Liver cancer data screening and marking method and system based on TCGA database and deep learning

By combining TCGA database and deep learning technology, preprocessing and screening of liver cancer data, solving the problems of early accuracy and non-invasiveness in liver cancer diagnosis, and the GLS gene was discovered as a biomarker of liver cancer, improving the early screening of liver cancer and diagnostic accuracy.

CN119068995APending Publication Date: 2024-12-03BEIJING DITAN HOSPITAL CAPITAL MEDICAL UNIVERSTY
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410934057.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing diagnosis methods for liver cancer lack early accuracy and non-invasiveness, which leads to liver cancer being often found in the middle and late stages, with poor treatment effects in patients, and the existing technology is not accurate enough in data screening and labeling.

Method used

Based on TCGA database and deep learning technology, deep learning models are built through data preprocessing and abnormal data identification and clustering processing, and liver cancer data are screened and marked to improve the efficiency and accuracy of data screening.

Benefits of technology

Efficient and accurate screening and marking of liver cancer data was achieved, and GLS gene was found to be highly expressed in the plasma of liver cancer patients, providing a biomarker for screening and diagnosis of non-invasive liver cancer, and improving the reliability of early diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068995A_ABST
    Figure CN119068995A_ABST
Patent Text Reader

Abstract

The invention discloses a liver cancer data screening and marking method and system based on a TCGA database and deep learning. The method comprises the following steps: acquiring liver cancer data from the TCGA database; performing data preprocessing on the liver cancer data to obtain preprocessed data; and inputting the preprocessed data into a pre-trained deep learning model for screening and marking, and outputting a screening and marking result. Rich liver cancer related data are obtained through the TCGA database, the data are processed, based on the deep learning technology, a large amount of liver cancer related gene data are deeply mined and analyzed, and the efficiency and accuracy of data screening are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a liver cancer data screening marker method and system based on TCGA database and deep learning. Background Art

[0002] Primary liver cancer is a malignant tumor that threatens the health and life of people all over the world. Hepatocellular carcinoma (HCC) (referred to as liver cancer) is the most important pathological type of primary liver cancer, which seriously threatens the health and life of all mankind. Its morbidity and mortality rates remain high all year round. Clinical data show that there are many risk factors for liver cancer, such as hepatitis B virus or hepatitis C virus infection, heavy drinking, long-term intake of aflatoxin, non-alcoholic fatty liver disease, etc. The rapid development of imaging and molecular biology technology has made ultrasound, blood alpha-fetoprotein, CT, nuclear magnetic resonance and other technologies attract much attention in the early diagnosis of liver cancer. At present, the treatment of liver cancer mainly relies on liver resection, liver transplantation, radiofrequency ablation, etc. However, due to the lack of early typical clinical symptoms and signs of liver cancer, when it is discovered, the tumor is often in the middle and late stages, and the patient misses the best treatment period. Even if patients with early liver cancer receive surgical treatment, the 5-year survival rate is only 50-70%, and the median survival time of patients with recurrence and metastasis after surgical treatment is only 7-16 months. Patients in the middle and late stages usually have no indications for surgery and have metastases to other organs and tissues. Such patients rely on radiotherapy, chemotherapy, immunotherapy or targeted drug therapy, and the 5-year survival rate is as low as 2.5%. Currently, although alpha-fetoprotein testing plays a certain auxiliary role in the diagnosis and treatment of liver cancer in clinical practice, about 30% of small liver cancers may have normal alpha-fetoprotein levels, and other liver diseases such as acute and chronic hepatitis, cirrhosis, and drug stimulation of the liver may increase alpha-fetoprotein levels. In addition, the gold standard for the diagnosis of liver cancer is puncture biopsy under the guidance of imaging, which is an invasive examination and an important means of diagnosing liver cancer and evaluating tumor progression. However, it has obvious disadvantages and limitations, such as: the operation is invasive and some patients cannot tolerate it; it relies on the rich experience of clinicians; it is not easy to obtain samples for small liver cancers, etc. Therefore, it is urgent to find a more accurate and non-invasive auxiliary diagnostic indicator to provide a basis for the clinical early diagnosis and treatment of liver cancer.

[0003] Therefore, it is necessary to analyze and diagnose based on liver cancer data, and extract useful data by screening and marking liver cancer data. In the prior art, only simple data screening is performed, resulting in inaccurate screened and marked data. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the above-mentioned technology to a certain extent. To this end, the first purpose of the present invention is to propose a liver cancer data screening marker method based on the TCGA database and deep learning, obtain rich liver cancer-related data through the TCGA database, process the data, and deeply mine and analyze a large amount of liver cancer-related gene data based on deep learning technology, thereby improving the efficiency and accuracy of data screening.

[0005] The second purpose of the present invention is to propose a liver cancer data screening marker system based on TCGA database and deep learning.

[0006] To achieve the above objectives, the first embodiment of the present invention proposes a liver cancer data screening marker method based on TCGA database and deep learning, comprising:

[0007] Obtain liver cancer data from the TCGA database;

[0008] Performing data preprocessing on liver cancer data to obtain preprocessed data;

[0009] The preprocessed data is input into a pre-trained deep learning model for screening and marking, and the screening and marking results are output.

[0010] According to some embodiments of the present invention, data preprocessing is performed on liver cancer data, including:

[0011] Performing data recognition on the liver cancer data, determining and eliminating abnormal data, and obtaining first data;

[0012] Data clustering is performed on the first data to obtain second data as preprocessed data.

[0013] According to some embodiments of the present invention, data recognition is performed on liver cancer data, abnormal data is determined and removed, and first data is obtained, including: inputting the liver cancer data into a pre-trained abnormal data recognition model, determining and removing abnormal data based on the output results, and obtaining the first data.

[0014] According to some embodiments of the present invention, a method for obtaining an abnormal data recognition model includes:

[0015] Obtain a training data set, and determine the attribute weights of the training data set based on a correlation analysis method;

[0016] Build a binary tree and perform segmentation based on the binary tree to divide the training data set into normal data and abnormal data;

[0017] An initial model is established, and optimization training of the initial model is performed based on the training data set. Model parameters are adjusted based on cross-validation and grid search techniques to obtain an abnormal data recognition model.

[0018] According to some embodiments of the present invention, performing data clustering processing on first data to obtain second data includes:

[0019] extracting data features of the first data;

[0020] Data clustering is performed based on a hierarchical clustering method according to data features to obtain second data.

[0021] According to some embodiments of the present invention, the further comprising:

[0022] Calculate each cluster set in the second data, determine the cluster value of each cluster set, and judge whether all of them are greater than a preset threshold value; when it is determined that all cluster values ​​are greater than the preset threshold value, it indicates that the clustering process is qualified; otherwise, it indicates that the clustering process is unqualified;

[0023] Determine the clustering value for each cluster set, including:

[0024] Calculate the eigenvalue of each data in the cluster set;

[0025]

[0026] Among them, Q is the characteristic value of the data; z is the semantic position of the data in the semantic tree; x is the length of the data;

[0027] Compute the clustering value for each set of clusters:

[0028]

[0029] Among them, H(i) is the clustering value of the i-th cluster set; n i is the number of data included in the i-th cluster set; Q(i, j) is the characteristic value of the j-th data in the i-th cluster set.

[0030] According to some embodiments of the present invention, a method for obtaining a deep learning model includes:

[0031] Based on multiple data screening dimensions included in the pre-generated target screening strategy, obtain screening parameters corresponding to each data screening dimension;

[0032] Based on the multiple data screening dimensions and the screening parameters corresponding to each data screening dimension, obtaining a sample data set from a target database;

[0033] Extract the association features between the sample data included in the sample data set, and build an association tree as a screening marker structure for the deep learning model.

