Use of reagents for detecting genomic compositions in the preparation of a product for predicting distal margin length in colorectal cancer patients

Abstract

The application discloses application of a reagent for detecting a gene combination in preparation of a product for predicting a length of a distal incisal edge of a colorectal cancer patient. The application provides genes and clinical features closely related to selection of the length of the distal incisal edge of the colorectal cancer, and a prediction model of the length of the distal incisal edge of the colorectal cancer patient is constructed by combining a machine learning algorithm, with an AUC as high as 0.83 or above, so that a data-driven and structure-defined guidance method for judging the length of the distal incisal edge is provided for clinical implementation of colorectal cancer resection, and has popularization and application values.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of biomedical technology, in particular, to the application of a reagent for detecting a gene combination in the preparation of a product for predicting the length of a distal resection margin of a colorectal cancer patient. BACKGROUND

[0002] Colorectal cancer (CRC) is one of the top three high-incidence malignant tumors in the world, and radical (R0) resection is the current standard treatment for colorectal cancer. In order to reduce the risk of local recurrence and at the same time preserve organ function as much as possible, the nearest distance between the specimen margin after tumor resection and the tumor tissue, i.e. the surgical margin, needs to be determined during radical resection according to the location of the tumor. If the surgical margin is too long, it may not be possible to preserve the function of the anus, and if the surgical margin is too short, it may not be possible to achieve R0 resection of the tumor. However, in clinical practice, many patients are unable to accept the impact of lifelong stoma on quality of life and therefore give up R0 resection surgery. Therefore, a more precise method for controlling the surgical margin is needed to improve the success rate of R0 resection and improve the prognosis of patients with colorectal cancer.

[0003] The surgical margin is divided into a distal resection margin (close to the anus side), a proximal resection margin (close to the colon side), and a circumferential resection margin (the boundary of the soft tissue around the tumor). In current clinical practice, a distal resection margin length of 1-2 cm is considered a safe margin for low rectal cancer radical resection, and 0.5-1 cm is the shortest acceptable distal resection margin length. However, the determination of the length of the surgical margin required for R0 resection usually relies on the experience of the doctor and technical assistance such as imaging and pathological examination, and there is a large subjective influence, so the technical personnel in the field have not yet reached a consensus on the length of the distal resection margin. Studies have shown that the distal resection margin length selected based on traditional pathological characteristics (tumor differentiation, type, and R0 resection or not) is not strongly associated with the probability of recurrence or metastasis of colorectal cancer. Even if a longer distal resection margin is implemented, about 20% of R0 resection patients still have a risk of recurrence or metastasis, while some patients who choose a shorter distal resection margin also have a good prognosis. This indicates that there are obvious limitations in deciding the length of the distal resection margin based only on pathological characteristics, which cannot fully reflect individual differences and is difficult to accurately guide the selection of the length of the distal resection margin for a specific patient.

[0004] More and more studies have shown that genomic differences are related to prognostic factors of colorectal cancer patients, including tumor differentiation, vascular invasion, microresidual lesions, invasion, and metastasis. Currently, there is no related technology that systematically uses genomic information to predict the length of the distal resection margin, and there is also no individualized prediction method for the length of the distal resection margin of colorectal cancer based on genomics. SUMMARY

[0005] To address the lack of quantifiable, repeatable, preoperatively guiding, and multidimensional feature-based intelligent auxiliary decision-making tools in existing technologies to support individualized assessment of distal resection margins in colorectal cancer, this invention provides the application of reagents for detecting gene compositions in the preparation of products for predicting distal resection margin length in colorectal cancer patients.

[0006] Another objective of this invention is to provide a method for constructing a predictive model for the distal resection margin length in colorectal cancer patients.

[0007] Another object of the present invention is to provide a system for predicting the length of the distal resection margin in patients with colorectal cancer.

[0008] Another object of the present invention is to provide a computer device.

[0009] Another object of the present invention is to provide a computer-readable storage medium.

[0010] To achieve the above objectives, the present invention is implemented through the following solution:

[0011] This invention combines the somatic mutation characteristics of multiple genes and constructs a model to predict the distal resection margin length of colorectal cancer patients through machine learning algorithms. This not only improves the accuracy of resection margin length determination but also provides data support for precision treatment during surgery.

[0012] The application of reagents for detecting gene compositions in the preparation of products for predicting the distal resection margin length in colorectal cancer patients, wherein the gene compositions consist of ABCD1, MAP7D3, MET, MYO10, SYTL2 and ZNF257, namely the six characteristic genes described in this invention.

[0013] A method for constructing a predictive model for the distal resection margin length of colorectal cancer patients involves using the somatic mutation characteristics of colorectal cancer patients as input features and the distal resection margin length as the output result to train a regression model. The AUC (Area Under the Curve) is used as the performance metric for the model to obtain the predictive model for the distal resection margin length of the colorectal cancer patients.

