A method for screening a triple negative breast cancer prognosis gene marker

Abstract

The application discloses a method for screening a triple-negative breast cancer prognosis gene marker, uses gene expression data of triple-negative breast cancer, obtains genes interacting with a cell proliferation marker, and screens prognosis-related genes by means of single-factor Cox regression, and the application uses a Cox regression method with minimum maximum concave penalty to construct a prognosis risk assessment model, and obtains a prognosis gene marker and a prognosis assessment method capable of predicting and assessing the risk of triple-negative breast cancer. The application proposes a prognosis marker gene identification technology for triple-negative breast cancer which is a high-heterogeneity disease, and the technology can be applied and popularized in clinical treatment.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of high-throughput technology and screening of tumor prognosis risk gene markers, and aims to screen prognosis gene markers and distinguish the stratification effect of patient prediction, and specifically relates to a method for screening prognosis gene markers of triple-negative breast cancer. BACKGROUND

[0002] Breast cancer is one of the malignant tumors that endanger women's health, and its new cases and mortality rates are among the top, with high incidence, unknown etiology, no obvious early symptoms, metastasis ability, and poor prognosis for patients with distant metastasis during diagnosis. It has a bad influence on women's life and health, family economic burden, and social economy. Breast cancer is a malignant disease with high heterogeneity, and its molecular typing includes: luminal A type (Luminal A), luminal B type (Luminal B), Her-2 overexpression type (Her2-overexpression), basal-like type (Basal-like, BLBC), and normal-like type (Normal-like). Most of the basal-like type is triple-negative breast cancer (Triple-negative breast cancer, TNBC), and accounts for 10%-20% of breast cancer patients. The marker identification of triple-negative breast cancer is the lack of estrogen receptor (estrogen receptor, ER), progesterone receptor (progesterone receptor, PR), and human epidermal growth factor receptor-2 (Human epidermal growth factor receptor-2, Her2). At present, the treatment of breast cancer mostly adopts traditional conservative treatment, and early breast cancer usually adopts breast-conserving surgery and radiotherapy, while advanced breast cancer is more invasive and needs to resect the breast after induction chemotherapy, but the tumor is prone to distant metastasis and recurrence. Therefore, it is very meaningful to study the internal molecular mechanism of triple-negative breast cancer, and accurate identification of the subtype of triple-negative breast cancer has very important research value and application prospect for further understanding the biological characteristics, clinical manifestations and individualized treatment of triple-negative breast cancer.

[0003] KI67 is a molecular marker for evaluating tumor proliferation, which is closely related to cell cycle progression: it accumulates in cells during mitosis S, G2 and M, and KI67 is also used in breast cancer subtype division, indicating the importance of cell proliferation indicators as prognostic and predictive markers. Since the evaluation of positive KI67 is based on the evaluation of experienced clinical experts and given the corresponding percentage, but the empirical evaluation still has some bias, the gene MKI67 expressing KI67 is used to construct a molecular typing index and screen prognostic gene markers, and the disease is typed by genes that characterize cell expansion capacity. In clinical practice, patient stratification and marker screening are conducive to further understanding of the disease and exploring the "classification and treatment" scheme for patients.

[0004] The interaction database (Biological General Repository for Interactionh Datasets, BioGRID: https: / / thebiogrid.org) mainly records and sorts the data of protein, genetic and chemical interactions, and stores a large number of reliable known interactions, which can be used to construct complex networks and assist medical workers in studying human health and diseases. All the contents contained in the database are supported by experimental evidence of biomedical literature. The genes screened by the present application for interacting with cell proliferation markers are derived from the public database.

[0005] With the rapid development of high-throughput genome sequencing technology, the sequencing cost is decreasing, while the sequencing depth and sequencing quality are constantly improving. Whole genome sequencing provides effective information for exploring the occurrence and development mechanism of diseases from the perspective of genetic information. With the help of bioinformatics means, the biological code contained in big data is analyzed and converted into reliable results with clinical guidance value. The mining and application of bioinformatics data have a prominent role in the treatment of diseases.

[0006] The Cox regression model is mainly used for prognosis analysis of tumors and chronic diseases, and can be used for the influence of continuous independent factors on survival time and state. The purpose of single-factor Cox regression analysis of multiple factors one by one is to screen a subset of variable factors related to the target variable. In the screening process, single-factor Cox regression can analyze data with censored survival time, and survival data is not controlled by the survival distribution type of the estimated data. Based on the above characteristics, Cox regression model is widely used in medical analysis field.