[0034] According to some embodiments of the present invention, extracting association features between sample data included in a sample data set and constructing an association tree includes:

[0035] All branches of the association tree are constructed based on the association feature settings, and each branch corresponds to a branch code set;

[0036] Mark the p-th branch encoding set L p And divided into m associated segments;

[0037] Calculate the p-th branch encoding set L p The amount of data;

[0038]

[0039] Among them, b p Represents the p-th branch encoding set L p The amount of data; j Represents the p-th branch encoding set L p The data volume of the jth associated segment in ; m represents the number of the associated segments;

[0040] Calculate the normalized data volume of all branch encoding sets;

[0041]

[0042] Among them, b p ′ represents the normalized data volume of all branch encoding sets, n represents the number of branches, b i represents the amount of data corresponding to the branch described in the i-th item, and the value of i is 1, 2, 3, ..., p ... n, b max Indicates the maximum amount of data corresponding to n branches, b min Indicates the minimum amount of data corresponding to n branches, Δa i represents the amount of lost data corresponding to the branch described in the i-th item;

[0043] An association tree is constructed based on the normalized data amounts of all the branch encoding sets.

[0044] According to some embodiments of the present invention, the further comprising:

[0045] Load the preprocessed data into memory and apply for index classification storage area;

[0046] The preprocessed data is distributed to the corresponding classes in the index classification storage area, and then the index is sorted, and the duplicate data after the index sorting is deleted to generate a matrix hash index table.

[0047] To achieve the above-mentioned purpose, the second embodiment of the present invention proposes a liver cancer data screening marker system based on TCGA database and deep learning, comprising:

[0048] The acquisition module is used to obtain liver cancer data from the TCGA database;

[0049] A preprocessing module, used for preprocessing liver cancer data to obtain preprocessed data;

[0050] The screening and marking module is used to input the pre-processed data into a pre-trained deep learning model for screening and marking, and output the screening and marking results.

[0051] The present invention proposes a liver cancer data screening marker method and system based on the TCGA database and deep learning. Rich liver cancer-related data are obtained through the TCGA database, the data is processed, and a large amount of liver cancer-related gene data is deeply mined and analyzed based on deep learning technology, thereby improving the efficiency and accuracy of data screening.

[0052] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0053] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 is a flow chart of a method for screening liver cancer data markers based on TCGA database and deep learning according to an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the expression pattern and clinical characteristics of GLS1 in the TCGA liver cancer dataset;

[0057] Figure 3 is a schematic diagram of the expression levels of GLS1 in plasma and plasma exosomes;

[0058] Figure 4 This is a schematic diagram of the CCK-8 experiment to detect the proliferation activity of liver cancer cells with knockdown and overexpression of GLS1;

[0059] Figure 5 It is a schematic diagram of the detection of invasion and migration ability of liver cancer cells with knockdown and overexpression of GLS1;

[0060] Figure 6 is a schematic diagram of the effects of exosomes from Hep3B cells overexpressing GLS1 on HCC cell viability, invasion, and migration abilities;

[0061] Figure 7It is a schematic diagram of the effect of overexpression of Hep3B cell supernatant exosomes on the biological function of HUVEC;

[0062] Figure 8 It is a block diagram of a liver cancer data screening marker system based on TCGA database and deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0064] like Figure 1 The first embodiment of the present invention proposes a method for screening liver cancer data markers based on TCGA database and deep learning, including steps S1-S3:

[0065] S1. Obtain liver cancer data from the TCGA database;

[0066] S2, performing data preprocessing on the liver cancer data to obtain preprocessed data;

[0067] S3. Input the preprocessed data into a pre-trained deep learning model for screening and marking, and output the screening and marking results.

[0068] The working principle of the above technical solution is: download liver cancer genomic data from the TCGA database, and the platform is RNAseq sequencing data. First, we screen the liver cancer data in the TCGA database: select "TCGA-LIHC" (Liver hepatocellular carcinoma) in Cases, select "Transcriptome Profiling" in Files, select Gene Expression Quantification as the Data Type, select "RNA-Seq" in Experimental Strategy, and select "HTSeq-Counts" as the Workflow Type, and then add the file to the shopping cart. Use R language to download hepatocellular carcinoma data, use the limma package to screen differential genes, and download a total of 371 liver cancer tissues and 50 normal control tissues.

[0069] The liver cancer data is preprocessed to improve the accuracy of the acquired liver cancer data. The preprocessed data is input into a pre-trained deep learning model for screening and labeling, and the screening and labeling results are output.

[0070] The beneficial effects of the above technical solution are: obtaining rich liver cancer-related data through the TCGA database, processing the data, and deeply mining and analyzing a large amount of liver cancer-related genetic data based on deep learning technology, thereby improving the efficiency and accuracy of data screening.

[0071] According to some embodiments of the present invention, data preprocessing is performed on liver cancer data, including:

[0072] Performing data recognition on the liver cancer data, determining and eliminating abnormal data, and obtaining first data;

[0073] Data clustering is performed on the first data to obtain second data as preprocessed data.

[0074] According to some embodiments of the present invention, data recognition is performed on liver cancer data, abnormal data is determined and removed, and first data is obtained, including: inputting the liver cancer data into a pre-trained abnormal data recognition model, determining and removing abnormal data based on the output results, and obtaining the first data.

[0075] According to some embodiments of the present invention, a method for obtaining an abnormal data recognition model includes:

[0076] Obtain a training data set, and determine the attribute weights of the training data set based on a correlation analysis method;

[0077] Build a binary tree and perform segmentation based on the binary tree to divide the training data set into normal data and abnormal data;

[0078] An initial model is established, and optimization training of the initial model is performed based on the training data set. Model parameters are adjusted based on cross-validation and grid search techniques to obtain an abnormal data recognition model.

[0079] Working principle and beneficial effects of the above technical solution: Obtain a training data set: a training data set containing normal and abnormal data samples. The data set should contain multiple attributes or features that are related to the classified object. Label the data set: ensure that each data point is labeled as normal or abnormal. Determine attribute weights based on correlation analysis methods: Calculate the correlation between the attributes in the data set. This can be done by using methods such as Pearson correlation coefficient, Spearman rank correlation coefficient, or mutual information. According to the results of the correlation analysis, determine the importance or weight of each attribute. Attributes with higher correlation with the target variable (normal / abnormal) are given higher weights. Establish a binary tree and perform data segmentation: Binary trees are often used in decision tree algorithms, and the training data set is fitted using a selected tree algorithm (such as random forest). During the fitting process, the algorithm builds a tree structure based on the weight of the attributes and the distribution of the data, and determines the segmentation conditions for each node. Through the node segmentation of the tree, the data is recursively divided into smaller subsets until a certain stopping condition is met (such as too few samples in the node or the purity is high enough). Divide normal data and abnormal data: In the process of building a binary tree (or ensemble tree), the data has been naturally divided into different subsets. These subsets can be marked as normal or abnormal based on their position in the tree structure or the prediction results of the node. A threshold, such as the proportion of abnormal samples under a node, can be set to determine whether the data under the node is considered abnormal. Build an initial model and optimize training: Machine learning models based on anomaly detection, such as Isolation Forest, One-Class SVM variants of Support Vector Machine (SVM), Autoencoder in deep learning, etc. Use the training data set to train the initial model. During the training process, adjust the model's hyperparameters, such as learning rate, number of iterations, regularization strength, etc. Adjust model parameters based on cross-validation and grid search techniques: Use cross-validation techniques to evaluate the performance of the model. Divide the training data set into multiple subsets (such as K-fold cross-validation), use K-1 subsets for training each time, and the remaining subset for validation. For each hyperparameter combination, calculate its average performance in cross validation (such as accuracy, recall, F1 score, etc.). Use grid search technology to traverse all possible hyperparameter combinations and find the hyperparameter combination that optimizes model performance. Obtain anomaly data recognition model: Use the optimal hyperparameter combination found in cross validation and grid search to train the final model to obtain an accurate anomaly data recognition model.