[0014] The somatic mutation characteristic of the gene is the determination result of whether each gene of the gene composition has a somatic mutation in the intestine of the colorectal patient; the gene composition consists of ABCD1, MAP7D3, MET, MYO10, SYTL2 and ZNF257.

[0015] Preferably, the somatic mutations are obtained by combining genomic sequencing data of colorectal cancer patients with reference transcript analysis of each gene in the gene composition.

[0016] The sequencing technologies used include, but are not limited to, whole exome sequencing, whole genome sequencing, and probe panel sequencing.

[0017] More preferably, the sequencing technology used is whole exome sequencing technology.

[0018] More preferably, the reference transcript for ABCD1 is NM_000033; the reference transcript for MAP7D31 is NM_001173517 or NM_024597; the reference transcript for MET is NM_000245 or NM_001127500; the reference transcript for MYO10 is NM_012334; the reference transcript for SYTL2 is NM_032943 or NM_206927; and the reference transcript for ZNF257 is NM_001316996 or NM_033468.

[0019] Preferably, the somatic mutation characteristics of the genes include: the determination results of whether SYTL2, MAP7D3, MET, and ABCD1 have somatic mutations in intestinal tumor tissue (referred to as SYTL2_T, MAP7D3_T, MET_T, and ABCD1_T, respectively); and the determination results of whether ZNF257, ABCD1, and SYTL2 have somatic mutations in intestinal tissue located 0.8cm to 1.2cm from the lower edge of the tumor (referred to as ZNF257_P1, ABCD1_P2 ... The results of determining whether somatic mutations of MET and ZNF257 exist in intestinal tissue at a distance of 1.8cm to 2.2cm from the lower edge of the tumor (referred to as MET_P2 and ZNF257_P2, respectively); the results of determining whether somatic mutations of MET, ZNF257, and MYO10 exist in intestinal tissue at a distance of 4.8cm to 5.2cm from the lower edge of the tumor (referred to as MET_P5, ZNF257_P5, and MYO10_P5, respectively).

[0020] In this invention, the lower edge of the tumor refers to the lowest point where the tumor in the intestine extends towards the anus, that is, the edge of the tumor closest to the anus. In clinical imaging and pathological diagnosis, it is usually used as the starting point for measuring the distance of the tumor from the anus or the distance of the tumor from the distal resection margin.

[0021] More preferably, the gene somatic mutation characteristics include: the determination results of whether SYTL2, MAP7D3, MET and ABCD1 have somatic mutations in intestinal tumor tissue; the determination results of whether ZNF257, ABCD1 and SYTL2 have somatic mutations in intestinal tissue 1 cm from the lower edge of the tumor; the determination results of whether MET and ZNF257 have somatic mutations in intestinal tissue 2 cm from the lower edge of the tumor; and the determination results of whether MET, ZNF257 and MYO10 have somatic mutations in intestinal tissue 5 cm from the lower edge of the tumor.

[0022] Preferably, the method for obtaining the somatic mutation characteristics of the genes includes: if each gene of the gene composition has a somatic mutation in the intestine of the colorectal patient, then the somatic mutation characteristic of the corresponding gene is assigned a value of 1; if each gene of the gene composition does not have a somatic mutation in the intestine of the colorectal patient, then the somatic mutation characteristic of the corresponding gene is assigned a value of 0.

[0023] This invention also incorporates multiple clinical features on top of gene somatic mutation characteristics to jointly predict the distal resection margin length of colorectal cancer patients, resulting in higher prediction accuracy.

[0024] Preferably, the input features also include the clinical features of the colorectal cancer patient; the clinical features include gender, height, weight, alcohol consumption history, smoking history and tumor markers; the tumor markers include NSE, CA50, CEA and CA72-4.

[0025] More preferably, the method for obtaining the clinical characteristics includes: if the colorectal cancer patient is male, then the sex value is assigned 1; otherwise, the sex value is assigned 0; if the colorectal cancer patient's height is greater than 168cm, then the height value is assigned 1; otherwise, the height value is assigned 0; if the colorectal cancer patient's weight is greater than 65kg, then the weight value is assigned 1; otherwise, the weight value is assigned 0; if the colorectal cancer patient has a history of alcohol consumption, then the alcohol consumption history value is assigned 1; otherwise, the alcohol consumption history value is assigned 0; if the colorectal cancer patient has a history of smoking, then the smoking history value is assigned 1; otherwise, the weight value is assigned 0. If the patient's smoking history is positive, the CEA value is 1; otherwise, the CEA value is 0. If the patient's CA50 test result is positive, the CA50 value is 1; otherwise, the CA50 value is 0. If the patient's CA72-4 test result is positive, the CA72-4 value is 1; otherwise, the CA72-4 value is 0. If the patient's NSE test result is positive, the NSE value is 1; otherwise, the NSE value is 0.

