Method of developing risk score model based on predictive markers to assess likelihood of liver cancer recurrence and Liver cancer recurrence prediction system

Abstract

The present invention relates to a method for constructing a risk index model using liver cancer recurrence prediction markers and a liver cancer recurrence prediction system. According to the present invention, a method for constructing a risk index model comprises the steps of: acquiring a plurality of genomic data sets including genomic expression profiles by utilizing clinical information of a patient; selecting a plurality of common genes that show a common difference in expression between a relapse group and a non-relapse group from the plurality of genomic data sets; inputting the selected common genes into a regression analysis algorithm to select a plurality of genes significant as independent variables for relapse and calculating regression coefficients for each gene; and developing a risk index (RS) model based on the selected plurality of genes and regression coefficients.

Description

Technology Field

[0001] The present invention relates to a method for constructing a risk index model using a liver cancer recurrence prediction marker and a liver cancer recurrence prediction system. More specifically, the invention relates to a method for constructing a risk index model and a liver cancer recurrence prediction system capable of predicting liver cancer recurrence by deriving a gene signature capable of predicting liver cancer recurrence and designing and modeling a risk score (RS) for liver cancer recurrence. Background Technology

[0002] According to a 2021 report by the National Cancer Information Center, the relative survival rate for liver cancer over the past five years was only 37.7%, making it the cancer with the lowest survival rate along with lung cancer (34.7%).

[0003] This is only half the level of the overall cancer survival rate of 70.7%, and even if radioembolization or hepatectomy is performed, the cumulative recurrence rate over two years reaches 50.0% and 58.3%, respectively.

[0004] Therefore, since liver cancer recurrence lowers the success rate of hepatectomy and increases drug resistance, thereby reducing treatment efficacy, it is urgent to understand the mechanisms of recurrence in order to improve the survival rate of liver cancer patients.

[0005] Since there are currently no known RNA markers for recurrence, which are a major factor in the survival rate of liver cancer patients, the discovery of candidate substances related to liver cancer recurrence and research into their mechanisms are absolutely necessary.

[0006] The technology forming the background of the present invention is disclosed in Korean Registered Patent No. 10-2188115 (published Dec. 07, 2020). The problem to be solved

[0007] The technical problem that the present invention aims to solve is to provide a method for constructing a risk score model capable of predicting liver cancer recurrence and a liver cancer recurrence prediction system by deriving a gene signature capable of predicting liver cancer recurrence and designing and modeling a risk score (RS) for liver cancer recurrence. means of solving the problem

[0008] According to an embodiment of the present invention for achieving such technical challenges, a method for constructing a risk index model comprises the steps of: acquiring a plurality of genomic data sets including genomic expression profiles by utilizing clinical information of a patient; selecting a plurality of common genes that show a common difference in expression between a relapse group and a non-relapse group from the plurality of genomic data sets; inputting the selected common genes into a regression analysis algorithm to select a plurality of genes significant as independent variables for relapse status and calculating regression coefficients for each gene; and developing a risk index (RS) model based on the selected plurality of genes and regression coefficients.

[0009] The above plurality of genome data sets may include a genome data set obtained from GSE14520, a genome data set obtained from GSE114564, and a genome data set obtained from TCGA LIHC.

[0010] The step of selecting the above genes involves selecting genes for each genome data set that have a significance probability less than a preset value, and using the selected genes, selecting multiple common genes that appear in common across three genome data sets.

[0011] The step of selecting the above genes can be performed by using the Glmnet package provided by R to set a value of 0 or 1 as the response variable based on whether there is a recurrence, and to set multiple common genes as predictor variables.

[0012] The step of selecting the above genes involves inputting the response variable and the predictor variable into an Elastic Net regression algorithm to allow the Elastic Net regression algorithm to select independent variables that appear significantly in response to cancer recurrence.

[0013] The above risk score can be calculated using the following mathematical formula.

[0014] Risk score = (0.03208 * GULP1) + (-0.04351 * LCAT) + (0.29381 * PPAT) + (-0.05648 * LPXN) + (0.09575 * NOP56) + ( -0.03938 * CD4) + (0.12171 *ZC2HC1A) + (0.42938 * PPIA) + (0.08035 * CST7) + (-0.04075 * PRKCQ) + (0.12525 * PHF20) + (0.03416 * RAB23) + (0.07323 * PCDHB6) + (-0.17598 * CXCR6) + (-0.09678 * SLC4A10)

[0015] It may further include a step of analyzing the predictive performance of the above-developed risk index (RS) model.

[0016] The step of analyzing the predictive performance of the above risk index model may analyze the predictive performance of the risk index model using any one of the following methods: DCA (Decision Curve Analysis), the optimal cutoff value of the risk index (RS), ROC analysis based on whether recurrence occurs with the RS value, and Disease-free survival (DFS).