[0080] According to some embodiments of the present invention, performing data clustering processing on first data to obtain second data includes:

[0081] extracting data features of the first data;

[0082] Data clustering is performed based on a hierarchical clustering method according to data features to obtain second data.

[0083] The working principle and beneficial effects of the above technical solution: Extract the data features of the first data; Hierarchical clustering: Calculate the distance: In hierarchical clustering, define the distance between data points. Distance metrics include Euclidean distance, Manhattan distance, etc. Select linkage criteria: Commonly used linkage criteria include single linkage, complete linkage, and average linkage. Perform clustering: Starting with each data point as a separate cluster, merge or split clusters based on the selected linkage criteria and distance metrics until a certain stop condition is met (such as reaching a specified number of clusters or the distance between clusters exceeds a certain threshold). Obtain the second data: The result of hierarchical clustering is a tree structure (cluster tree), in which each node is a cluster. Select clusters according to the different levels of the cluster tree or the size of the clusters, and use the data in these clusters as the second data. You can set a threshold (such as cluster size or distance between clusters) to automatically select clusters. It is easy to obtain the second data.

[0084] According to some embodiments of the present invention, the further comprising:

[0085] Calculate each cluster set in the second data, determine the cluster value of each cluster set, and judge whether all of them are greater than a preset threshold value; when it is determined that all cluster values ​​are greater than the preset threshold value, it indicates that the clustering process is qualified; otherwise, it indicates that the clustering process is unqualified;

[0086] Determine the clustering value for each cluster set, including:

[0087] Calculate the eigenvalue of each data in the cluster set;

[0088]

[0089] Among them, Q is the characteristic value of the data; z is the semantic position of the data in the semantic tree; x is the length of the data;

[0090] Compute the clustering value for each set of clusters:

[0091]

[0092] Among them, H(i) is the clustering value of the i-th cluster set; n i is the number of data included in the i-th cluster set; Q(i, j) is the characteristic value of the j-th data in the i-th cluster set.

[0093] The working principle and beneficial effects of the above technical solution are as follows: the cluster set in the second data is calculated to determine whether the cluster processing is qualified. If the cluster processing is unqualified, the cluster processing needs to be re-performed. When determining the cluster value of each cluster set, the characteristic value of each data in the cluster set is first calculated. Each data is included in the semantic tree, and the distance from the root node to the node of the data in the semantic tree is used as the semantic position. According to the characteristic value of each data in the cluster set, the cluster value of each cluster set is calculated, which facilitates the accurate calculation of the cluster value of each cluster set and improves the accuracy of judging the size of the cluster value and the preset threshold.

[0094] According to some embodiments of the present invention, a method for obtaining a deep learning model includes:

[0095] Based on multiple data screening dimensions included in the pre-generated target screening strategy, obtain screening parameters corresponding to each data screening dimension;

[0096] Based on the multiple data screening dimensions and the screening parameters corresponding to each data screening dimension, obtaining a sample data set from a target database;

[0097] Extract the association features between the sample data included in the sample data set, and build an association tree as a screening marker structure for the deep learning model.

[0098] The working principle and beneficial effects of the above technical solution: Obtain screening parameters: Based on the pre-generated target screening strategy, the strategy contains multiple data screening dimensions. Each data screening dimension usually corresponds to a certain attribute or attribute combination of the data, which can be numerical, categorical or time series, etc. Parse the screening dimension: Parse the various data screening dimensions defined in the screening strategy. Determine the screening parameters: For each screening dimension, determine its corresponding screening parameters. These parameters may be specific values, ranges, conditions, etc. Obtain a sample data set from the target database: Connect to the database: Use an appropriate database connection tool or library (such as SQLAlchemy, pymysql, etc.) to connect to the target database. Construct a query: Based on the screening dimensions and screening parameters, construct an SQL query statement or other database query statement. Execute the query: Execute the query statement to obtain a sample data set that meets the conditions from the database. Extract associated features and construct an association tree: Feature selection: Select or calculate features that can reflect the association relationship between sample data. Including various statistics, correlation indicators, distance metrics, etc. Construct an association tree: Determine the association relationship: Analyze the sample data based on graph theory, network analysis or clustering algorithm to determine which features or data points have an association relationship. Constructing a tree structure: Representing the determined associations in a tree structure. The nodes of the tree can represent data points or features, and the edges represent the associations between them. Optimizing the tree structure: According to actual needs, the association tree can be optimized, such as pruning, merging nodes, etc., to reduce the complexity of the tree and improve readability. As a screening tag structure for deep learning models: Integrate into deep learning models: Use the constructed association tree as a component or input of the deep learning model. Model training: Use a sample data set with association tree tags to train the deep learning model. During the training process, the model can learn the associations between sample data and use these relationships for more accurate screening and tagging.

[0099] According to some embodiments of the present invention, extracting association features between sample data included in a sample data set and constructing an association tree includes:

[0100] All branches of the association tree are constructed based on the association feature settings, and each branch corresponds to a branch code set;

[0101] Mark the p-th branch encoding set L p And divided into m associated segments;

[0102] Calculate the p-th branch encoding set L p The amount of data;

[0103]

[0104] Among them, b p Represents the p-th branch encoding set L pThe amount of data; j Represents the p-th branch encoding set L p The data volume of the jth associated segment in ; m represents the number of the associated segments;

[0105] Calculate the normalized data volume of all branch encoding sets;

[0106]

[0107] Among them, b p ′ represents the normalized data volume of all branch encoding sets, n represents the number of branches, b i represents the amount of data corresponding to the branch described in the i-th item, and the value of i is 1, 2, 3, ..., p ... n, b max Indicates the maximum amount of data corresponding to n branches, b min Indicates the minimum amount of data corresponding to n branches, Δa i represents the amount of lost data corresponding to the branch described in the i-th item;

[0108] An association tree is constructed based on the normalized data amounts of all the branch encoding sets.

[0109] The working principle and beneficial effects of the above technical solution: All branches of the association tree are constructed based on the association feature setting, and each branch corresponds to a branch code set; mark the pth branch code set L p And split into m associated segments, calculate the p-th branch code set L p The amount of data, calculate the normalized amount of data for all branch encoding sets, and the association tree is a tree structure used for classification or regression. In the association tree, each internal node represents a test of a feature or attribute, each branch represents a possible value of the feature or attribute, and each leaf node represents a category or regression value. Use algorithms such as ID3, C4.5, CART, etc. to build association trees. These algorithms select features to split at each node based on criteria such as information gain, gain rate, or Gini impurity. When building an association tree, make sure to use the normalized amount of data. This usually means that when calculating information gain, gain rate, or Gini impurity, the normalized feature values ​​are used. Tree pruning: In order to prevent overfitting, the constructed association tree can be pruned. Pruning is the process of simplifying the tree structure by removing some branches of the tree, usually by deleting some subtrees or replacing them with leaf nodes. Pruning can be achieved by pre-pruning (during tree construction) or post-pruning (after tree construction is completed). Pre-pruning is usually based on some heuristic rules, such as setting the maximum depth of the tree or the minimum number of samples required for node splitting; post-pruning is usually based on cross-validation or an independent validation set to evaluate the performance of the pruned tree, facilitating the construction of the associated tree.