[0026] The present invention constructs a predictive model for the distal resection margin length of colorectal cancer patients, which can be implemented using machine learning methods such as support vector machine (SVM), XGBoost, neural network, random forest, LASSO regression, logistic regression, and Bayesian classifier.

[0027] Preferably, the regression model includes a LASSO regression model.

[0028] In some specific implementations, the somatic mutation characteristics of the gene are used as input features, and the distal resection margin length is used as the output result to train a LASSO regression model. AUC is used as the performance criterion for the model, and the output is the model with the best performance for predicting the distal resection margin length of colorectal cancer patients.

[0029] As a specific implementation method, the model with the best performance for predicting the distal resection margin length of colorectal cancer patients is shown in Equation (1);

[0030] Formula (1): logit(P) = -1.535 + 1.7388 × SYTL2_T + 1.3954 × MET_T - 10.3082 × MAP7D3_T + 2.8344 × ZNF257_P1 - 3.1685 × MET_P2 - 4.1649 × ZNF257_P2 + 2.9768 × MET_P5 + 1.7647 × ZNF257_P5 + 2.8959 × MYO10_P5.

[0031] More preferably, the regression model includes a LASSO regression model and a generalized linear model.

[0032] In some specific implementations, the somatic mutation characteristics of the gene are used as input features, and the distal resection margin length is used as the output result to train a LASSO regression model. AUC is used as the performance criterion for the model, and the output is a first model with the best performance in predicting the distal resection margin length of colorectal cancer patients. The clinical characteristics are used as input features, and the distal resection margin length is used as the output result to train a LASSO regression model. AUC is used as the performance criterion for the model, and the output is a second model with the best performance in predicting the distal resection margin length of colorectal cancer patients. The output results of the first model and the second model are used as input features to train a generalized linear model. AUC is used as the performance criterion for the model, and the output is a third model with the best performance in predicting the distal resection margin length of colorectal cancer patients.

[0033] As a specific implementation, the first model, the second model and the third model are respectively as shown in formulas (1) to (3);

[0034] Formula (1): logit(P) = -1.535 + 1.7388 × SYTL2_T + 1.3954 × MET_T - 10.3082 × MAP7D3_T + 2.8344 × ZNF257_P1 - 3.1685 × MET_P2 - 4.1649 × ZNF257_P2 + 2.9768 × MET_P5 + 1.7647 × ZNF257_P5 + 2.8959 × MYO10_P5;

[0035] Formula (2): logit(C) = -1.0332 + 0.9839 × CEA + 0.4280 × weight + 0.0495 × drinking history - 1.4711 × smoking history - 1.5233 × gender + 1.4218 × height + 5.0847 × NSE - 0.5493 ​​× CA72 - 4 + 5.2670 × CA50.

[0036] Formula (3): logit(F) = 0.9548 + 1.1519 × logit(P) + 1.3417 × logit(C).

[0037] More preferably, the sigmoid function is used to map the output of the LASSO regression model to the 0-1 interval to obtain the predicted probability of the distal resection margin length for colorectal cancer patients; based on the predicted probability, a longer distal resection margin length is determined for colorectal cancer patients, and the specific determination criteria are as follows:

[0038] If the predicted probability is >0.5, the distal resection margin of the colorectal cancer patient is determined to be greater than 1 cm and less than or equal to 5 cm; if the predicted probability is ≤0.5, the distal resection margin of the colorectal cancer patient is determined to be greater than or equal to 0 cm and less than or equal to 1 cm.

[0039] A system for predicting the distal resection margin length in colorectal cancer patients includes a data acquisition module, an analysis module, and an output module;

[0040] The data acquisition module is used to acquire genomic sequencing data and clinical information of colorectal cancer patients;

[0041] The analysis module is a prediction model for the distal resection length of colorectal cancer patients obtained by the construction method. The genomic sequencing data and clinical information obtained by the data acquisition module are used as input variables and input into the prediction model for the distal resection length of colorectal cancer patients to obtain the distal resection length of colorectal cancer patients.

[0042] The output module is used to output the distal resection margin length of the colorectal cancer patient obtained by the analysis module.

[0043] A computer device includes a memory and a processor, the memory storing a computer program executable on the processor; when the computer program is executed by the processor, the system for predicting the distal resection margin length of a colorectal cancer patient operates.