[0017] In addition, according to an embodiment of the present invention, a risk index model construction system includes an input unit that receives a plurality of genetic information for a target patient, a risk index calculation unit that calculates a risk index of the target patient through a previously constructed risk index model based on the input genetic information, and a result output unit that provides predictive information regarding the recurrence of liver cancer in the target patient according to the risk index. Effects of the invention

[0018] As such, according to the present invention, RNA indicator factors for recurrence, which are a major factor in the survival rate of liver cancer patients, can be extracted, thereby improving the performance of the risk index prediction model. Brief explanation of the drawing

[0019] FIG. 1 is a flowchart of a method for constructing a risk index model according to an embodiment of the present invention. Figure 2 is an example diagram showing the result of Lasso-Cox regression in step S130 shown in Figure 1. Figure 3 shows the cross-validation results. Figure 4 is a graph showing the DCA results of the risk index (RS) model of the present invention and the conventional model. Figure 5 is a graph showing the distribution of recurrences according to risk index in three genomic data sets. Figure 6 is a graph showing the results of ROC analysis based on recurrence using 15 risk indices in three genomic datasets. Figure 7 is a graph showing the results of disease-free survival analysis in three genome datasets. Figure 8 is a graph showing the results of Cox regression analysis for disease-free survival analysis in three genomic datasets. FIG. 9 is a configuration diagram of a liver cancer recurrence prediction system according to another embodiment of the present invention. Specific details for implementing the invention

[0020] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification have been given similar reference numerals.

[0021] Throughout the specification, when a part is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0022] Then, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the invention.

[0023] Hereinafter, a method for constructing a risk index model using a liver cancer recurrence prediction marker according to an embodiment of the present invention will be explained in more detail using FIG. 1.

[0024] FIG. 1 is a flowchart of a method for constructing a risk index model according to an embodiment of the present invention.

[0025] Referring to FIG. 1, first, a plurality of genomic data sets are obtained using gene expression profiles obtained from patients who have undergone hepatectomy (S110). In an embodiment of the present invention, gene expression profile data is obtained for each of three gene data sets (GSE14520, GSE114564, TCGA LIHC) in order to differentiate and analyze patients who have undergone hepatectomy and patients who have not experienced recurrence for more than two years.

[0026] According to an embodiment of the present invention, the genomic data set obtained from GSE14520 includes sample data obtained from 89 non-relapsed patients and 83 relapsed patients, and the genomic data set obtained from GSE114564 includes sample data obtained from 31 non-relapsed patients and 24 relapsed patients. Additionally, the genomic data set obtained from TCGA LIHC includes sample data obtained from 48 non-relapsed patients and 164 relapsed patients.

[0027] Next, common genes are selected from the acquired multiple genome data sets (S120). In an embodiment of the present invention, genes are selected from each genome data set in which a statistically significant difference in gene expression is confirmed in the relapse group compared to the non-relapse group. For example, a significance probability indicating a statistically significant difference in expression ( P Assuming the -value) is 0.05, the present invention selects genes with a significance probability of less than 0.05 among a plurality of genes included in a genome data set.

[0028] Selected genes for each genomic data set are compared and analyzed to select genes that show common expression differences across all data sets. In an embodiment of the present invention, 50 genes that show common expression differences were selected from three genomic data sets, and the selected 50 genes were named as common genes.

[0029] Next, in the present invention, training data is generated, and the generated training set is input into a regression analysis (Net regression) algorithm to select independent variables (S130).

[0030] First, in an embodiment of the present invention, a training set is generated by randomly extracting data corresponding to 80% of the total data from the TCGA LIHC, which has the largest number of samples in the recurrence group.

[0031] Next, using the data included in the generated training set, the response variable, and the predictor variable, an Elastic Net regression analysis that combines Ridge regression (L2) and Lasso regression (L1) is performed to filter and select a list of genes in which each gene is significantly identified as an independent variable for recurrence.

[0032] Here, the response variable represents a value of 0 or 1 depending on whether there is a recurrence, and the predictor variable represents 50 common genes selected at step S120.

[0033] Meanwhile, in an embodiment of the present invention, genes included in a training set are set as a feature set, and an Elastic Net, a normalized regression method, is performed considering overfitting. That is, to reduce multicollinearity, highly correlated duplicate genes are excluded to construct a minimal feature set, thereby increasing the explanatory power of gene markers and enabling efficient universal application to various data sets.

[0034] Figure 2 is an example diagram showing the Lasso-Cox regression result in step S130 shown in Figure 1, and Figure 3 shows the cross-validation result.

