device (PLD), a field-programmable gate array (FPGA), a controller, a microcontroller, a microprocessor, or other electronic components.

[0163] Example 3:

[0164] This application proposes a computer-readable storage medium storing executable instructions that, when executed, cause a processor to perform the following steps:

[0165] Acquire gene sequencing data; the gene sequencing data uses miRNA data, which can be a CPM file or a raw data file.

[0166] Gene sequencing data were grouped to obtain gene sequencing data from normal sample group and nasopharyngeal carcinoma patient sample group;

[0167] Based on the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group, distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and grouping statistical plots were generated respectively.

[0168] Based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics input from the graph generation module, the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are preprocessed to obtain and store the preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0169] Visualize the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and preview the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group;

[0170] Differentially expressed genes in nasopharyngeal carcinoma were identified from preprocessed gene sequencing data.

[0171] The aforementioned storage media include: flash memory, hard disks, multimedia cards, card-type memory (e.g., SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR) memory), random access memory (RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, disks, optical discs, servers, APP (Application) app stores, and other media capable of storing program verification codes, on which computer programs are stored.

[0172] Example 4:

[0173] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, performs the following steps:

[0174] Acquire gene sequencing data; the gene sequencing data uses miRNA data, which can be a CPM file or a raw data file.

[0175] Gene sequencing data were grouped to obtain gene sequencing data from normal sample group and nasopharyngeal carcinoma patient sample group;

[0176] Based on the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group, distribution box plots, inter-sample correlation heatmaps, gene expression heatmaps, PCA dimensionality reduction plots, and grouping statistical plots were generated respectively.

[0177] Based on the distribution box plot, inter-sample correlation heatmap, gene expression heatmap, PCA dimensionality reduction plot and grouping statistics input from the graph generation module, the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are preprocessed to obtain and store the preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group.

[0178] Visualize the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group, and preview the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group;

[0179] Differentially expressed genes in nasopharyngeal carcinoma were identified from preprocessed gene sequencing data.

[0180] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0181] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0182] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of this disclosure and its equivalents, then the intent of this disclosure also includes these modifications and variations.

Claims

1. A system for automatic detection of nasopharyngeal carcinoma differentially expressed genes based on genetic algorithm, characterized in that, The system comprises a gene sequencing data acquisition module, a data grouping module, a graph generation module, a data preprocessing module and a differential expression gene detection module. The gene sequencing data acquisition module is configured to acquire gene sequencing data as input of the data preprocessing module; the gene sequencing data uses miRNA data, which is a CPM file or an original data file. The data grouping module is configured to group the gene sequencing data to obtain gene sequencing data of a normal sample group and a nasopharyngeal carcinoma patient sample group as input of the graph generation module. The graph generation module is configured to generate distribution box plots, sample correlation heat maps, gene expression heat maps, PCA dimension reduction plots and grouping statistical plots from the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group as input of the data preprocessing module. The data preprocessing module is configured to preprocess the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group according to the distribution box plots, the sample correlation heat maps, the gene expression heat maps, the PCA dimension reduction plots and the grouping statistical plots input by the graph generation module to obtain and store the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group after preprocessing as input of the differential expression gene detection module. The differential expression gene detection module is configured to identify nasopharyngeal carcinoma differential expression genes from the input gene sequencing data after preprocessing.

2. The system of claim 1, wherein the system is characterized by, The system further comprises a visualization module configured to visualize and preview the CPM files of the normal sample group and the nasopharyngeal carcinoma patient sample group after preprocessing. 3.The system of claim 1, wherein, The CPM file is configured to represent the standardized expression level of genes in each sample; in the CPM file, the first column "ID" is a gene identification column containing the gene name of miRNA; the column after "ID" is a sample identification column; the matrix cells after the first row are CPM values of genes in corresponding samples; and the original data file refers to a probe level original fluorescence information file of gene sequencing.

4. The system of claim 1, wherein the system is characterized by, The grouping statistical plot is configured to represent the number of samples and genes included in the normal sample group and the nasopharyngeal carcinoma patient sample group.

5. The system of claim 1, wherein the system is based on genetic algorithm. The method for grouping and preprocessing the CPM file comprises the following steps: A1: loading and reading the CPM file acquired by the gene sequencing data acquisition module and converting the CPM file into a DataFrame format; A2: dividing all samples into the normal sample group and the nasopharyngeal carcinoma patient sample group according to the sample categories in the CPM file and counting the number of samples in the normal sample group and the nasopharyngeal carcinoma patient sample group; the sample categories include a nasopharyngeal carcinoma patient category and a normal sample category; A3: determining whether the CPM values of genes in each sample are missing, filling the missing values with the average CPM value of genes in the corresponding group of the sample, determining whether the CPM values of genes are less than a set minimum CPM threshold, and filling the CPM values of genes less than the set minimum CPM threshold with the average CPM value of genes in the corresponding group of the sample. A4: Set the minimum sample number of the normal sample group and the nasopharyngeal carcinoma patient sample group respectively, and screen the samples in the filled normal sample group and the nasopharyngeal carcinoma patient sample group according to the minimum sample number of each group respectively, and retain at least the minimum sample number of samples; A5: Perform data standardization processing on the screened normal sample group and nasopharyngeal carcinoma patient sample group respectively; A6: Determine whether to perform feature selection according to the requirement, if yes, perform feature selection on the filled normal sample group and nasopharyngeal carcinoma patient sample group respectively according to the distribution box plot, sample correlation heat map, gene expression heat map, PCA dimension reduction plot and grouping statistical diagram, and screen the genes in the samples, otherwise directly execute A7; A7: Save all genes and their CPM values as a gene sample according to the sample they belong to as a separate file.