[0044] A computer-readable storage medium storing a computer program executable by a processor, wherein when executed by the processor, the computer program implements the operation of the system for predicting the distal resection margin length in patients with colorectal cancer.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] This invention provides genes and clinical features closely related to the selection of distal resection margin length in colorectal cancer, and constructs a model to predict the distal resection margin length of colorectal cancer patients by combining machine learning algorithms. The AUC is as high as 0.83 or more, providing a data-driven and structurally well-defined guiding method for determining the distal resection margin length in clinical colorectal cancer resection, and has the value for widespread application. Attached Figure Description

[0047] Figure 1 The receiver operating characteristic (ROC) curve is the training set of the model used in Example 1 to predict the distal resection margin length in colorectal cancer patients.

[0048] Figure 2 The ROC curve for the test set of the model used in Example 1 to predict the distal resection margin length in colorectal cancer patients.

[0049] Figure 3 The training set ROC curves for the model in Comparative Example 1 that predicts the distal resection margin length in colorectal cancer patients.

[0050] Figure 4 The test set ROC curves for the model in Comparative Example 1 that predicts the distal resection margin length in colorectal cancer patients.

[0051] Figure 5 The training set ROC curve for the model used in Example 2 to predict the distal resection margin length in colorectal cancer patients.

[0052] Figure 6 The ROC curve for the test set of the model used in Example 2 to predict the distal resection margin length in colorectal cancer patients.

[0053] Figure 7 The validation set ROC curve for the model used in Example 2 to predict the distal resection margin length in colorectal cancer patients. Detailed Implementation

[0054] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. These embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Unless otherwise specified, the experimental methods used in the following embodiments are conventional methods; the materials and reagents used, unless otherwise specified, are commercially available.

[0055] Example 1: A method for predicting the distal resection margin length of colorectal cancer patients based on gene somatic mutation characteristics.

[0056] 1. Characteristic genes used to predict distal resection margin length in colorectal cancer patients

[0057] This invention uses six genes (ABCD1, MAP7D3, MET, MYO10, SYTL2 and ZNF257) as characteristic genes for predicting the distal resection margin length in colorectal cancer patients.

[0058] 2. Data Acquisition and Processing

[0059] (1) Sample source

[0060] This embodiment collected distal resection margin lengths (DRM, 0cm≤DRM≤5cm) from 80 colorectal cancer patients who underwent R0 resection surgery, and collected four types of clinical tissue samples: colorectal tumor tissue (T), intestinal tissue 1cm below the lower edge of the tumor (P1), intestinal tissue 2cm below the lower edge of the tumor (P2), and intestinal tissue 5cm below the lower edge of the tumor (P5). All patients were enrolled in accordance with the standards and guidelines of the "National Health Commission's Guidelines for the Diagnosis and Treatment of Colorectal Cancer in China," and the testing process strictly followed the relevant provisions of these guidelines regarding specimen collection and pathological evaluation, including requirements for specimen fixation, collection, and processing.

[0061] Eighty colorectal cancer patients were randomly divided into two groups: group 1 (60 patients) and group 2 (20 patients).

[0062] (2) Genome sequencing and data analysis

[0063] Whole-exome sequencing was used to sequence four types of clinical tissue samples from 80 colorectal cancer patients. Following the standard analysis procedures and methods outlined in the GATK (Genome Analysis Toolkit) Best Practices guide, the somatic mutation characteristics of 12 genes across six characteristic genes were analyzed. Specifically, the presence of somatic mutations in four genes (SYTL2, MAP7D3, MET, and ABCD1) in colorectal tumor tissue was analyzed (denoted as SYTL2_T, MAP7D3_T, MET_T, and ABCD1_T, respectively). The presence of somatic mutations in three genes (ZNF257, ABCD1, and SYTL2) in intestinal tissue located 1 cm below the lower border of the tumor was also analyzed. Cellular mutations (referred to as ZNF257_P1, ABCD1_P1, and SYTL2_P1, respectively); whether somatic mutations of the MET and ZNF257 genes exist in intestinal tissue 2 cm from the lower edge of the tumor (referred to as MET_P2 and ZNF257_P2, respectively); and whether somatic mutations of the MET, ZNF257, and MYO10 genes exist in intestinal tissue 5 cm from the lower edge of the tumor (referred to as MET_P5, ZNF257_P5, and MYO10_P5, respectively).

[0064] (3) Data conversion

[0065] Based on the sequencing data corresponding to the somatic mutation characteristics of 12 genes and the reference transcripts used in data analysis, the somatic mutation characteristics of each gene in each colorectal cancer patient were assigned values. Specifically: when a gene's somatic mutation characteristic indicates the presence of a somatic mutation (i.e., each characteristic gene exhibits a somatic mutation in the corresponding tissue of the colorectal cancer patient), the somatic mutation characteristic value for that gene is assigned 1; when a gene's somatic mutation characteristic indicates the absence of a somatic mutation (i.e., each characteristic gene does not exhibit a somatic mutation in the corresponding tissue of the colorectal cancer patient), the somatic mutation characteristic value for that gene is assigned 0. In this embodiment, the somatic mutation condition refers to the presence of mutations, including but not limited to those shown in Table 2, in each characteristic gene compared to its reference transcript.