[0110] According to some embodiments of the present invention, the further comprising:

[0111] Load the preprocessed data into memory and apply for index classification storage area;

[0112] The preprocessed data is distributed to the corresponding classes in the index classification storage area, and then the index is sorted, and the duplicate data after the index sorting is deleted to generate a matrix hash index table.

[0113] The working principle and beneficial effects of the above technical solution: Load preprocessed data into memory: load the preprocessed data into the memory of the storage module. Apply for index classification storage area: create an index classification storage area for data, which is a data structure, such as a hash table (HashMap), a dictionary (Dictionary) or a custom data structure, for indexing data according to its characteristics or classification. Distribute data to index classification storage area: distribute the data loaded into the memory to the corresponding class of the index classification storage area according to one or more characteristics (or called "keys"). Involves traversing each element in the data set and placing the data in the corresponding index position according to its key value. Perform index sorting and delete duplicate data: after distributing the data to the index classification storage area, delete the duplicate data. Generate a matrix hash index table based on the data in the index classification storage area, which is a two-dimensional structure, in which each element contains a reference (such as a pointer or key) to the data in the index classification storage area. The matrix hash index table allows fast access to data through the row and column indexes of the matrix, which facilitates the storage and fast indexing of data.

[0114] Enhanced glutamine metabolism is the main feature of tumor amino acid metabolism. Tumors take up a large amount of glutamine to provide energy for their own rapid growth and division. Enhanced glutamine decomposition and increased uptake in tumor cells are key features of metabolic reprogramming. Among them, glutaminase (GLS1) controls the first step of glutamine decomposition. It is the initiating enzyme and rate-limiting enzyme of glutamine decomposition. It is encoded by the GLS1 gene and catalyzes the decomposition of glutamine into glutamate and ammonium ions (NH4+). The GLS1 gene is located on chromosome 2 and consists of 19 exons. The specific splicing process of GLS1 precursor mRNA is regulated by the alternative polyadenylation (APA) regulator CFIm25, producing three splicing variants including KGA, GAC and GAM. GAM has no catalytic activity. Studies have found that GLS1 expression can be observed to be upregulated in a variety of malignant tumors, such as breast cancer, small cell lung cancer, thyroid cancer, colorectal cancer, cervical cancer, lung cancer and melanoma. Triple-negative breast cancer and HER2+ cell lines are highly dependent on glutamine and overexpress GLS1. Tumor cells deprive tumor-infiltrating cells of the glutamine they need, weakening the anti-tumor immune response. Tumor-selective targeting of glutamine metabolism may be a promising therapeutic strategy for triple-negative breast cancer. Non-small cell lung cancer cell lines rely on Gln decomposition to varying degrees, and GAC expression is significantly increased, making them a potential target for cancer metabolic therapy. In addition, studies have found that GLS1 depletion inhibits the proliferation and migration of colorectal cancer cells through a redox / Nrf2 / autophagy-dependent pathway. Studies have shown that dysregulated GLS1 expression plays a key role in a variety of malignant tumors that are highly dependent on glutamine metabolism and has a potential pro-cancer effect.

[0115] Exosomes are extracellular vesicles with a diameter of 30-200nm secreted by living cells. They have a phospholipid bilayer structure similar to the cell membrane structure, and contain rich biologically active components such as proteins, lipids, nucleic acids (DNA, lncRNA, microRNA, circRNA, etc.) and metabolites. They participate in intercellular material exchange, mediate cell information exchange, and regulate various physiological and pathological processes such as immune response, angiogenesis, cardiovascular disease, nervous system diseases and malignant tumors. The various mRNA components contained in exosomes can effectively reflect the dynamic characteristics of the source cells. Exosomal mRNA has obvious heterogeneity between normal cells and tumor cells, so it has great potential to become a biomarker for tumor diagnosis, treatment and prognosis assessment.

[0116] Based on the present invention, liver cancer data was screened and labeled, and it was found that the GLS gene was highly expressed in the plasma of liver cancer patients, and this part of the highly expressed GLS gene was mainly derived from plasma exosomes. As a biomarker, the GLS gene can be used for non-invasive screening, diagnosis and evaluation of liver cancer. In addition, the GLS gene and protein can be used as drug intervention targets for the screening and development of liver cancer drugs. Specifically, by detecting the expression level of GLS gene mRNA in the patient's plasma, it can be suitable for the development of liver cancer gene detection related kits. When the expression level of GLS gene mRNA is significantly higher than the standard of the healthy control population, it can be used as an auxiliary judgment on whether the patient has liver cancer, thereby guiding clinicians to intervene early, and has good application prospects.

[0117] Based on the present invention, primers for liver cancer biomarker GLS1 are determined.

[0118] Among them, the forward primer sequence is TGGGCAACAGTGTTAAGGGA; the reverse primer sequence is AAGGAATGCCTTTGATCACCAC.

[0119] The present invention found that the GLS gene was highly expressed in the plasma of liver cancer patients, and this part of the highly expressed GLS gene was mainly derived from plasma exosomes. Specifically, the following experimental steps were performed according to the attached Figure 2-7 Provide explanation.

[0120] In such Figure 2 A: Expression level of GLS1 in liver cancer tissues and normal control tissues; B: Expression level of GLS1 in each liver cancer tissue and its control tissue; C: Relationship between GLS1 expression level and liver cancer stage; D: Relationship between GLS1 expression level and survival of liver cancer patients; E: Multivariate Cox regression analysis of risk factors for liver cancer patients; F: Expression of GLS1 in liver cancer tissues and normal liver tissues in the HPA database.

[0121] In such Figure 3 A: qRT-PCR experiment was used to detect the expression level of GLS1 mRNA in the plasma of normal subjects and liver cancer patients; B: qRT-PCR experiment was used to detect the expression level of GLS1 mRNA in plasma exosomes of normal subjects and liver cancer patients; C: qRT-PCR experiment was used to detect the expression level of GLS1 mRNA in plasma exosomes of liver cancer patients and the remaining plasma components after collecting exosomes; D: Transmission electron microscopy was used to detect the morphology of plasma exosomes (scale bar: 100 nm); E: Protein immunoblotting experiment was used to detect plasma exosome marker proteins; F: The process of extracting and purifying plasma exosomes by ultracentrifugation.

[0122] In such Figure 4In the figure, A: CCK-8 assay was used to detect the viability of HepG2 cells with GLS1 knocked down; B: CCK-8 assay was used to detect the viability of Li-7 cells with GLS1 knocked down; C: CCK-8 assay was used to detect the viability of Hep3B cells with GLS1 overexpressed.

[0123] In such Figure 5 In the figure, A: Cell scratch assay to detect the migration ability of HepG2 cells with GLS1 knockdown; B: Cell scratch assay to detect the migration ability of Li-7 cells with GLS1 knockdown; C: Cell scratch assay to detect the migration ability of Hep3B cells with GLS1 overexpression; D: Transwell assay to detect the migration ability of HepG2 cells with GLS1 knockdown; E: Transwell assay to detect the migration ability of Li-7 cells with GLS1 knockdown; F: EdU assay to detect the proliferation ability of Hep3B cells with GLS1 overexpression. Scale bar: 200μm.

[0124] In such Figure 6 A: CCK-8 assay to detect the effect of exosomes from Hep3B cells overexpressing GLS1 on HCC cell viability; B: Cell scratch assay to detect the effect of exosomes from Hep3B cells overexpressing GLS1 on HCC cell migration ability (Scale bar: 200 μm); C: Transwell assay to detect the effect of exosomes from Hep3B cells overexpressing GLS1 on HCC cell invasion and migration ability (Scale bar: 200 μm).

