Tumor classification and potential target prediction method based on personalized metabolic model

Abstract

The method for tumor typing and potential target prediction based on personalized metabolic model disclosed in the present invention comprises the following steps: Step 1: constructing a personalized metabolic network model for tumor patients; Step 2: performing accurate molecular typing based on the metabolic characteristics of the personalized metabolic network model for tumor patients described in Step 1; Step 3: predicting genes that have an important impact on tumor cell growth as potential targets based on the personalized metabolic network model for tumor patients described in Step 1. The present invention has the following advantages: 1. The survival curves corresponding to the molecular typing scheme established by the present invention are more significantly different. 2. The prognostic markers discovered based on the metabolic hierarchy can be detected through body fluids or blood, which is more convenient and easy to promote. 3. The potential targets predicted by the present invention are obtained based on quantitative simulation of tumor cell growth. Patients of different typing have personalized targets, which can improve the drug effectiveness response.

Description

Technical Field

[0001] The present invention relates to the field of cancer subtyping, and in particular to a method for tumor typing and potential target prediction based on an individualized metabolic model. It distinguishes different tumor patients based on the metabolic flux change characteristics predicted by the individualized metabolic network model, performs precise molecular typing, and discovers genes that have an important impact on tumor cell growth as candidate targets through computational simulation. Background Art

[0002] The key to precision medicine for tumors is personalization. Personalized diagnosis and treatment measures require accurate molecular typing, accurate prognostic markers and therapeutic targets. Due to the high heterogeneity, poor prognosis, high mortality and difficulty in early diagnosis of malignant tumors, there is still a lack of effective typing and targeted diagnosis and treatment methods.

[0003] As people gain a deeper understanding of tumor biology and the complexity of tumor metabolism, it is found that cancer cells have a large amount of metabolic heterogeneity compared with normal tissues. In addition to the Warburg effect, cancer cells also have many differences in metabolism from normal cells. For example, p53-deficient colon cancer cells activate the mevalonate pathway to promote the synthesis of ubiquinone, which is essential for maintaining mitochondrial electron transport for respiration and pyrimidine synthesis in a metabolically impaired environment. Studies have shown that abnormal mitochondrial metabolism in the human liver is one of the carcinogenic factors. Immune checkpoints in liver cancer regulate the mitochondrial metabolic energy, tumor microenvironment, and tumor-specific immune response of tumors. This regulation forms a positive cycle between the metabolic reprogramming of cancer cells and immune cells. Cancer progression is closely related to metabolic reprogramming. The metabolic phenotype and metabolic dependence of cancer change during the process of cancer development from precancerous lesions to local infiltration to metastasis. The phenomenon of tumor cell changes in "metabolic reprogramming" is considered to be one of the ten hallmarks of cancer.

[0004] There are some methods for molecular classification of cancer based on omics information. For example, researchers have used information from early hepatocellular carcinoma proteomics and phosphorylation proteomics to divide early hepatocellular carcinoma into three subtypes: SI, S-II, and S-III. The biggest feature of the S-III subtype is impaired cholesterol homeostasis. In addition, molecular classification of HBV-induced liver cancer patients based on proteomic data revealed that the three subtypes had significantly different prognostic survival, and their protein expressions were enriched in metabolic reprogramming, immune microenvironment, and cell proliferation processes. However, there is currently no method for classifying cancer based on metabolic hierarchical characteristics, especially a universal analysis method for various cancers.

[0005] At present, the clinical staging schemes for cancer mainly include BCLC staging, TNM staging, etc. Although there are objective measurement standards, the heterogeneity of patients in the same stage is still significant, and patients in different stages cannot show significantly different survival conditions, which affects the prognosis assessment and the choice of treatment options. Therefore, the present invention performs a more effective classification of cancer patients at the molecular level, and implements individualized treatment methods for patients with different classifications accordingly. In addition, clinically, the effective response of cancer patients to targeted drugs is limited. For example, the effective response of liver cancer to first-line drugs such as sorafenib and lefitinib is only about 30%. Therefore, finding more effective individualized treatment targets is a bottleneck to be broken through in the field of precision medicine for liver cancer.