[0066] Table 2. Mutation patterns of gene somatic mutation characteristics (partial)

[0067]

[0068] In addition, according to the standards and guidelines of the "Diagnosis and Treatment Specification for Colorectal Cancer of the National Health Commission", the distal resection margin length of each colorectal cancer patient was binarized as follows: when 0 cm ≤ DRM ≤ 1 cm, the DRM of this colorectal cancer patient was assigned a value of 0; when 1 cm < DRM ≤ 5 cm, the DRM of this colorectal cancer patient was assigned a value of 1.

[0069] (4)Data retention

[0070] Retain the converted data corresponding to the above 12 gene somatic mutation characteristics of each colorectal cancer patient in Set 1 and Set 2 of this embodiment for the next step of dataset division.

[0071] 3. Dataset division

[0072] For the data obtained in the previous step, the data of colorectal cancer patients belonging to Set 1 of this embodiment was used as the training set, and the data of colorectal cancer patients belonging to Set 2 of this embodiment was used as the test set.

[0073] 4. Construction of a model for predicting the distal resection margin length of colorectal cancer patients based on gene somatic mutation characteristics and performance evaluation

[0074] For the training set of this embodiment, using the Least Absolute Shrinkage and Selection Operator (LASSO) regression algorithm, with 12 gene somatic mutation characteristics (ZNF257_P1, MET_P5, SYTL2_T, ZNF257_P5, ABCD1_P1, MAP7D3_T, MET_T, ABCD1_T, MET_P2, SYTL2_P1, MYO10_P5 and ZNF257_P2) as the input features of the model and DRM as the output result, 5-fold cross-validation was used for training, and AUC was used as the model performance judgment index to output the model with the optimal performance, that is, a model for predicting the distal resection margin length of colorectal cancer patients based on gene somatic mutation characteristics was obtained, as shown in formula (1) specifically.

[0075] Formula (1): logit(P)=-1.535 + 1.7388×SYTL2_T + 1.3954×MET_T - 10.3082×MAP7D3_T + 2.8344×ZNF257_P1 - 3.1685×MET_P2 - 4.1649×ZNF257_P2 + 2.9768×MET_P5 + 1.7647×ZNF257_P5 + 2.8959×MYO10_P5.

[0076] After that, the sigmoid function is used to map logit(P) to the interval of 0-1, and the predicted probability of the distal resection margin length of colorectal cancer patients (denoted as Z) is obtained. When Z > 0.5, it is determined that the colorectal cancer patient has a long distal resection margin length, that is, 1 cm < DRM ≤ 5 cm; when Z ≤ 0.5, it is determined that the colorectal cancer patient has a short distal resection margin length, that is, 0 cm ≤ DRM ≤ 1 cm.

[0077] According to the DRM determination results and actual values of colorectal cancer patients in the training set of this embodiment, the ROC curve shown in Figure 1 is constructed. After calculation, the AUC is 0.830.

[0078] The test set of this embodiment is input into formula (1) to obtain the DRM determination results of colorectal cancer patients. After that, combined with the actual values of DRM, the ROC curve shown in Figure 2 is constructed. After calculation, the AUC is 0.880.

[0079] It can be seen that the AUC of the training set and test set of this embodiment are both above 0.83, indicating that the model constructed in this embodiment for predicting the distal resection margin length of colorectal cancer patients has excellent classification performance and can accurately predict the distal resection margin length of colorectal cancer patients.

[0080] Comparative Example 1 A method for predicting the distal resection margin length of colorectal cancer patients based on clinical features

[0081] 1. Data collection and processing

[0082] (1) Sample source

[0083] In this comparative example, the distal resection margin length (DRM, 0 cm ≤ DRM ≤ 5 cm) of 190 colorectal cancer patients (including 80 cases in Example 1) who underwent R0 resection surgery and 9 clinical features (3 individual indicators (gender, height, and weight), 4 tumor biochemical indicators (CEA, CA50, CA72-4, and NSE), and 2 personal medical histories (drinking history and smoking history)) are collected.

[0084] 109 colorectal cancer patients are randomly divided into Set 1 and Set 2, where Set 1 has 87 colorectal cancer patients and Set 2 has 22 colorectal cancer patients.