[0007] Feature selection is used to evaluate the correlation between input variables and target variables for screening variables and selecting the most relevant subset of predictive variables. LASSO regression is considered as the most commonly used variable selection method to remove the problem of multicollinearity between variables, which compresses the coefficients with smaller absolute values in the model to 0, so as to achieve the purpose of parameter evaluation and variable selection, but the defect is that the coefficients with larger absolute values are also compressed in the compression process, which causes the deviation of the model. The smooth slice absolute deviation penalty SCAD function solves this problem, and the MM algorithm is used to solve the problem that the variables screened out of the model cannot enter. The minimum maximum concave penalty model, namely MCP, is a simplified model based on SCAD but retains its characteristics for variable screening. SUMMARY

[0008] In view of the above research status, the purpose of the present application is to provide a classification method and therapeutic marker for diseases with heterogeneity, for prognosis stratification and prognosis gene marker screening in triple negative breast cancer patients with high heterogeneity, to solve the problem of difficult treatment due to unclear target, to divide subgroups and precise target to improve treatment effectiveness, and to reduce overtreatment.

[0009] The present application provides a method for screening prognosis gene markers of triple negative breast cancer, which specifically comprises the following steps:

[0010] Step one, calculate the expression amount of all genes of the target sample from the sequencing data of the sample, filter the genes with low expression amount, and add the expression amount of each different transcript of the same gene name to represent the expression amount of the gene.

[0011] Step two, obtain genes interacting with the cell proliferation marker MKI67 gene, and screen out genes possibly related to survival prognosis as a target gene set by using the single factor Cox regression method, wherein the genes interacting with the proliferation marker are obtained from the biological database BioGRID.

[0012] Step three, for further screening of more core genes related to prognosis, the Cox regression model based on MCP is used for the target gene set obtained in step two, and the risk coefficient corresponding to the prognosis gene marker is also obtained from the model, and the prognosis risk evaluation formula is constructed by using the expression amount data of the prognosis gene marker and the coefficient of the prognosis gene marker.

[0013] The prognosis risk evaluation formula is used to calculate the prognosis risk score of the patient, the patient is divided into a low-risk prognosis group and a high-risk prognosis group according to the prognosis risk score, and the effectiveness of screening the prognosis gene marker is verified by survival analysis.

[0014] Preferably, the calculation of the gene expression quantity in step one is first to align the RNA-seq data to the human reference genome by using BWA, then sambamba is used to filter the repeated, unaligned and multiple aligned fragments, and then the featurecounts tool is used for the quantification of the transcriptome, and the FPKM value of the gene is calculated by using the exonsBy function in the GenomicFeatures package to calculate the length of the gene, so as to represent the expression level of each gene.

[0015] Preferably, in the single factor Cox regression in step two, the survival data of each target sample is obtained, the survival data includes the survival time and the survival state corresponding to each sample, the P value calculation method of the regression analysis is selected as likelihood-p screening, and the threshold is set as P<0.05.

[0016] Preferably, in step three, the genes having influence on the prognosis are further screened, and the weight of each gene in the prognosis is obtained, the minimum maximum concave penalty is used to compress the coefficient by using the cv.ncvsurv function in the ncvreg package, and the parameter is selected as the MCP penalty.

[0017] Preferably, in step four, the risk score of each sample is calculated, and the prognosis risk group of the patient is divided, the product sum of the penalty coefficient and the corresponding gene expression quantity of each sample is used to construct a prognosis risk evaluation formula, the best cut-off value is obtained by using the surv_cutpoint function in the survival package, the samples are stratified by using the best cut-off value, wherein the samples with a prognosis risk score less than or equal to the best cut-off value are divided into a low-risk prognosis group, and the samples with a prognosis risk score greater than the best cut-off value are divided into a high-risk prognosis group, the survival package is used to compare the survival state difference between the two groups by using the survfit function and the ggsurvplot function, and the P value obtained is less than 0.05, indicating that there is a significant difference between the two groups.