[0006] In addition, Chinese patents have disclosed some methods for classifying tumor patients, such as a tumor classification method based on radiomics disclosed in China Invention Patent Publication No. CN108765411A; a cancer subtyping method and device based on RNA targeted sequencing and machine learning disclosed in China Invention Patent Publication No. CN110400601A; a method for constructing an radiomics model for pancreatic cancer prediction disclosed in China Invention Patent Publication No. CN111311558A; a method for identifying the infiltration classification of lung adenocarcinoma disclosed in China Invention Patent Publication No. CN110415206A; and a method for lung cancer classification disclosed in China Invention Patent Publication No. CN107849613A. However, none of them are methods for classifying cancer based on metabolic hierarchical characteristics. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a method for tumor classification and potential target prediction based on an individualized metabolic model in response to the shortcomings of current methods for classification and potential target prediction of tumor patients. It solves the challenge of accurate molecular classification of cancer based on metabolic levels, and predicts potential targets conveniently and quickly based on quantitative simulation of the model, providing an important reference for accurate diagnosis and treatment of tumors.

[0008] In order to achieve the purpose of the present invention, the technical solution adopted by the present invention is:

[0009] A method for tumor classification and potential target prediction based on an individualized metabolic model comprises the following steps:

[0010] Step 1: Construct a personalized metabolic network model for cancer patients;

[0011] Step 2: Perform precise molecular typing based on the metabolic characteristics of the tumor patient's individualized metabolic network model described in step 1;

[0012] Step 3: Based on the tumor patient's individualized metabolic network model described in step 1, genes that have important effects on tumor cell growth are predicted as potential targets;

[0013] The step 2 and step 3 may be performed separately or simultaneously.

[0014] In a preferred embodiment of the present invention, the specific method for constructing a personalized metabolic network model for cancer patients in step 1 is: selecting the human genome-scale metabolic network model Human-GEM as a general model, sorting the patient's transcriptome expression data or proteome expression data according to the expression level, and dividing them into high, medium, low, and non-expressed expression spectrum data, using the tINIT algorithm to integrate the expression spectrum data with the Human-GEM model, setting constraints on metabolic reaction flow according to the expression situation, thereby obtaining a personalized metabolic network model for cancer patients.

[0015] In a preferred embodiment of the present invention, the patient's transcriptome expression data is divided into a tumor transcriptome corresponding to the tumor tissue and a normal transcriptome corresponding to the adjacent cancer tissue; the proteome expression data is divided into a tumor proteome corresponding to the tumor tissue and a normal proteome corresponding to the adjacent cancer tissue; the tumor patient's personalized metabolic network model is divided into a personalized normal metabolic model and a personalized tumor metabolic model.

[0016] In a preferred embodiment of the present invention, the specific method of performing accurate molecular typing based on the metabolic characteristics of the individualized metabolic network model of tumor patients described in step 1 in step 2 is as follows: First, based on the metabolic flux information obtained by simulation, hierarchical clustering is performed on tumor cells and normal cells to identify key metabolic characteristics that can distinguish tumors from normal cells. Then, hierarchical clustering or k-means clustering is performed on all tumor models, and the Kaplan-Meie survival analysis test is performed using clinical data to obtain molecular typing with significant survival differences, and metabolic reactions and metabolites with significant differences between different subtypes are found.

[0017] In a preferred embodiment of the present invention, the specific method of predicting potential targets based on the individualized metabolic network model of tumor patients in step 1 in step 3 is: for the constructed metabolic network model of patient-individualized tumor tissue, taking tumor cell growth as the optimal objective function, a flow balance analysis is performed to obtain the original growth rate grWT, as shown in formula (1):

[0018] subject to: S·v = 0 (1)

[0019] a j ≤ v ≤ b j

[0020] S represents the stoichiometric matrix, V represents the metabolic flux of all reactions, that is, the reaction rate, S·v satisfies the conservation of mass, a jand b j represent the upper and lower constraints of the reaction flow rate respectively;

[0021] Then all genes were knocked out one by one to obtain the growth rate after knockout grKO, and the growth rate change of tumor cells was calculated according to formula (2):

[0022] grRatio=grWT / grKO (2)

[0023] Here, the range of grRatio is [0,1]. Usually, when the ratio grRatio <= 0.5, the gene is considered to be a lethal gene for tumor cells and can significantly inhibit the growth of tumor cells. At the same time, in order to avoid damage to normal cells, the candidate gene is not required to be an essential gene for growth in the normal cell metabolism model, and the candidate gene is required to appear in at least 1 / 3 of the samples, which can be used as a potential therapeutic target; the simulation of gene combination knockout is similar to single gene knockout, that is, all pairwise gene combination knockouts are traversed to obtain the change ratio of tumor cell growth.