[0125] In such Figure 7 In the figure, A: the effect of overexpression of exosomes from Hep3B cell supernatant on HUVEC cell viability; B: the effect of overexpression of exosomes from Hep3B cell supernatant on HUVEC cell migration ability (scale bar: 200 μm); C: the effect of overexpression of Hep3B cell supernatant on HUVEC cell tubule formation ability (scale bar: 500 μm).

[0126] Step 1: Cell culture;

[0127] HepG2 hepatoma cell line was cultured in MEM medium containing 10% FBS and 1% NEAA, Hep3B was cultured in DMEM medium containing 10%, Li-7 was cultured in RPMI-1640 medium containing 10% FBS, and normal hepatocyte L02 cell line was cultured in RPMI-1640 medium containing 10% FBS. Cell culture was cultured in a 5% CO2, 37°C incubator, and cells in the logarithmic growth phase were used for subsequent experiments.

[0128] Step 2: Plasma exosome extraction (ultracentrifugation)

[0129] The collected plasma was centrifuged at 400g, 4℃, for 10min, the supernatant was collected into a new tube, and the precipitate was discarded; 2000g, 4℃, centrifuged for 20min, the supernatant was collected into a new tube, and the precipitate was discarded; 10000g, 4℃, centrifuged for 40min, the supernatant was collected into a new tube, and the supernatant was collected and the precipitate was discarded. The supernatant was first filtered with a 0.22μm filter membrane, then transferred to an 8mL ultracentrifuge tube, sealed, and balanced in an ultracentrifuge, centrifuged at 140000g, 4℃ for 3h30min. The supernatant was carefully aspirated, and the bottom precipitate (mostly invisible) was fully resuspended with 1-2mL PBS (pipette about 200 times), and the suspension was transferred to a new ultracentrifuge tube, supplemented with PBS to 8mL, aspirated again to mix, and sealed. 140000g, 4℃, centrifuged again for 3h30min. Carefully discard the supernatant, and mix the beige or white precipitate at the bottom with 100 μL PBS to obtain the exosome suspension. Take 90 μL of the exosome suspension for subsequent experiments or freeze at -80°C for long-term storage; take 10 μL of the exosome suspension to observe the exosome morphology through a transmission electron microscope; take 10 μL of the exosome suspension for BCA protein quantification experiment.

[0130] Step 3: Identification of plasma exosomes

[0131] (1) Transmission electron microscopy (negative staining method)

[0132] Place the extracted exosome PBS resuspension on ice, cut the sealing film of appropriate size, draw 10μL of exosome suspension and drop it on the sealing film, carefully pick up the copper mesh with tweezers, place it on the exosome droplet, contact for 10 minutes, and use filter paper to absorb excess liquid from the edge of the copper mesh. Add 20μL of 3% phosphotungstic acid solution to the copper mesh covered with exosomes, stain for 2-3 minutes, and use filter paper to absorb excess dye from the edge of the copper mesh. Dry the residual liquid on the copper mesh under incandescent light, observe under electron microscope, and the accelerating voltage is 80kV. Find a suitable field of view and take pictures of exosomes.

[0133] (2) Nanoparticle tracking experiment

[0134] PBS was used to dilute the extracted exosome suspension 3 times to ensure that the concentration of exosome particles was 1×10 7 -1×10 9 The number and size of exosome particles in the sample were measured using a Zetaview instrument under a 405nm laser, and the nanoparticle size distribution diagram was drawn.

[0135] (3) Western Blot assay to detect plasma exosome markers

[0136] ① Protein sample preparation: The collected plasma exosome suspension was mixed with exosome suspension and RIPA lysis buffer (containing 1mM PMSF) at a volume ratio of 3:1, and fully lysed on ice for 30 minutes. Centrifuge at 12000g, 4°C for 5 minutes. Collect the supernatant into a new EP tube, discard the precipitate, and the supernatant is the exosome protein suspension.

[0137] ② The obtained protein supernatant suspension was quantified by the BCA (bicinchoninic acid) method to obtain the protein concentration, and 1 / 4 volume of 5× protein loading buffer was added. The sample was heated in a water bath and boiled for 10 minutes. The sample was placed on ice for subsequent Western Blot experiments.

[0138] ③Electrophoresis: Take out the precast gel from the refrigerator, install it in the electrophoresis tank after balancing to room temperature, add 1× electrophoresis buffer, and vertically pull out the comb in the precast gel in the buffer. According to the protein concentration obtained by the BCA protein concentration detection method, ensure that the protein mass of each loading well is 30 mg, and the loading volume is controlled at 20-25 μL. Add 5-10 μL of pre-stained protein marker to the edge wells of the gel plate as a marker. Add the protein sample to be tested to the middle well. Perform electrophoresis at 80V for about 150 minutes, and stop electrophoresis according to the position of the protein marker band separation.

[0139] ④ Transfer: Take out the plastic clip prepared for electrophoresis, and place the sponge, filter paper, protein gel block, PVDF membrane, filter paper, and sponge on the black side from bottom to top. Remove the bubbles between the gel block and the PVDF membrane in time. Clamp the plastic clip and put it into the electrophoresis tank. The black side of the plastic clip corresponds to the black side of the transfer tank, and the red side of the plastic clip corresponds to the red side of the transfer tank. Pour freshly prepared electrophoresis buffer containing methanol into the tank, turn on the power, and the electrophoresis conditions are constant voltage 125V, and the electrophoresis is 110min. Since the whole process generates a lot of energy, the electrophoresis tank needs to be placed at low temperature.

[0140] ⑤ Blocking: After transfer, the PVDF membrane was placed in 5% skimmed milk powder blocking solution and incubated on a shaker at room temperature for 1.5 hours.

[0141] ⑥ Primary antibody incubation: 5% skim milk powder was prepared in 1×TBST, and the primary antibody was prepared in 5% skim milk powder, with a dilution ratio of 1: 1000. According to the marker position and the molecular weight of the target protein, the PVDF membrane was cut into appropriate strips, and the strips were placed in the primary antibody solution, shaken on a shaker at room temperature for 1 hour, and then incubated at 4°C overnight.

[0142] ⑦ Secondary antibody incubation: After the primary antibody incubation, wash the membrane 5 times with 1×TBST, 5 minutes each time. The secondary antibody is prepared with 5% skim milk powder, and the dilution multiple is 1:2000 (anti-mouse) and 1:3000 (anti-rabbit). The membrane washed with PBS is placed in the secondary antibody and shaken at low speed for 1 hour at room temperature. After the secondary antibody incubation, wash the membrane 4 times with 1×TBST, 7 minutes each time.

[0143] ⑧ Development: The developer is the ECL kit, and the A solution and the B solution are prepared in a 1:1 ratio. Evenly add the developer onto the PVDF membrane, and after fully reacting for 1 minute, turn on the chemiluminescence program, expose and develop, and save the image.

[0144] Step 4: Extraction of total RNA from plasma or exosomes

[0145] Thaw plasma or cell supernatant exosome suspension samples on ice, add 5 times the volume of lysis buffer of exosome suspension, and vortex to mix. Incubate at room temperature for 5 minutes. Add an equal volume of chloroform, cover the tube tightly, and shake vigorously for 15 seconds. Incubate at room temperature for 2-3 minutes. Centrifuge at 4°C, 12000×g for 10 minutes. Transfer the upper aqueous phase to a new EP tube. Add 1.5 times the volume of anhydrous ethanol and mix thoroughly by pipetting. Pipette 700μL of sample into the centrifuge column and centrifuge at 12000×g at room temperature for 15 seconds. Discard the lower layer of liquid until all the liquid is processed. Add 700μL of RWT solution to the centrifuge column and centrifuge at 12000×g at room temperature for 15 seconds. Discard the lower layer of liquid. Add 500μL of RPE solution to the centrifuge column and centrifuge at 12000×g at room temperature for 15 seconds. Discard the lower layer of liquid. Add 500 μL 80% ethanol to the centrifuge column and centrifuge at room temperature at 12,000 × g for 2 min. Discard the column and the lower layer of liquid. Place the centrifuge column in a new 2 mL collection tube and spin for 5 min to fully dry it. Transfer the centrifuge tube to a new EP tube, add 14 μL enzyme-free water to the center of the centrifuge column, and centrifuge at room temperature at 12,000 × g for 1 min. Use a UV spectrophotometer to detect the total RNA concentration and purity.