[0085] (2) Data conversion

[0086] To achieve the standardized expression of continuous variables in the model, in this comparative example, combined with clinical practice and the universal threshold (median) in the research field, the 9 clinical features of each colorectal cancer patient are binary converted as follows:

[0087] When the gender of a colorectal cancer patient is male, the gender is assigned a value of 1; otherwise, the gender is assigned a value of 0;

[0088] When the height of a colorectal cancer patient is greater than 168 cm, the height is assigned a value of 1; otherwise, the height is assigned a value of 0;

[0089] When the weight of a colorectal cancer patient is greater than 65 kg, the weight is assigned a value of 1; otherwise, the weight is assigned a value of 0;

[0090] When the CEA test result of a colorectal cancer patient is positive, CEA is assigned a value of 1; otherwise, CEA is assigned a value of 0;

[0091] When the CA50 test result of a colorectal cancer patient is positive, CA50 is assigned a value of 1; otherwise, CA50 is assigned a value of 0;

[0092] When the CA72-4 test result of a colorectal cancer patient is positive, CA72-4 is assigned a value of 1; otherwise, CA72-4 is assigned a value of 0;

[0093] When the NSE test result of a colorectal cancer patient is positive, NSE is assigned a value of 1; otherwise, NSE is assigned a value of 0;

[0094] When a colorectal cancer patient has a history of alcohol consumption, the alcohol consumption history is assigned a value of 1; otherwise, the alcohol consumption history is assigned a value of 0;

[0095] When a colorectal cancer patient has a history of smoking, the smoking history is assigned a value of 1; otherwise, the smoking history is assigned a value of 0.

[0096] In addition, using the same method as in Example 1, binary conversion is performed on the distal resection margin length of each colorectal cancer patient, specifically: when 0 cm ≤ DRM ≤ 1 cm, the DRM of this colorectal cancer patient is assigned a value of 0; when 1 cm < DRM ≤ 5 cm, the DRM of this colorectal cancer patient is assigned a value of 1.

[0097] (3)Data retention

[0098] Retain the converted data corresponding to the above 12 somatic mutation characteristics of each colorectal cancer patient in Set 1 and Set 2 of this comparative example for the next step of dataset division.

[0099] 3. Dataset division

[0100] For the data obtained in the previous step, use the data of colorectal cancer patients belonging to Set 1 of this comparative example as the training set and the data of colorectal cancer patients belonging to Set 2 of this comparative example as the test set.

[0101] 4. Construct a model for predicting the distal resection margin length of colorectal cancer patients based on clinical characteristics and performance evaluation

[0102] For the training set of this comparative example, using the LASSO algorithm, nine clinical features (CEA, body weight, drinking history, smoking history, height, NSE, CA50, gender, and CA72-4) were used as the input features of the model, and DRM was used as the output result. Five-fold cross-validation was used for training, and AUC was used as the model performance judgment index. The model with the best performance was output, that is, a model for predicting the distal resection margin length of colorectal cancer patients based on clinical features was obtained, as shown in formula (2) specifically.

[0103] Formula (2): logit(C)=-1.0332 + 0.9839×CEA + 0.4280×body weight + 0.0495×drinking history - 1.4711×smoking history - 1.5233×gender + 1.4218×height + 5.0847×NSE - 0.5493×CA72-4 + 5.2670×CA50.

[0104] After that, the sigmoid function was used to map logit(C) to the interval of 0-1 to obtain the prediction probability (Z) of the distal resection margin length of colorectal cancer patients. When Z>0.5, it was determined that colorectal cancer patients had a long distal resection margin length, that is, 1 cm < DRM ≤ 5 cm; when Z ≤ 0.5, it was determined that colorectal cancer patients had a short distal resection margin length, that is, 0 cm ≤ DRM ≤ 1 cm.

[0105] According to the DRM prediction results and actual values of colorectal cancer patients in the training set of this comparative example, the ROC curve shown in Figure 3 was constructed. After calculation, the AUC was 0.800.

[0106] The test set of this comparative example was input into formula (2) to obtain the DRM determination results of colorectal cancer patients. After that, combined with the actual values of DRM, the ROC curve shown in Figure 4 was constructed. After calculation, the AUC was 0.590.

[0107] It can be seen that the AUCs of the training set and test set of this comparative example are both below 0.8, indicating that the classification performance of the model for predicting the distal resection margin length of colorectal cancer patients constructed in this comparative example is poor and cannot accurately predict the distal resection margin length of colorectal cancer patients as accurately as the model for predicting the distal resection margin length of colorectal cancer patients constructed in Example 1.

[0108] Example 2 A method for predicting the distal resection margin length of colorectal cancer patients based on gene somatic mutation characteristics and clinical characteristics

[0109] 1. Data division

[0110] (1) Sample source

[0111] The 80 colorectal cancer patients in both Example 1 and Comparative Example 1 were randomly divided into set 1 and set 2, with set 1 containing 60 colorectal cancer patients and set 2 containing 20 colorectal cancer patients.