[0024] In a preferred embodiment of the present invention, the survival analysis is a test of survival analysis using clinical data to obtain molecular typing with significant survival differences and to discover metabolic reactions and metabolites with significant differences between different subtypes.

[0025] In a preferred embodiment of the present invention, the tumor is liver cancer.

[0026] Due to the adoption of the above technical solution, the present invention has the following advantages:

[0027] 1. The survival curves corresponding to the molecular typing scheme established in the present invention are more significantly different.

[0028] 2. Prognostic markers discovered based on metabolic levels can be detected through body fluids or blood, which is more convenient and easy to promote.

[0029] 3. The potential targets predicted by the present invention are obtained based on the quantitative simulation of tumor cell growth. Patients of different types have personalized targets, which can improve the drug effectiveness response. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Schematic diagram of the process of tumor classification and potential target prediction based on personalized metabolic model.

[0031] Figure 2 Schematic diagram of molecular typing and survival analysis results of hepatocellular carcinoma samples. DETAILED DESCRIPTION

[0032] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0033] See also Figure 1 and Figure 2 The method for tumor classification and potential target prediction based on the personalized metabolic model of the present invention comprises the following steps:

[0034] Step 1: Construct a personalized metabolic network model for cancer patients;

[0035] Step 2: Perform precise molecular typing based on the metabolic characteristics of the tumor patient's individualized metabolic network model described in step 1;

[0036] Step 3: Based on the personalized metabolic network model of tumor patients described in step 1, genes that have important effects on tumor cell growth are predicted as potential targets.

[0037] 1. Constructing a personalized metabolic network model for cancer patients

[0038] The original metabolic network model used for the construction of personalized metabolic network needs to cover all known biochemical reactions and pathways as much as possible. The latest human genome-scale metabolic network model Human-GEM model is selected, integrated with cancer transcriptome or proteome expression data from databases such as TCGA or a certain research test, and the tINIT algorithm is used to build a personalized model.

[0039] First, the transcriptome expression data or proteome expression data of the patient are sorted by expression level and converted into categorized data of four conditions: high, medium, low, and no expression. Then, the expression spectrum data is integrated with the Human-GEM model using the tINIT algorithm. The core of the algorithm is to delete inactive genes and reactions in the model according to the expression of genes or proteins, and set the metabolic function tasks that the generated individualized model needs to meet. If the deleted genes and reactions will cause the failure of the metabolic function tasks, these genes and reactions must be retained in the individualized model. For each patient's expression spectrum data of cancer and adjacent tissues, metabolic models of specific cancer cells and adjacent cells are constructed according to the above method.

[0040] 2. Molecular typing based on metabolic change characteristics of individualized metabolic models

[0041] First, based on the metabolic flux information obtained from the simulation, tumor cells and normal cells were hierarchically clustered to identify key metabolic features that can distinguish tumors from normal cells. Then, all tumor models were hierarchically clustered or k-means clustered, and the Kaplan-Meie survival analysis test was performed using clinical data to obtain molecular subtypes with significant survival differences, and metabolic reactions and metabolites with significant differences between different subtypes were found. Figure 2 Shown are the results of classification and survival analysis using hepatocellular carcinoma as an example.

[0042] Then, potential metabolic markers are further screened from the exchange metabolites. These metabolites can be detected extracellularly and are relatively easier to detect clinically. The orthogonal partial least squares method OPLS is used to analyze the exchange reactions between tumor samples and adjacent samples, determine the metabolites of this part of the reaction exchange, and select the exchange metabolites with variable projection importance VIP>1 or 2 (depending on the specific tumor data) as markers to distinguish cancer from adjacent samples. OPLS analysis and VIP value screening are then performed on cancer samples of different types to obtain prognostic markers that significantly distinguish different survival patients.

[0043] 3. Predict genes that have important effects on tumor cell growth as potential targets based on personalized metabolic models

[0044] For the metabolic network model of the patient's individualized tumor tissue, tumor cell growth is taken as the optimal objective function, and the flow balance analysis is performed to obtain the original growth rate grWT, as shown in formula (1).

[0045] subject to: S·v = 0 (1)

[0046] a j ≤ v ≤ b j

[0047] S represents the stoichiometric matrix, V represents the metabolic flux of all reactions, that is, the reaction rate, S·v satisfies the conservation of mass, a j and b j Represent the upper and lower constraints of the reaction flow rate respectively.