[0018] The present application has the following effects:

[0019] The present application identifies markers related to cell proliferation and prognosis of triple negative breast cancer by using transcriptome sequencing data of triple negative breast cancer, and uses the markers for prognosis stratification, and verifies the effectiveness of prognosis stratification by survival analysis; the present application proposes an innovation point, that is, to study genes that meet both the interaction with cell proliferation markers and the prognosis related two conditions, and to use them for stratification of patients after being determined as triple negative breast cancer prognosis markers, the proposed stratification index further enriches the new ideas for classification and treatment in clinic; at the same time, since the number of genes used to construct the prognosis risk assessment formula is relatively small, the cost of the detection stage is reduced, which can be further popularized in clinical application, and also becomes an auxiliary means for targeted treatment of low-risk group and high-risk group in clinic; the finally identified prognosis gene marker is a gene related to cell proliferation ability, which can assist researchers to develop new drugs to control cancer cell proliferation for the treatment of TNBC patients, and at the same time, the method can be expanded to the research of other malignant tumors, and provides a new direction for the drug research and development of malignant tumors. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 Flowchart of triple negative breast cancer prognosis gene marker screening method;

[0021] Figure 2 Specific implementation flowchart of triple negative breast cancer prognosis gene marker screening and prognosis stratification;

[0022] Figure 3 Cross-validation for MCP to construct a prognosis risk assessment formula;

[0023] Figure 4 Density distribution diagram of each level of prognosis stratification sample and selection statistical diagram of optimal cutoff value;

[0024] Figure 5 Kaplan-Meier survival analysis diagram for high-risk group and low-risk group of prognosis. DETAILED DESCRIPTION

[0025] In order to solve the problem that the treatment effect is poor due to the heterogeneity among patients and the inaccuracy of targets in the treatment process of malignant tumors, the idea of classifying and treating and precise treatment by evaluating and stratifying the prognosis of patients and exploring the characteristics of each level is used to solve the technical problem that the traditional group treatment mode leads to poor prognosis. The embodiment of the present application provides a method for screening a triple negative breast cancer prognosis gene marker and disease stratification, which comprises the following steps: screening a gene set related to the prognosis of triple negative breast cancer and the cell proliferation capacity; constructing a prognosis risk assessment formula according to the target gene set screened and obtaining the prognosis risk score of each patient, and the prognosis risk score is used for stratifying the triple negative breast cancer patients into a high-risk prognosis group and a low-risk prognosis group. The specific process diagram of the overall scheme is shown in Figure 1 .

[0026] The technical scheme of the present application is further described in detail by means of the accompanying drawings:

[0027] Figure 1 is a flowchart of a triple negative breast cancer prognosis gene marker screening method, mainly including obtaining gene expression, obtaining cell amplification marker interaction genes, screening prognosis gene markers, constructing a prognosis risk assessment formula based on the Cox regression model of MCP and calculating the prognosis risk score and the prognosis stratification of patients.

[0028] Figure 2 is a detailed display of the implementation process of prognosis stratification and prognosis marker screening using sequencing data of triple negative breast cancer, including calculating gene expression from sequencing data, obtaining cell amplification marker interaction genes, screening genes with prognostic effects from cell amplification marker interaction genes, prognosis stratification and verifying the effectiveness of the classification formula.

[0029] Figure 3 is the number of variables and the corresponding model obtained by cross-validation to construct a prognosis risk assessment model, the horizontal axis is the number of variables and the lambda value included in each prognosis risk assessment model, and the vertical axis is the error rate of cross-validation. The model with the minimum error rate is the optimal prognosis risk assessment model for constructing the prognosis risk score formula.

[0030] Figure 4is a schematic diagram of distribution of prognosis risk score and the calculation process of optimal cutoff value, wherein the abscissa of the upper graph represents the continuous variable prognosis risk score, the ordinate represents the density distribution of the sample, the optimal cutoff value divides the sample to obtain the sample density distribution in each prognosis risk score interval, and the sample size difference in the two groups of prognosis stratification is also shown; the abscissa of the lower graph is still the prognosis risk score, and the ordinate is the calculated standardized statistic value. The algorithm is to set each prognosis risk score as a threshold to divide groups, and calculate a standardized statistic value through the two groups divided, the statistic value represents the difference between the two groups, and finally the corresponding standardized statistic value calculated after each prognosis risk score is set as a threshold to divide groups is obtained, wherein the maximum statistic value is the optimal cutoff value, and thus the sample classification value is obtained.