[0112] In addition, this embodiment also included 19 colorectal cancer patients who underwent R0 resection surgery as set 3. These patients did not overlap with the patients in Example 1 and Comparative Example 1. The enrollment of all the above patients met the standards and guidelines of the "National Health Commission's Guidelines for the Diagnosis and Treatment of Colorectal Cancer in China". The testing process strictly followed the relevant provisions of the guidelines regarding specimen collection and pathological evaluation, including the requirements for specimen fixation, collection, and processing.

[0113] For each colorectal cancer patient in sets 1–3, the distal resection margin length (DRM, 0 cm ≤ DRM ≤ 5 cm) and nine clinical characteristics (CEA, weight, alcohol consumption history, smoking history, height, NSE, CA50, sex, and CA72-4) were collected, and four types of clinical tissue samples were collected (tumor tissue of the colorectal region (T), intestinal tissue 1 cm from the lower edge of the tumor (P1), intestinal tissue 2 cm from the lower edge of the tumor (P2), and intestinal tissue 5 cm from the lower edge of the tumor (P5)).

[0114] (2) Genome sequencing and data analysis

[0115] Using the same method as in Example 1, genome sequencing was performed on four types of clinical tissue samples from each colorectal cancer patient in sets 1 to 3, and the presence of somatic mutations in 12 genes (ZNF257_P1, MET_P5, SYTL2_T, ZNF257_P5, ABCD1_P1, MAP7D3_T, MET_T, ABCD1_T, MET_P2, SYTL2_P1, MYO10_P5, and ZNF257_P2) was detected.

[0116] (3) Data conversion

[0117] Using the same method as in Example 1 and Comparative Example 1, data conversion was performed on the distal resection margin length, sequencing data corresponding to 12 gene somatic mutation features, and 9 clinical features of each colorectal cancer patient in sets 1 to 3.

[0118] 3. Dataset partitioning

[0119] For the data obtained in the previous step, the data of colorectal cancer patients belonging to sets 1 to 3 of this embodiment will be used as the training set, test set, and validation set in sequence.

[0120] 4. Constructing and evaluating a model for predicting distal resection margin length in colorectal cancer patients based on gene somatic mutation characteristics and clinical features.

[0121] For the training set of this example, the model for predicting the distal resection margin length of colorectal cancer patients constructed based on gene somatic mutation features in Example 1 (i.e., formula (1)) and the model for predicting the distal resection margin length of colorectal cancer patients constructed based on clinical features in Comparative Example 1 (i.e., formula (2)) were respectively combined to sequentially obtain the logit(P) value and logit(C) value of each colorectal cancer patient. Then, as input features, using the Generalized Linear Model (GLM) algorithm, with DRM as the output result, 5-fold cross-validation was used for training, and AUC was used as the model performance judgment index to output the model with the optimal performance, that is, the model for predicting the distal resection margin length of colorectal cancer patients constructed based on gene somatic mutation features and clinical features, as specifically shown in formula (3).

[0122] Formula (3): logit(F) = 0.9548 + 1.1519×logit(P) + 1.3417×logit(C).

[0123] After that, the sigmoid function was used to map the logit(F) value to the interval of 0-1 to obtain the predicted probability (Z) of the distal resection margin length of colorectal cancer patients. When Z > 0.5, it was determined that the colorectal cancer patient adopted a long distal resection margin length, that is, 1 cm < DRM ≤ 5 cm; when Z ≤ 0.5, it was determined that the colorectal cancer patient adopted a short distal resection margin length, that is, 0 cm ≤ DRM ≤ 1 cm.

[0124] According to the DRM determination results and actual values of colorectal cancer patients in the training set of this example, the ROC curve shown in Figure 5 was constructed. After calculation, the AUC was 0.940.

[0125] The test set of this example was input into formula (3) to obtain the DRM determination result of colorectal cancer patients. Then, combined with the actual value of DRM, the ROC curve shown in Figure 6 was constructed. After calculation, the AUC was 0.920.

[0126] The validation set of this example was input into formula (3) to obtain the DRM determination result of colorectal cancer patients. Then, combined with the actual value of DRM, the ROC curve shown in Figure 7 was constructed. After calculation, the AUC was 0.910.

[0127] As can be seen, the AUC of the training set, test set and validation set of this embodiment are all above 0.9, indicating that the classification performance of the model for predicting the distal resection length of colorectal cancer patients constructed in this embodiment is excellent. Moreover, compared with the model for predicting the distal resection length of colorectal cancer patients based on gene somatic mutation features constructed in Example 1, it has a higher AUC and can more accurately predict the distal resection length of colorectal cancer patients.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description and ideas, and it is neither necessary nor possible to exhaustively describe all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. The application of a reagent for detecting gene compositions in the preparation of products for predicting the distal resection margin length in colorectal cancer patients, characterized in that, The gene composition consists of ABCD1, MAP7D3, MET, MYO10, SYTL2 and ZNF257.