[0146] Step 5: Real-time fluorescence quantitative PCR of plasma exosome GLS1 mRNA

[0147] (1) Reaction for removing genomic DNA (gDNA): Take out samples and reagents from a -80°C freezer and thaw on ice. The reagent components are as shown in the following table. Incubate at 42°C for 2 min.

[0148]

[0149] (2) Reverse transcription reaction: Add 10 μL of prepared reaction solution B to the above system. The reaction solution B system is as shown in the table below. Mix gently and immediately start reverse transcription reaction. The reaction procedure is: 37°C, 15 min; 85°C, 5 s; 4°C storage. The obtained is the cDNA reaction solution.

[0150]

[0151] (3) Preparation of qRT-PCR reaction system: The qPCR reaction system was prepared according to the following reagent ratios. The system was prepared in 8 rows of qPCR reaction tubes, with 3 replicate wells in each group.

[0152]

[0153]

[0154] (4) qPCR reaction: Place the reaction plate on the quantitative fluorescence PCR instrument and set the reaction parameters according to the following procedure.

[0155]

[0156] (5) After the reaction is completed, GAPDH is used as an internal reference to calculate the relative expression level of the target gene based on the experimental results.

[0157] (6) The primers used in this experimental system were designed as follows. The GLS1 and GAPDH gene sequences were searched using the GenBank section of the NCBI database. Primers for the GLS1 and GAPDH genes were designed using the primer design software Primer5.0. The matching degree between the designed upstream and downstream primers and the gene sequences was compared using the Primer-Blast software. The primers were designed and purified by Shanghai Biotechnology Co., Ltd., and their sequences are shown in the following table:

[0158]

[0159] Step 6: Extraction of exosomes from cell supernatant

[0160] To eliminate the influence of exosomes in cell culture serum, cells were cultured with serum-free basal medium for 24-36h before collecting cell supernatant for exosome extraction experiment. Cells in logarithmic growth phase were inoculated in T75 culture flasks and cultured normally. When the cell fusion rate reached 70%-80%, the medium was changed. The cells were rinsed with PBS 3 times, and 15mL basal medium (serum-free) was added to each culture flask, and incubated in an incubator for 24-36h. A total of 150mL of cell supernatant was collected and transferred to a 50mL centrifuge tube for centrifugation. 400g, 4℃, centrifuged for 10min, cells were removed, and supernatant was collected into a new tube; 2000g, 4℃, centrifuged for 20min, dead cells were removed, and supernatant was collected into a new tube; 10000g, 4℃, centrifuged for 40min, residual cell debris was removed, and supernatant was collected into a new tube. The supernatant was aspirated and filtered with a 0.22μm filter membrane to collect as much filtrate as possible, and the centrifuge tube was placed on ice. Add the supernatant to a 20mL 100kD ultracentrifuge tube, centrifuge at 4000g, concentrate the supernatant to 2mL, and perform size exclusion chromatography (SEC) exosome extraction experiments based on Izon qEV columns. Take out the qEV column from the 4℃ refrigerator at least 30min in advance, place it upright on a tube rack, and equilibrate it to room temperature. Elute the column with one column volume (about 50mL) of 1×PBS. Add 2mL of concentrated cell supernatant suspension, and add 12mL of 1×PBS after it completely passes through the separation column. At this time, the liquid is not collected. Add 10mL of PBS to collect the separation liquid below, which is the exosome fraction obtained. Concentrate the cell supernatant to 90μL again using a 100kD ultrafiltration concentration tube. 60μL is frozen and stored at -80℃ for long-term storage for subsequent experiments; 10μL is used for transmission electron microscopy to identify the morphology of exosomes, and 20μL is used for BCA protein quantitative detection experiments.

[0161] Step 7: CCK8 experiment

[0162] CCK8 assay was used to detect the effects of GLS1 and exosomes on the proliferation of liver cancer cells. Briefly, 4 × 10 3 After culturing for 24h, 48h, 72h, and 96h, 10μL CCK-8 reagent was added to the test wells and incubated in a 37℃ incubator for 2h in the dark. The optical density (OD) of the test wells was measured at a wavelength of 450nm, and the cell growth curve was drawn with time (unit: h) as the horizontal axis and OD value as the vertical axis.

[0163] Step 8: Cell scratch test

[0164] Use a marker pen to draw 5 parallel horizontal lines on the back of the 6-well plate, with a spacing of 0.5 cm between each horizontal line. Select cells with good growth conditions, dilute to an appropriate concentration, evenly plate them in a 6-well plate, and culture overnight. The next day, when the cell density grows to 90%, scratch the cell surface with a 200μL pipette tip. Wash the cells 3 times with PBS to remove the detached cells, add 1mL of serum-free culture medium to each well, observe the cell condition at the scratch position, take pictures, and record them as 0h. After culturing for 24h and 48h, take pictures at the same position as the 0h picture and save the pictures. Use Image J software to measure the cell migration distance and calculate the cell migration ratio.

[0165] Step 9: Transwell assay

[0166] Migration assay: adjust the cell concentration to 5 × 10 5 / mL. Add 600μL of culture medium containing 15% FBS to the 24-well plate, place the Transwell chamber in the 24-well plate, evenly inoculate 200μL of cell suspension in the upper chamber of the chamber, and culture in the incubator for 24h. Wash the upper chamber with PBS 3 times, and gently remove the cells that have not passed through the membrane in the upper chamber with a cotton swab. Fix the chamber with 4% paraformaldehyde for 30min. Stain in 0.1% crystal violet solution for 30min. Wash the chamber with PBS until there is no dye, invert it to air dry, and then observe and take pictures under a microscope.

[0167] For the invasion experiment, the matrix gel should be transferred from -20℃ to 4℃ to thaw one day in advance, and the matrix gel and serum-free medium mixture should be prepared at a ratio of 1:8. 40μL of matrix gel mixture should be added to each well of the Transwell chamber and placed in the incubator for 12h to allow it to completely solidify. The cell concentration should be adjusted to 5×10 5 / mL. Subsequent steps were the same as those in the migration experiment.

[0168] Step 10: Matrigel Tubule Formation

[0169] Obtain HUVEC cells in the logarithmic growth phase and resuspend them in a culture medium containing 0.5-5% FBS to obtain a cell suspension. Thaw the extracellular matrix solution on ice. Equilibrate the wash solution to 37°C in advance. Add 50 μL of extracellular matrix to each well of a 96-well plate and place at 37°C for 1 hour to form a gel. Inoculate 4,000 HUVECs per well of a 96-well plate, add 100 μL of culture medium, and culture in a 37°C incubator for 18 hours. Aspirate the culture medium and wash the cells with 100 μL of wash solution. Add 100 μL of dye solution to each well and culture in a 37°C incubator for 30 minutes. Observe the formation of tubules under an inverted fluorescence microscope.

[0170] Step 11: Statistical analysis

[0171] All experimental results were performed with at least 3 biological replicates and are presented as mean ± standard deviation. The results are shown in Table 1. SPSS 22.0 software was used for data processing and statistical analysis. GraphPad 8.0 software was used to draw images. The independent sample t test was used for comparison between the two groups. P < 0.05 indicated that the results were statistically significant.