[0031] Figure 5 is a Kaplan-Meier survival analysis curve for stratifying patients according to prognosis risk score, wherein the abscissa is survival time, and the ordinate is survival probability. The group with slow decline in the curve is the group with longer survival period and higher survival rate, indicating that the prognosis risk of the group is low; the group with greater decline in the curve is the group with shorter survival period and lower survival rate, indicating that the prognosis risk of the group is high.

[0032] The present embodiment provides a method for screening prognosis gene markers of triple negative breast cancer, and finally obtains prognosis gene markers for constructing a TNBC prognosis risk assessment formula, which are eight genes ATRIP, KNOP1, LDHA, METTL13, PAPD5, PAXBP1, PTCD3 and U2AF2. The present embodiment is carried out under the framework of the overall process, and specific details are shown in Figure 2 The main methods include:

[0033] 1. Calculate gene expression

[0034] In this step, before screening the genes required for constructing the prognosis risk assessment formula, the gene expression data information of the patient population needs to be calculated first, and the specific calculation method is as follows:

[0035] All the RNA-seq sequences of the triple negative breast cancer samples were aligned to the human reference genome using the MEM algorithm of the Burrows-Wheeler Aligner (BWA) and then converted from SAM format to BAM format using the samtools tool. The RNA-seq sequence data of the samples were downloaded from NCBI (https: / / www.ncbi.nlm.nih.gov / sra), and the data set included 504 TNBC patients who received breast surgery at Fudan University Shanghai Cancer Center (FUSCC) from July 2007 to December 2014. The samples consisted of primary tumor tissues and blood samples. In addition to the sequencing data, the data containing clinical information were also downloaded. The clinical follow-up data related to 352 pieces of human information matched with the whole transcriptome sequencing data, including: recurrence-free survival time, recurrence-free survival status, tumor size, surgical treatment method, and the number of positive lymph nodes, and other important effective statistical data. The FASTA file of the human reference genome sequence and the human genome annotation GTF file used for alignment were also from NCBI, and the hg38 reference genome was selected.

[0036] After alignment to the reference genome, unaligned sequences, multiple alignment sequences, and repetitive sequences need to be filtered, which is achieved using sambamba. Next, the featurecount tool is used to calculate the number of fragments aligned to each corresponding gene, and the featurecount calculation program is derived from the subread program package.

[0037] Finally, the FPKM normalization processing of the obtained fragments count data is performed to obtain the quantitative data of all sequencing coverage genes of each patient sample. The specific calculation process is as follows:

[0038] According to the annotation file, the exonsBy function in the GenomicFeatures package in the programming language R is used to calculate the start and end positions of all exons on each gene, so as to obtain the length of all exons.

[0039] Finally, the normalized processing result of the gene is obtained according to the calculation formula of FPKM. The FPKM value of each gene of each sample is used to represent the gene expression data for subsequent analysis, so as to eliminate the influence of gene length and sequencing depth. FPKM is defined as the number of fragments per thousand bases per million mapped reads. According to the statistical fragment matrix, gene length, and total number of fragments aligned to the genome, the normalized gene expression level is calculated, and the calculation is shown in Formula 1:

[0040]

[0041] Wherein F represents the total number of fragments matched to the gene; L represents the total length of all exons of the gene; N represents the total number of fragments matched to the genome.

[0042] After calculating the expression of all genes of all samples, in order to reduce the difference of gene expression between samples but need to keep the relative distance of expression value between genes unchanged, the expression is further standardized.

[0043] Specifically, the expression of all samples is logarithmically processed with base 2 to reduce the expression value to a certain range. Before that, in order to avoid the phenomenon of negative infinity and negative number when taking logarithm of the genes that are not expressed or rarely expressed in some patients, all gene expression values of all samples are uniformly processed by adding 1 without affecting the overall distribution analysis of expression. Further processing is the standardization of the same gene between samples, that is, the same gene of all samples is a set of data. The expression of the gene of each patient in the set is subtracted from the average of the expression values in the set, and the result is divided by the standard deviation of the expression values in the set. The calculation is shown in Formula Two. The scale function in R language is selected for calculation, which is a method of dividing by standard deviation after numerical centralization. Finally, a set of gene expression information with values near 0 and variance of 1 is obtained to scale the data.

[0044]

[0045] Wherein, FPKM ni is the standardized FPKM value of the gene, FPKM i is the FPKM value of the gene of the patient, mean(FPKM) is the average of the FPKM values of the gene in all patients, and sd(FPKM) is the standard deviation calculated using the FPKM values of the gene of all patients.