6. The system of claim 1, wherein the system is based on genetic algorithm. The grouping and preprocessing method for the original data file is: B1: Read the original data file, and only retain the key column data, including: gene name, gene ID, signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity, and take the gene name, gene ID, signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity of each gene as a gene sample; B2: Sort the screened gene samples according to the gene ID, and save the sorted gene samples into the normal sample group and the nasopharyngeal carcinoma patient sample group under the new preprocessing directory; B3: Merge all gene samples in the normal sample group, and at the same time, merge all gene samples in the nasopharyngeal carcinoma patient sample group, to obtain the normal sample data file and the nasopharyngeal carcinoma patient data file respectively; B4: Confirm whether there are missing values in the normal sample data file and the nasopharyngeal carcinoma patient data file, and fill the missing values with the average value of other gene samples in the column, and output the descriptive statistics of the normal sample data file and the nasopharyngeal carcinoma patient data file respectively, including the standard deviation, average value and inter-group average correlation of the key column data; B5: Perform Z-score standardization processing on the signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity in the normal sample data file and the nasopharyngeal carcinoma patient data file respectively, to obtain the standardized normal sample data file and the nasopharyngeal carcinoma patient data file and save them; B6: Apply 3-Sigma rule to the signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity in the standardized normal sample data file and the nasopharyngeal carcinoma patient data file respectively for abnormal value processing; B7: Calculate the score data of each gene sample in the normal sample data file and the nasopharyngeal carcinoma patient data file, and for the case of repeated genes, take the average of the score data and merge the gene samples, and use the score data to replace the signal median, signal mean, signal background, signal-to-noise ratio and total signal intensity, finally each gene sample includes gene name, gene ID and score data; The calculation method of the score data is: First, calculate the signal intensity value of each gene by integrating the background value: ; wherein, representative signal intensity values, representative signal mean values, representative signal median values, representative signal background; Then normalize the signal intensity value of each gene: ; wherein, is the normalized signal intensity value, represents the signal intensity value, represents the total signal intensity; Finally, correct the signal-to-noise ratio to obtain the score data of the gene: ; wherein, is the score data of the gene, and SNR is the signal-to-noise ratio; B8: Save the normal sample data file and nasopharyngeal carcinoma patient data file to the output directory according to the uniform naming rule.

7. The system of claim 1, wherein the system is based on genetic algorithm. The processing process of the differential expression gene detection module is: C1: Read the gene sequencing data of the preprocessed normal sample group and nasopharyngeal carcinoma patient sample group, randomly select a certain proportion of gene samples as a gene subset, encode each gene subset to generate an individual, and then generate a certain number of individuals to obtain an initialized population; C2: Set the current iteration number t=1; C3: Evaluate the initialized population using multiple pre-trained machine learning models, calculate the classification accuracy and individual SHAP value of each individual under different machine learning models; the machine learning models include XGBoost, LightGBM, CatBoost, ElasticNet, Ngboost, KNN and SVM; the individual SHAP value includes the SHAP value of each gene in the individual; The calculation of the classification accuracy and individual SHAP value of each individual under different machine learning models is as follows: input an individual and the gene sequences of several patients into a machine learning model, and the machine learning model classifies the gene sequences of the patients as normal samples or nasopharyngeal carcinoma patient samples according to the genes included in the individual; the classification accuracy represents the two-classification accuracy of the machine learning model when classifying the gene sequences of the patients through the genes included in the input individual; C4: Aggregate the classification accuracy and individual SHAP value obtained by multiple machine learning models to obtain the fitness score of the individual using the Robust Rank Aggregation method; The fitness score calculation formula is: ; wherein, a fitness score of the individual, is an average of classification accuracies of the respective machine learning models, is a preset number of evolution generations, and Current_Generation is a current iteration number of the operation, is a number of genes of the current individual, is an average SHAP value of all genes in the current individual, and is a preset parameter value at static initialization, is a dynamic penalty term for controlling the number of genes; C5: Use the tournament selection method to randomly select a certain number of individuals from the current population for fitness score comparison, and select the individual with the highest fitness score as the parent individual; C6: Merge the genes in the selected parent individual as a gene set, calculate the absolute SHAP value of each gene in the gene set as the selection probability of the crossover operation, and randomly select the genes in the gene set according to their selection probability. The selected genes are used as the genes of the new generation of offspring individuals. If the number of final offspring individuals is less than the expected number of differential expression genes, randomly supplement from the remaining all genes to finally obtain the offspring individuals; C7: Perform gene bit flipping on the offspring individuals according to the preset mutation probability of static initialization; C8: Repeat C4-C7 to obtain several newly generated offspring individuals, and select a certain number of individuals with the highest fitness score from the previous generation of individuals to form a new population. If the number of individuals in the new population is greater than the number of individuals in the initialized population, remove the previous generation of individuals with low fitness score one by one until the number of individuals in the new population is equal to that in the initialized population; C9: Record the fitness score of each individual in the new population, the number of genes of each individual, the classification accuracy of each individual, and the SHAP value of the genes in each individual as the evolution history. C10: judging whether the current iteration number reaches the preset evolution generation number, if yes, outputting individuals in the current population and performing C11, otherwise, increasing the iteration number by 1 and returning to C3; C11: sorting the genes of all individuals in the population according to SHAP values, and selecting a set number of genes with the largest SHAP values as nasopharyngeal carcinoma differential expression genes.