[0035] Each line shown in Figure 2 represents a gene (variable), and the Y-axis represents the coefficient of that variable. In this case, the Lasso-Cox regression process adjusts the lambda value to extract the optimal lambda value that fits the model so that the variables are not overfitted.

[0036] To elaborate, as shown in Figure 3, when each variable is applied to the Lasso-Cox regression algorithm, the estimated deviation extracted, i.e., the value on the Y-axis, is used to calculate the optimal lambda value that minimizes that value.

[0037] However, in the present invention, the lambda value for 50 common genes is calculated by applying Elastic Net regression, which is a compromise model of Ridge Regression and Lasso Regression.

[0038] As a result, among the lambda values ​​calculated in the present invention, the value with the smallest error was 0.1829059, and among the 50 common genes, there were 15 genes that could be selected as independent variables.

[0039] The regression coefficients for the 15 genes selected as independent variables are as shown in Table 1 below.

[0040]

[0041] Subsequently, a risk score model is developed based on 15 independent variables (S140).

[0042] First, using the respective regression coefficients corresponding to the 15 independent variables, the following formula for calculating the risk score is developed as a risk score model.

[0043] Risk score = (0.03208 * GULP1) + (-0.04351 * LCAT) + (0.29381 * PPAT) + (-0.05648 * LPXN) + (0.09575 * NOP56) + (-0.03938 * CD4) + (0.12171 *ZC2HC1A) + (0.42938 * PPIA) + (0.08035 * CST7) + (-0.04075 * PRKCQ) + (0.12525 * PHF20) + (0.03416 * RAB23) + (0.07323 * PCDHB6) + (-0.17598 * CXCR6) + (-0.09678 * SLC4A10)

[0044] Subsequently, the predictive performance of the developed risk index (RS) model is analyzed (S150).

[0045] When predicting recurrence using a risk index model and an existing model, Decision Curve Analysis (DCA) is performed to determine which model offers greater benefits.

[0046] Figures 4 to 8 below show the results of comparing and evaluating a risk index (RS) model according to an embodiment of the present invention with a conventional model using the three genome data sets described above.

[0047] Figure 4 is a graph showing the DCA results of the risk index (RS) model of the present invention and a conventional model, and Figure 5 is a graph showing the number of recurrences according to the risk index in three genomic data sets.

[0048] As shown in Figure 5, risk index values ​​were calculated for each sample included in three genomic data sets, and the optimal cutoff value was applied to the calculated risk index to divide them into a high risk group and a low risk group.

[0049] At this time, the optimal cutoff value is obtained using the surv_cutpoint function of the R package "survminer". As a result, it was confirmed that in the TCGA, GSE14520, and GSE114564 datasets, the number of non-recurrence groups was higher as the risk index was lower, and the number of recurrence groups was more distributed as the risk index was higher.

[0050] Figure 6 is a graph showing the results of ROC analysis based on recurrence with 15 risk index values ​​in three genomic datasets.

[0051] As shown in Figure 6, the AUC for GSE14520 was 0.70 (95% CI: 0.62-0.78), the AUC for GSE114564 was 0.85 (95% CI: 0.74-0.95), and the AUC for TCGA was 0.79 (95% CI: 0.73-0.86), all three genome datasets were evaluated to have an accuracy of 0.7 or higher.

[0052] Figure 7 is a graph showing the results of disease-free survival analysis in three genome datasets.

[0053] As shown in Figure 7, disease-free survival (DFS) analysis was performed on three genomic datasets using the Kaplan-Meier Estimation method. Based on the median RS value for all patients, the data were divided into a low RS group (RS_Low) and a high RS group (RS_High). The results showed that all three datasets had a Logrank P-value of 0.001, indicating high significance. For GSE14520, the hazard ratio (HR) in the high RS group was 3.22 (95% CI=2.08-4.98); for GSE114564, the HR was 7.75 (95% CI=3.26-18.44); and for TCGA, the HR was 2.81 (95% CI=2.04-3.88).

[0054] Figure 8 is a graph showing the results of Cox regression analysis for disease-free survival analysis in three genomic datasets.

[0055] As shown in Figure 8, Cox regression analysis of disease-free survival was performed on three datasets, and the risk index (RS) of 15 genes was found to act as a statistically significant independent variable in all three datasets.

[0057] FIG. 9 is a configuration diagram of a liver cancer recurrence prediction system according to another embodiment of the present invention.

[0058] Referring to FIG. 9, the liver cancer recurrence prediction system (900) may include an input unit (910), a risk index calculation unit (920), and a result output unit (930).

[0059] The input unit (910) receives multiple genetic information for a target patient. Then, the risk index calculation unit (920) calculates the risk index of the target patient using a pre-established risk index model based on the input multiple genetic information. The risk index model predicts whether liver cancer will recur based on the risk index calculated using regression coefficients corresponding to the input genetic information.