[0046] 2、Obtain cell proliferation marker interaction gene

[0047] After obtaining the required standardized expression information and follow-up information of TNBC patients, further extract genes related to cell proliferation markers. One of the indicators in tumor index detection is Ki67, which can be used to identify the proportion of cells in the division period. The sample with high Ki67 labeling rate has relatively strong tumor cell proliferation ability and relatively poor prognosis. The gene MKI67 expressing the antigen and its related genes are considered to contribute to the prognosis of tumor. Therefore, the expression data of cell proliferation marker interaction genes are obtained for further analysis, as follows:

[0048] The gene set list interacting with the MKI67 gene was downloaded using the BioGRID database, the list with gene names was picked out, and the R language basic command was used to exclude repeated genes to form a new gene set with unique gene name for each gene name, and then the obtained gene set was matched with the gene name in the FUSCC data set. The genes not appearing in the interaction gene set were combined with the alias of the gene queried on the NCBI website. The gene name in the FUSCC TNBC data set was matched again through the alias. Thus, all genes interacting with the amplification marker MKI67 were screened out. Finally, 476 genes interacting with the gene MKI67 were obtained, and the amplification marker MKI67 gene was used for the next step analysis, and there were 477 genes.

[0049] Finally, the gene expression amount information of the genes interacting with the amplification marker was extracted from the expression data set of each TNBC patient for subsequent analysis.

[0050] 3、Screening of genes interacting with cell markers and having prognostic effect

[0051] After obtaining the gene set interacting with the MKI67 gene, genes satisfying both the interaction with the MKI67 gene and the correlation with prognosis were screened out for feature selection analysis to screen out genes contributing to prognosis. In this case, the single factor Cox regression method was used to select the prognosis genes by using the single factor Cox regression model.

[0052] Specifically, the survival information was used to construct a survival data object by using the Surv function in the survival package, wherein the survival information includes: the survival time and survival state of each TNBC patient in clinical follow-up. The survival time is the relapse-free survival (RFS), which is defined as the time from complete remission of the patient to the first local, regional or distant invasive metastasis recurrence, death for any reason or follow-up cutoff. The survival state consists of numerical values "0" and "1", i.e. the sample marked as 0 is the patient still alive at the end of follow-up, and the sample marked as 1 is the patient who has not survived before the end of follow-up.

[0053] Then, combined with the constructed survival data object, the coxph function in the survival package was used for regression analysis, the genes interacting with the amplification marker MKI67 gene were taken as a factor into the regression model for fitting, and the calculated likelihood-p corresponding to each gene was extracted.

[0054] Finally, the genes that affect the prognosis of TNBC are selected by the first screening condition, and thus 32 genes related to prognosis are obtained in the FUSCC TNBC dataset as shown in Table 1, wherein the first screening condition is set as P < 0.05, that is, P < 0.05 is statistically significant. The names of the 32 genes related to prognosis and interacting with the MKI67 gene are as follows:

[0055] Table 1 Genes related to the amplification marker MKI67 gene interaction and prognosis in FUSCC TNBC

[0056]

[0057] 4, patient prognosis stratification

[0058] According to the obtained amplification marker interaction genes and prognosis-related genes, the next step is to construct a prognosis risk assessment formula for the gene set, and calculate the prognosis risk score for each sample to evaluate the effectiveness of the screened prognostic gene markers. In this case, the cv.ncvsurv function in the R package ncvreg is selected, and the model penalty method is selected as the minimum maximum concave penalty MCP method. Specifically as follows:

[0059] Here, the expression data of all samples interacting with the MKI67 gene and having a prognosis contribution, that is, the expression of all samples corresponding to the 32 genes, and the survival data object constructed by all samples are obtained, and the expression level of the gene and the survival data object satisfy the required data set of the minimum maximum concave penalty Cox regression model.

[0060] Next, the MCP-based Cox regression model constructs a prognosis risk assessment formula for the genes interacting with MKI67 and related to prognosis. Specifically, the cv.ncvsurv function in the ncvsurv package is used to construct a prognosis risk assessment model for the selected gene expression level and survival data object. To solve the problem of large coefficients being compressed in the coefficient compression process, the penalty parameter adopts the MCP method. As shown in Figure 3 To evaluate the prediction accuracy of the model, cross-validation is performed.