8. An electronic device, comprising: Comprise: one or more processors, and a memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the following steps: obtaining gene sequencing data; the gene sequencing data uses miRNA data, which is a CPM file or an original data file; grouping the gene sequencing data to obtain gene sequencing data of normal sample groups and nasopharyngeal carcinoma patient sample groups; generating distribution box plots, sample correlation heat maps, gene expression heat maps, PCA dimension reduction maps and grouping statistical maps according to the gene sequencing data of the normal sample groups and the nasopharyngeal carcinoma patient sample groups, respectively; According to the distribution box plot, the sample correlation heat map, the gene expression heat map, the PCA dimension reduction map and the grouping statistical map input by the figure generation module, the gene sequencing data of the normal sample groups and the nasopharyngeal carcinoma patient sample groups are preprocessed to obtain the gene sequencing data of the normal sample groups and the nasopharyngeal carcinoma patient sample groups after preprocessing and storage; Visualize the CPM file of the normal sample groups and the nasopharyngeal carcinoma patient sample groups after preprocessing, and preview the CPM file of the normal sample groups and the nasopharyngeal carcinoma patient sample groups after preprocessing; Identify nasopharyngeal carcinoma differential expression genes from the preprocessed gene sequencing data.

9. A computer-readable storage medium, characterized in that, It stores executable instructions, which when executed cause the processor to perform the following steps: obtaining gene sequencing data; the gene sequencing data uses miRNA data, which is a CPM file or an original data file; grouping the gene sequencing data to obtain gene sequencing data of normal sample groups and nasopharyngeal carcinoma patient sample groups; generating distribution box plots, sample correlation heat maps, gene expression heat maps, PCA dimension reduction maps and grouping statistical maps according to the gene sequencing data of the normal sample groups and the nasopharyngeal carcinoma patient sample groups, respectively; According to the distribution box plot, the sample correlation heat map, the gene expression heat map, the PCA dimension reduction map and the grouping statistical map input by the figure generation module, the gene sequencing data of the normal sample groups and the nasopharyngeal carcinoma patient sample groups are preprocessed to obtain the gene sequencing data of the normal sample groups and the nasopharyngeal carcinoma patient sample groups after preprocessing and storage; Visualize the CPM file of the normal sample groups and the nasopharyngeal carcinoma patient sample groups after preprocessing, and preview the CPM file of the normal sample groups and the nasopharyngeal carcinoma patient sample groups after preprocessing; Identify nasopharyngeal carcinoma differential expression genes from the preprocessed gene sequencing data.

10. A computer program product, characterised in that, It includes a computer program or instructions that are executed by a processor to implement the following steps: obtaining gene sequencing data; the gene sequencing data uses miRNA data, which is a CPM file or an original data file; grouping the gene sequencing data to obtain gene sequencing data of normal sample groups and nasopharyngeal carcinoma patient sample groups; According to the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group, a distribution box plot, a sample correlation heat map, a gene expression heat map, a PCA dimension reduction plot and a grouping statistical plot are generated respectively; According to the distribution box plot, the sample correlation heat map, the gene expression heat map, the PCA dimension reduction plot and the grouping statistical plot input by the figure generation module, the gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group are preprocessed to obtain the preprocessed gene sequencing data of the normal sample group and the nasopharyngeal carcinoma patient sample group and store them; The CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group are visualized, and the CPM files of the preprocessed normal sample group and the nasopharyngeal carcinoma patient sample group are previewed; Nasopharyngeal carcinoma differentially expressed genes are identified from the preprocessed gene sequencing data.

Citation Information

Patent Citations

  • RNA-seq differential expression gene determination method and application

    CN114023385A

  • Differential expression gene prediction system based on layered self-attention mechanism

    CN114283888A

  • Identification of esophageal cancer related characteristic pathways and construction method of early diagnosis model

    CN109841280A

  • Industrial control system intrusion detection method based on improved genetic algorithm feature selection

    CN115688097A

  • Biomarker identification method based on genetic algorithm

    CN118053501A