2. A method for constructing a predictive model for the distal resection margin length in colorectal cancer patients, characterized in that, Using the somatic mutation characteristics of colorectal cancer patients as input features and the distal resection margin length as output, a regression model was trained, and AUC was used as the performance metric to obtain a predictive model for the distal resection margin length of the colorectal cancer patients. The prediction model for the distal resection margin length of colorectal cancer patients is shown in formula (1); Formula (1): logit(P) = -1.535 + 1.7388 × SYTL2_T + 1.3954 × MET_T - 10.3082 × MAP7D3_T + 2.8344 × ZNF257_P1 - 3.1685 × MET_P2 - 4.1649 × ZNF257_P2 + 2.9768 × MET_P5 + 1.7647 × ZNF257_P5 + 2.8959 × MYO10_P5; The somatic mutation characteristic of the gene refers to the determination result of whether each gene of the gene composition in claim 1 has a somatic mutation in the intestine of the colorectal patient, including: Results of determining whether SYTL2, MAP7D3, MET and ABCD1 have somatic mutations in intestinal tumor tissue; The results of determining whether ZNF257, ABCD1, and SYTL2 exist in intestinal tissue at a distance of 0.8 cm to 1.2 cm from the lower edge of the tumor; The results of determining whether somatic mutations of MET and ZNF257 exist in intestinal tissue located 1.8cm to 2.2cm from the lower margin of the tumor; The results of determining whether somatic mutations of MET, ZNF257, and MYO10 exist in intestinal tissue located 4.8 cm to 5.2 cm from the lower margin of the tumor; The methods for obtaining the somatic mutation characteristics of the gene include: If each gene in the gene composition of claim 1 has a somatic mutation in the intestine of the colorectal patient, then the somatic mutation characteristic of the corresponding gene is assigned a value of 1; if each gene in the gene composition of claim 1 does not have a somatic mutation in the intestine of the colorectal patient, then the somatic mutation characteristic of the corresponding gene is assigned a value of 0.

3. The construction method according to claim 2, characterized in that, The input features also include the clinical characteristics of the colorectal cancer patient; The clinical characteristics include sex, height, weight, alcohol consumption history, smoking history, and tumor markers; the tumor markers include NSE, CA50, CEA, and CA72-4.

4. The construction method according to claim 3, characterized in that, The clinical features are transformed into data before being used as input features; The methods for data transformation of the clinical features include: If the colorectal cancer patient is male, the gender value is 1; otherwise, the gender value is 0. If a colorectal cancer patient's height is greater than 168cm, the height value is assigned as 1; otherwise, the height value is assigned as 0. If a colorectal cancer patient weighs more than 65 kg, the weight is assigned a value of 1; otherwise, the weight is assigned a value of 0. If a colorectal cancer patient has a history of alcohol consumption, the alcohol consumption history is assigned a value of 1; otherwise, the alcohol consumption history is assigned a value of 0. If a colorectal cancer patient has a history of smoking, the smoking history value is assigned as 1; otherwise, the smoking history value is assigned as 0. If a colorectal cancer patient tests positive for CEA, the CEA value is assigned as 1; otherwise, the CEA value is assigned as 0. If a colorectal cancer patient has a positive CA50 test result, the CA50 value is assigned to 1; otherwise, the CA50 value is assigned to 0. If a colorectal cancer patient tests positive for CA72-4, the CA72-4 value is assigned as 1; otherwise, the CA72-4 value is assigned as 0. If a colorectal cancer patient has a positive NSE test result, the NSE value is assigned as 1; otherwise, the NSE value is assigned as 0.

5. The construction method according to any one of claims 2 to 4, characterized in that, The regression model includes the LASSO regression model.

6. A system for predicting the length of the distal resection margin in patients with colorectal cancer, characterized in that, It includes a data acquisition module, an analysis module, and an output module; The data acquisition module is used to acquire genomic sequencing data and clinical information of colorectal cancer patients; The analysis module is a prediction model for the distal resection length of colorectal cancer patients obtained by the construction method described in any one of claims 2 to 4. The genomic sequencing data and clinical information obtained by the data acquisition module are used as input variables and input into the prediction model for the distal resection length of colorectal cancer patients to obtain the distal resection length of colorectal cancer patients. The output module is used to output the distal resection margin length of the colorectal cancer patient obtained by the analysis module.

7. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program that can be executed on the processor; when the computer program is executed by the processor, it implements the operation of the system of claim 6.

8. A computer-readable storage medium storing a computer program executable by a processor, characterized in that, When the computer program is executed by the processor, it implements the operation of the system of claim 6.

Images