[0061] With the shrinkage of the MCP Cox regression model to each factor coefficient, the gene coefficient with a prominent prognosis contribution is retained, and the factor with a coefficient shrinkage to 0 is excluded from the construction of the prognosis risk assessment model. Finally, 8 prognostic gene markers are retained for constructing a prognosis risk assessment formula, which are ATRIP, KNOP1, LDHA, METTL13, PAPD5, PAXBP1, PTCD3, and U2AF2.

[0062] Finally, the eight genes were used to construct a prognostic risk assessment formula. The expression levels of the eight prognostic gene markers in all samples were matched, and the coefficients of the prognostic gene markers calculated by the Cox regression model with the minimum-maximum concave penalty were used to represent the proportion in the construction formula. The gene expression information and the coefficients were multiplied to calculate the matrix, i.e., the coefficient of each gene in each sample was multiplied by the expression level of the gene, and then the products of each gene were added to obtain the prognostic risk score of each patient. The specific prognostic risk score calculation is shown in Formula Three:

[0063] riskscore = (-0.08066567) * ATRIP + 0.62438558 * KNOP1 + 0.42998045 * LDHA + (-0.64895377) * METTL13 + (0.02456758) * PAPD5 + 0.01029721 * PAXBP1 + 0.35394749 * PTCD3 + (-0.36885638) * U2AF2 Formula Three

[0064] wherein ATRIP, KNOP1, LDHA, METTL13, PAPD5, PAXBP1, PTCD3, and U2AF2 represent the expression levels of the genes. The above genes and their functions are shown in Table 2:

[0065] Table 2: Prognostic risk genes and their functions

[0066]

[0067] Finally, the prognostic risk scores were used to stratify patients, and the samples were divided by the optimal cutoff value. The samples with a prognostic risk score ≤ the optimal cutoff value were classified as a low-risk group, and the samples with a prognostic risk score > the optimal cutoff value were classified as a high-risk group. The optimal cutoff value was determined using the survival package function surv_cutpoint, as follows:

[0068] Using the survival status, survival time, and prognostic risk score data of the FUSCC TNBC patients, the optimal threshold was calculated to be 0.4857975 using the surv_cutpoint function. The sample was stratified by the optimal threshold, and the number distribution of the high-risk group and the low-risk group is shown in Table 3: Figure 4As shown in FIG. 6, the number of samples in the high-risk group is less than that in the low-risk group; then, the survival_categorize function in the survival package is used to divide each sample into a high-risk group and a low-risk group, which are marked as "high" and "low", respectively, and the optimal threshold is used as a cutoff value. The samples with a prognosis evaluation score less than or equal to the optimal threshold are marked as a low-risk group, and the samples with a prognosis evaluation score greater than the optimal threshold are marked as a high-risk group. Thus, 255 patients in the FUSCC TNBC patient group are classified into a low-risk group, and the remaining 97 patients are classified into a high-risk group.

[0069] 5. Survival analysis verifies the effectiveness of the classification formula and the prognosis gene marker

[0070] Finally, to verify the accuracy of the prognosis risk evaluation formula in sample classification and prognosis gene marker screening, the survival package and the survminer package are used to perform Kaplan-Meier survival analysis on the high-risk and low-risk groups to observe whether the calculated prognosis risk score can separate the samples.

[0071] Here, first, the survival time and the survival status of the data set are used to construct a survival data object using the survfit and Surv functions, wherein the prognosis risk grouping is used as a classification variable; then, the ggsurvplot function is used to visualize the survival data object, as shown in FIG. 7. Figure 5 As shown in FIG. 7, the horizontal axis of the survival curve represents the observation time, and the vertical axis represents the survival rate. According to the analysis of a large number of samples in this embodiment, the mortality rate of the low-risk group is lower than that of the high-risk group at any time, which is consistent with the concept of the calculated prognosis risk score. The survival rate of the sample group with a higher prognosis risk score is also poor, and the survival curve does not cross, indicating a good classification effect. The P value is much less than 0.05, indicating that the difference between the two survival curves is statistically significant due to the grouping based on the constructed prognosis risk score. The prognosis risk score has a good classification effect and can well stratify patients. It is proved that the prognosis gene marker of triple-negative breast cancer selected by the present application is very effective, which is conducive to further research on the prognosis gene markers selected from the two levels of patients to assist in the treatment and diagnosis of TNBC. The P value obtained by the survival analysis herein is used as the second determination condition. When P < 0.05, it can be concluded that the classifier can separate the samples based on the prognosis risk score.