[0172] The experimental results obtained are:

[0173] 1. GLS1 is highly expressed in liver cancer tissue and is associated with patient survival rate

[0174] In order to study the expression of GLS1 in liver cancer tissues, we downloaded liver cancer datasets from the TCGA database and used bioinformatics methods to analyze the expression and clinical significance of GLS1. The expression level of GLS mRNA in liver cancer tissues of patients with liver cancer was significantly higher than that in control tissues (Figures A and B). The results of GLS1 mRNA levels and liver cancer staging showed that the higher the malignancy of liver cancer, the higher the expression of GLS1 mRNA (Figure A and B). Figure 6 C). The results of survival analysis showed that the level of GLS1 mRNA in liver cancer tissue was closely related to the patient's survival period, and the survival rate of patients in the group with high expression of GLS1 mRNA was low. Using Cox regression to analyze the risk factors for liver cancer progression, it was found that GLS1 mRNA was an independent risk factor for the occurrence and development of liver cancer (Figure E). Subsequently, we used the HPA database to mine the expression levels of GLS1 protein in liver cancer tissue and liver tissue, and found that the expression of GLS1 protein in liver cancer tissue was increased and darkly stained (Figure F). This means that for liver cancer, high expression of GLS1 is an adverse factor.

[0175] 2. The expression of plasma GLS1 mRNA is elevated in patients with liver cancer, and the elevated GLS1 mainly originates from exosomes.

[0176] The expression of GLS1 mRNA in plasma and extracted exosomes from 60 HCC patients and 30 healthy controls was detected. First, qRT-PCR results showed that GLS1 mRNA was elevated in the plasma of patients with liver cancer (Figure A). In order to identify the source of this part of GLS1 mRNA, we detected the exosomes in the plasma and the remaining plasma after extracting the exosomes, and found that the GLS1 mRNA in the plasma exosomes was significantly higher than the GLS1 mRNA in the remaining plasma after collecting the exosomes. This result suggests that the GLS1 mRNA in plasma mainly comes from the exosome components in the plasma. The plasma exosomes extracted by ultracentrifugation were saucer-like, round or oval in shape, with light-stained membrane structures at the edges and dark-stained in the middle of the exosomes. The average diameter of plasma exosomes from HCC patients detected by NTA was 101.9 nm, and the concentration of exosomes in the sample was 4.7E+10 / mL after 3-fold dilution. Western Blot assay was used to detect positive markers CD69, TSG101 and negative marker Calnexin for exosomes, and the results showed that TSG101 and CD63 were positive, while Calnexin was negative in plasma exosomes from normal subjects and HCC patients. Compared with the expression level of GLS1 mRNA in plasma exosomes from 30 subjects, the expression level of GLS1 mRNA in plasma exosomes from 60 HCC patients was significantly increased, which was 3.42±0.12 of that in the control group. Compared with the expression level of GLS1 in plasma exosomes from 30 subjects, the expression level of GLS1 mRNA in plasma exosomes from 60 HCC patients was significantly increased, which was 3.42±0.12 of that in the control group. The mRNA expression level of GLS1 in the plasma of 60 HCC patients was also significantly increased, which was 2.58±0.09 times that of the control group.

[0177] 3. GLS1 has the ability to promote the proliferation, invasion and migration of liver cancer cells

[0178] Lentiviral vectors with specific overexpression sequence (GV492, element sequence is Ubi-MCS-3FLAG-CBh-gcGFP-IRES-puromycin) and knockdown sequence (GV493, element sequence is hU6-MCS-CBh-gcGFP-IRES-puromycin, which is shGLS2 in this example) were synthesized for the GLS1 sequence and transfected into the liver cancer cell line to overexpress and knockdown the GLS1 gene and protein levels in the liver cancer cell line, and then detect the biological function of the liver cancer cell line.

[0179] Figure 3 The results showed that the CCK8 reagent was used to detect the changes in the proliferation ability of liver cancer cells after knockdown and overexpression of GLS1. After knockdown of GLS1 in HepG2 and Li-7 cells, the OD values ​​of cells at 450nm decreased compared with those in the control virus vector transfection group. After overexpression of GLS1 in Hep3B cells, the OD values ​​of cells at 450nm increased compared with those in the control virus vector transfection group.

[0180] Figure 4 A and 4B show that the cell scratch experiment detected the changes in the scratch migration ratio of liver cancer cells after knockdown and overexpression of GLS1. The scratch migration ratios of HepG2 cells shGLS2-HepG2 and shGLS3-HepG2 with GLS1 knockdown were 69.223±1.568 and 67.537±6.406, respectively, which were significantly higher than 55.760±1.913 of the control group; the scratch migration ratios of Li-7 cells shGLS2-Li-7 and shGLS3-Li-7 with GLS1 knockdown were 73.680±2.191 and 77.683±3.175, respectively, which were significantly higher than 58.507±0.684 of the control group. Figure 4 C The migration ratio of liver cancer cells overexpressing GLS1 was 76.57±3.221, which was significantly higher than that of the control group (37.24±1.743). The results showed that the migration ability of liver cancer cells was enhanced after overexpression of GLS1.

[0181] Transwell assay was used to detect the changes in the migration ability of liver cancer cells after knockdown and overexpression of GLS1. The number of cells migrating through the Transwell chamber in HepG2 cells shGLS2-HepG2 and shGLS3-HepG2 with GLS1 knockdown was 19.33±2.867 and 15.67±1.247, which was significantly lower than that in the control group (66.67±7.409 ( Figure 3 D); The number of cells migrating through the Transwell chamber in Li-7 cells shGLS2-Li-7 and shGLS3-Li-7 with GLS1 knockdown was 25.33±1.247 and 21.67±2.867, which was significantly lower than that of the control group (81.33±1.7 ( Figure 3 E); The number of cells migrating through the Transwell chamber in liver cancer cells overexpressing GLS1 was 36.34±7.241, which was significantly higher than that in the control group (16.22±2.543).

[0182] Matrigel Transwell assay was used to detect changes in the invasion ability of liver cancer cells after knockdown and overexpression of GLS1. The number of cells migrating through the Transwell chamber of HepG2 cells shGLS2-HepG2 and shGLS3-HepG2 with GLS1 knockdown was 14.56±1.654 and 14.33±1.843, which were significantly lower than 64.33±2.055 of the control group; the number of cells migrating through the Transwell chamber of Li-7 cells shGLS2-Li-7 and shGLS3-Li-7 with GLS1 knockdown was 11.23±1.532 and 10.32±2.160, which were significantly lower than 55.332±6.236 of the control group; the number of cells migrating through the Transwell chamber of liver cancer cells overexpressing GLS1 was 23.52±3.546, which was significantly higher than 48.34±3.413 of the control group.

[0183] 4. GLS1 affects the proliferation, migration and invasion of liver cancer cells through exosomes Figure 5 The results showed that the CCK-8 experiment detected the changes in the viability of Hep3B cells after the exosomes of liver cancer cells overexpressing GLS1 (Hep3B-OE-exo) acted on them. After treatment with exosomes containing GLS1, the viability of Hep3B cells in the group adding Hep3B-OE-exo was significantly increased compared with that in the group adding exosomes from the control Hep3B cell supernatant at 96h ( Figure 5 A).

[0184] Figure 5 The results showed that the cell scratch assay detected the effect of exosomes from HCC cells overexpressing GLS1 on the migration ability of Hep3B cells with relatively low GLS1 expression. Compared with the control group, the migration distance ratio of Hep3B cells acted by exosomes from HCC cells overexpressing GLS1 was 67.214±5.893, which was significantly higher than that of the control group (43.543±3.745). Figure 5 B).