[0048] Then all genes were knocked out one by one to obtain the growth rate after knockout grKO, and the growth rate change of tumor cells was calculated according to formula (2):

[0049] grRatio=grWT / grKO (2)

[0050] Here, the range of grRatio is [0,1]. Usually, when the ratio grRatio <= 0.5, the gene is considered to be lethal to tumor cells and can significantly inhibit the growth of tumor cells. At the same time, in order to avoid damage to normal cells, the candidate gene is not required to be an essential gene for growth in the normal cell metabolism model, and the candidate gene is required to appear in at least 1 / 3 of the samples, which can be used as a potential therapeutic target. The simulation of gene combination knockout is similar to single gene knockout, that is, all pairwise gene combination knockouts are traversed to obtain the change ratio of tumor cell growth.

Claims

1. A method for tumor classification and potential target prediction based on an individualized metabolic model, characterized in that: The steps are as follows: Step 1: Construct a personalized metabolic network model for cancer patients; Step 2: Perform precise molecular typing based on the metabolic characteristics of the tumor patient's individualized metabolic network model described in step 1; Step 3: Based on the tumor patient's individualized metabolic network model described in step 1, genes that have important effects on tumor cell growth are predicted as potential targets; The steps 2 and 3 can be run separately or simultaneously; The specific method for constructing the personalized metabolic network model of tumor patients in the step 1 is: selecting the human genome-scale metabolic network model Human-GEM as a general model, sorting the transcriptome expression data or proteome expression data of the patient according to the expression amount, and dividing them into high, medium, low, and no expression spectrum data, integrating the expression spectrum data with the Human-GEM model using the tINIT algorithm, and setting the constraints of the metabolic reaction flow according to the expression situation, so as to obtain the personalized metabolic network model of tumor patients; The patient's transcriptome expression data is divided into a tumor transcriptome corresponding to the tumor tissue and a normal transcriptome corresponding to the paracancerous tissue; the proteome expression data is divided into a tumor proteome corresponding to the tumor tissue and a normal proteome corresponding to the paracancerous tissue; the tumor patient's personalized metabolic network model is divided into a personalized normal metabolic model and a personalized tumor metabolic model; The specific method for performing accurate molecular typing based on the metabolic characteristics of the individualized metabolic network model of tumor patients described in step 1 in step 2 is as follows: first, hierarchical clustering is performed on tumor cells and normal cells based on the metabolic flux information obtained by simulation to identify key metabolic characteristics that can distinguish tumors from normal cells, and then hierarchical clustering or k-means clustering is performed on all tumor models, and a Kaplan-Meie survival analysis test is performed using clinical data to obtain molecular typing with significant survival differences, and to discover metabolic reactions and metabolites that are significantly different between different subtypes; The survival analysis is a test for survival analysis using clinical data to obtain molecular typing with significant survival differences and to discover metabolic reactions and metabolites with significant differences between different subtypes; The specific method of predicting potential targets based on the individualized metabolic network model of tumor patients in step 1 in step 3 is: for the constructed metabolic network model of the patient's individualized tumor tissue, taking tumor cell growth as the optimal objective function, a flow balance analysis is performed to obtain the original growth rate grWT, as shown in formula (1): subject to: S·v = 0 (1) a j ≤ v ≤ b j S represents the stoichiometric matrix, V represents the metabolic flux of all reactions, that is, the reaction rate, S·v satisfies the conservation of mass, a j and b j represent the upper and lower constraints of the reaction flow rate respectively; Then all genes were knocked out one by one to obtain the growth rate after knockout grKO, and the growth rate change of tumor cells was calculated according to formula (2): grRatio=grWT / grKO (2) Here, the range of grRatio is [0,1]. When the ratio grRatio <= 0.5, the gene is considered to be lethal to tumor cells and can significantly inhibit the growth of tumor cells. At the same time, in order to avoid damage to normal cells, the candidate gene cannot be an essential gene for growth in the normal cell metabolism model, and the candidate gene is required to appear in at least 1 / 3 of the samples, then it can be used as a potential therapeutic target. For the simulation of gene combination knockout and single gene knockout, all pairwise gene combination knockouts are traversed to obtain the change ratio of tumor cell growth.

2. The method for tumor classification and potential target prediction based on an individualized metabolic model according to claim 1, characterized in that: The tumor is liver cancer.

Images