Claims

1. A method for screening prognostic gene markers for triple-negative breast cancer, characterized in that, Includes the following steps: (1) Calculation of gene expression levels Gene expression levels in triple-negative breast cancer patient samples were characterized using the MEM algorithm of sequence alignment BWA and FPKM normalization of each transcript. Logarithmic transformation and standardization were performed on these gene expression levels, and genes with low expression levels in all samples were removed. (2) Screening of genes that amplify biomarkers and interact with each other Genes that interact with common amplification markers were selected, with the MKI67 gene chosen as the amplification marker. Other genes that interact with this marker were obtained by querying the database, and the expression level data of these genes were extracted to complete the preliminary screening. The screening of genes that interact with MKI67 was conducted using the BioGRID biological database. (3) Screening of prognostic gene markers Univariate Cox regression was selected as the univariate feature selection method for screening prognostic genes. Univariate Cox regression analysis was implemented using the `coxph` function in the `survival` package of the R language, with the p-value of genes selected using `likelihood-p`. A prognostic risk assessment model was constructed using a minimum-maximum concavity penalty function to further screen prognostic genes selected by univariate Cox regression. The minimum-maximum concavity penalty model was implemented using the `cv.ncvsurv` function in the `ncvreg` package of the R language. The minimum-maximum concavity penalty model simultaneously yielded prognostic gene biomarkers and their coefficients in the prognostic risk assessment model. The validation method for prognostic gene biomarkers involves constructing a prognostic risk assessment formula by summing the products of the prognostic gene biomarkers and their corresponding gene expression levels within the sample. This formula is used to calculate prognostic risk scores, and each patient's score is assessed to determine if it exceeds the optimal stratification cutoff value. Patients are then stratified prognostively using the optimal cutoff value method. Patients with prognostic risk scores less than or equal to the optimal cutoff value are classified as low-risk, while those with scores greater than the optimal cutoff value are classified as high-risk. Survival analysis is performed based on the high- and low-risk group classifications. A p-value less than 0.05 indicates that the selected prognostic gene biomarkers are effective and prognostic. The optimal cutoff value is implemented using the `surv_cutpoint` function in the `survival` package. Transcriptome sequencing data from triple-negative breast cancer identified biomarkers related to cell proliferation and prognosis, and these biomarkers were used for prognostic stratification. The effectiveness of the prognostic stratification was validated through survival analysis.

2. The method for screening prognostic gene markers for triple-negative breast cancer according to claim 1, characterized in that, In step (1), the genes with low expression levels are defined as those whose gene expression level is 0, which account for 30% of all samples. Genes with low expression levels will be removed.

3. The method for screening prognostic gene markers for triple-negative breast cancer according to claim 1, characterized in that, In step (2), the amplification marker was selected as the MKI67 gene that expresses the proliferating cell antigen KI67. KI67 is usually used to assess the proliferation of tumor cells. The screening of genes that interact with MKI67 was conducted using the BioGRID database.

4. The method for screening prognostic gene markers for triple-negative breast cancer according to claim 1, characterized in that, The univariate Cox regression analysis in step (3) is implemented using the coxph function in the survival package of R language, where the p-value of the gene is screened using likelihood-p.

5. The method for screening prognostic gene markers for triple-negative breast cancer according to claim 1, characterized in that, The prognostic risk assessment model in step (3) is constructed using the cv.ncvsurv function of the ncvreg package in the R language program. The penalty method is the minimum maximum concavity penalty. The minimum maximum concavity penalty is a regression method in which the penalty rate is the same as LASSO at the beginning, but as the absolute value of the coefficient increases, the penalty rate will be slowly relaxed to zero.

6. The method for screening prognostic gene markers for triple-negative breast cancer according to claim 1, characterized in that, In step (3), further screening of prognostic genes is completed. That is, genes with a coefficient of 0 are considered to have a small impact on the prognosis. Genes with a non-zero prognostic gene coefficient are selected as prognostic gene markers, thus completing the screening of prognostic gene markers.

7. The method for screening prognostic gene markers for triple-negative breast cancer according to claim 1, characterized in that, The optimal cutoff value method and survival analysis are implemented by selecting the optimal cutoff value through the surv_cutpoint function in the survival package, and the survfit and ggsurvplot functions in the survminer package are used together to complete the survival verification.

Images