[0185] Figure 5 The results showed that the Transwell assay detected the effect of exosomes from HCC cells overexpressing GLS1 on the migration ability of Hep3B cells with relatively low GLS1 expression. After the exosomes from HCC cells overexpressing GLS1 were used to treat HCC cells with low GLS1 expression, the number of Hep3B cells that migrated was 173.253±10.687, which was significantly higher than that of the control group (53.873±5.195). Figure 5 C).

[0186] Figure 5The results showed that the Transwell assay detected the effect of exosomes from HCC cells overexpressing GLS1 on the invasion ability of Hep3B cells with relatively low GLS1 expression. After the exosomes from HCC cells overexpressing GLS1 were used to treat HCC cells with low GLS1 expression, the number of Hep3B cells that migrated was 86.027±5.536, which was significantly higher than that of the control group (37.213±2.044). Figure 5 D) The above results suggest that liver cancer cells can promote the increase of liver cancer cell vitality, migration ability and invasion ability by secreting exosomes containing GLS1.

[0187] 5. GLS1 affects endothelial cell proliferation, migration and tubulogenesis through exosomes

[0188] Figure 6 A shows that the CCK-8 assay detected the cell viability of HUVECs after treatment with exosomes containing GLS1 secreted by liver cancer cells. At 96 h, the cell viability of HUVECs was significantly higher than that of the control group ( Figure 6 A).

[0189] Figure 6 B shows that the Transwell assay detected the changes in the migration ability of HUVEC cells after co-culture of exosomes from liver cancer cells overexpressing GLS1 with HUVEC. After HUVEC cells were treated with exosomes from HCC cells overexpressing GLS1, the number of HUVEC cells that migrated was 234.45±5.674, which was significantly higher than that of the control group (34.87±1.239 ( Figure 6 B).

[0190] Figure 6 C shows that the matrix gel tubule formation experiment detects whether the exosomes GLS1 derived from liver cancer cells affects the ability of HUVEC to form tubules. After HUVEC was exposed to matrix gel for 18 hours, negative cells (with matrix gel, control exosomes added, and culture medium) HUVEC could not effectively form blood vessels. Compared with the exosomes in the supernatant of Hep3B cells, the exosomes of Hep3B cells overexpressing GLS1 were able to form more blood vessels after acting on HUVEC ( Figure 5 C). The above results suggest that GLS1 from HCC cell-derived exosomes can significantly promote HUVEC angiogenesis.

[0191] like Figure 8 As shown, the second embodiment of the present invention proposes a liver cancer data screening marker system based on TCGA database and deep learning, including:

[0192] The acquisition module is used to obtain liver cancer data from the TCGA database;

[0193] A preprocessing module, used for preprocessing liver cancer data to obtain preprocessed data;

[0194] The screening and marking module is used to input the pre-processed data into a pre-trained deep learning model for screening and marking, and output the screening and marking results.

[0195] The beneficial effects of the above technical solution are: obtaining rich liver cancer-related data through the TCGA database, processing the data, and deeply mining and analyzing a large amount of liver cancer-related genetic data based on deep learning technology, thereby improving the efficiency and accuracy of data screening.

[0196] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A liver cancer data screening marker method based on TCGA database and deep learning, characterized in that: include: Obtain liver cancer data from the TCGA database; Performing data preprocessing on liver cancer data to obtain preprocessed data; The preprocessed data is input into a pre-trained deep learning model for screening and marking, and the screening and marking results are output.

2. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 1, characterized in that: Perform data preprocessing on liver cancer data, including: Performing data recognition on the liver cancer data, determining and eliminating abnormal data, and obtaining first data; Data clustering is performed on the first data to obtain second data as preprocessed data.

3. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 1, characterized in that: Performing data recognition on liver cancer data, determining and removing abnormal data, and obtaining first data includes: inputting the liver cancer data into a pre-trained abnormal data recognition model, determining and removing abnormal data based on an output result, and obtaining the first data.

4. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 3, characterized in that: The method for obtaining the abnormal data recognition model includes: Obtain a training data set, and determine the attribute weights of the training data set based on a correlation analysis method; Build a binary tree and perform segmentation based on the binary tree to divide the training data set into normal data and abnormal data; An initial model is established, and optimization training of the initial model is performed based on the training data set. Model parameters are adjusted based on cross-validation and grid search techniques to obtain an abnormal data recognition model.

5. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 2, characterized in that: Performing data clustering processing on the first data to obtain second data includes: extracting data features of the first data; Data clustering is performed based on a hierarchical clustering method according to data features to obtain second data.

6. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 5, characterized in that: Also includes: Calculate each cluster set in the second data, determine the cluster value of each cluster set, and determine whether they are all greater than a preset threshold; When it is determined that all clustering values ​​are greater than the preset threshold, it means that the clustering process is qualified; On the contrary, it means that the clustering process is unqualified; Determine the clustering value for each cluster set, including: Calculate the eigenvalue of each data in the cluster set; Among them, Q is the characteristic value of the data; z is the semantic position of the data in the semantic tree; x is the length of the data; Compute the clustering value for each set of clusters: Among them, H(i) is the clustering value of the i-th cluster set; n i is the number of data included in the i-th cluster set; Q(i, j) is the characteristic value of the j-th data in the i-th cluster set.

7. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 1, characterized in that: Methods for obtaining deep learning models include: Based on multiple data screening dimensions included in the pre-generated target screening strategy, obtain screening parameters corresponding to each data screening dimension; Based on the multiple data screening dimensions and the screening parameters corresponding to each data screening dimension, obtaining a sample data set from a target database; Extract the association features between the sample data included in the sample data set, and build an association tree as a screening marker structure for the deep learning model.

8. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 7, characterized in that: Extract the association features between the sample data included in the sample data set and build an association tree, including: All branches of the association tree are constructed based on the association feature setting, and each branch corresponds to a branch code set; Mark the p-th branch encoding set L p And divided into m associated segments; Calculate the p-th branch encoding set L p The amount of data; Among them, b p Represents the p-th branch encoding set L p The amount of data; j Represents the p-th branch encoding set L p The data volume of the jth associated segment in ; m represents the number of the associated segments; Calculate the normalized data volume of all branch encoding sets; Among them, b p ′ represents the normalized data volume of all branch encoding sets, n represents the number of branches, b i represents the amount of data corresponding to the branch described in the i-th item, and the value of i is 1, 2, 3, ..., p ... n, b max Indicates the maximum amount of data corresponding to n branches, b min Indicates the minimum amount of data corresponding to n branches, Δa i represents the amount of lost data corresponding to the branch described in the i-th item; An association tree is constructed based on the normalized data amounts of all the branch encoding sets.

9. The method for screening liver cancer data markers based on TCGA database and deep learning according to claim 2, characterized in that: Also includes: Load the preprocessed data into memory and apply for index classification storage area; Distribute the preprocessed data to the corresponding classes in the index classification storage area, then sort the index, delete the duplicate data after the index sorting, and generate a matrix hash index table.

10. A liver cancer data screening marker system based on TCGA database and deep learning, characterized in that: include: The acquisition module is used to obtain liver cancer data from the TCGA database; A preprocessing module, used for preprocessing liver cancer data to obtain preprocessed data; The screening and marking module is used to input the pre-processed data into a pre-trained deep learning model for screening and marking, and output the screening and marking results.

Citation Information

Patent Citations

  • Decision tree feature selection method based on feature weights

    CN111488903A

  • Abnormal data screening method for internet financial risk control

    CN114511399A

  • Method for improving gastric cancer typing prognosis prediction precision based on gradient lifting depth feature selection algorithm

    CN116417070A