[0060] Finally, the result output unit (930) transmits prediction information regarding the recurrence of liver cancer in the target patient according to the risk index to the medical professional's terminal, and the medical professional can take appropriate measures using the prediction information regarding the recurrence of liver cancer.

[0061] As such, the method for constructing a risk index model according to the present invention can extract RNA indicator factors for recurrence, which are a major factor in the survival rate of liver cancer patients, and thus can improve the performance of the risk index prediction model.

[0063] The present invention has been described with reference to the embodiments illustrated in the drawings, but this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the present invention should be determined by the technical spirit of the appended claims. Explanation of the symbols

[0064] 900: Liver Cancer Recurrence Prediction System 910 : Input section 920 : Risk Index Calculation Unit 930 : Result output section

Claims

Claim 1 A method for constructing a risk index model performed by a processor comprises: a step of acquiring multiple genomic data sets including genomic expression profiles using clinical information of a patient; a step of selecting multiple common genes that show common expression differences between a relapse group and a non-relapse group from the multiple genomic data sets; a step of inputting the selected common genes into a regression analysis algorithm to select multiple genes significant as independent variables for relapse status and calculating regression coefficients for each gene; and a step of developing a risk index (RS) model based on the selected multiple genes and regression coefficients. The step of selecting common genes comprises selecting genes for each genomic data set whose significance probability is less than a preset value, and using the selected genes to select multiple common genes that appear commonly in the multiple genomic data sets. The step of calculating regression coefficients comprises setting a value of 0 or 1 according to relapse status as a response variable, setting multiple common genes as predictor variables, and inputting the response variable and predictor variables into an Elastic Net regression analysis algorithm to perform the Elastic Net regression analysis. A method for constructing a risk index model, wherein an algorithm selects independent variables that appear significantly in response to cancer recurrence, and the step of developing the risk index (RS) model derives a formula for calculating the risk index (RS) using multiple selected independent variables and regression coefficients for each independent variable. Claim 2 A method for constructing a risk index model according to claim 1, wherein the plurality of genomic data sets include a genomic data set obtained from GSE14520, a genomic data set obtained from GSE114564, and a genomic data set obtained from TCGA LIHC. Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 In claim 1, the above-mentioned risk score is a method for constructing a risk score model calculated through the following mathematical formula. Risk score = (0.03208 * GULP1) + (-0.04351 * LCAT) + (0.29381 * PPAT) + (-0.05648 * LPXN) + (0.09575 * NOP56) + (-0.03938 * CD4) + (0.12171 * ZC2HC1A) + (0.42938 * PPIA) + (0.08035 * CST7) + (-0.04075 * PRKCQ) + (0.12525 * PHF20) + (0.03416 * RAB23) + (0.07323 * PCDHB6) + (-0.17598 * CXCR6) + (-0.09678 * SLC4A10) Claim 7 A method for constructing a risk index model according to claim 1, further comprising the step of analyzing the predictive performance of the developed risk index (RS) model. Claim 8 In claim 7, the step of analyzing the predictive performance of the risk index model is a method for constructing a risk index model that analyzes the predictive performance of the risk index model using one of the following methods: DCA (Decision Curve Analysis), the optimal cutoff value of the risk index (RS), ROC analysis based on whether there is a recurrence with the RS value, and Disease-free survival (DFS). Claim 9 The system comprises an input unit that receives multiple genetic information regarding a target patient, a risk index calculation unit that calculates a risk index of the target patient through a pre-established risk index model based on the input genetic information, and a result output unit that provides predictive information regarding liver cancer recurrence in the target patient according to the risk index. The risk index model is developed by utilizing the patient's clinical information to acquire multiple genomic data sets including genomic expression profiles, selecting multiple common genes that exhibit common expression differences between a recurrence group and a non-recurrence group from the multiple genomic data sets, inputting the selected common genes into a regression analysis algorithm to select multiple genes significant as independent variables for recurrence status and calculating regression coefficients for each gene, and based on the selected multiple genes and regression coefficients, selecting genes for each genomic data set whose significance probability is less than a preset value, selecting multiple common genes that appear commonly in the multiple genomic data sets using the selected genes, setting a value of 0 or 1 based on recurrence status as a response variable, setting multiple common genes as predictor variables, and analyzing the response variable and predictor variables using an elastic net A liver cancer recurrence prediction system that inputs an Elastic Net regression algorithm to select independent variables that appear significantly in response to cancer recurrence, and derives a formula for calculating a risk index (RS) using the selected multiple independent variables and the regression coefficients for each independent variable.

Images