Marker for detecting tumor radiotherapy sensitivity and screening method and application thereof

Abstract

The invention provides a marker for detecting tumor radiotherapy sensitivity and a screening method and application thereof, and belongs to the technical field of medical diagnosis. Comprising the following steps: obtaining a processed sample set, an initial differential substance sample set, a second differential metabolite sample set, a third differential metabolite sample set, a trained logistic regression model, a trained random forest model, a trained support vector machine and a trained LASSO model; and by utilizing ROC analysis, according to the AUC value, determining an optimal model of logistic regression, LASSO, a support vector machine and a random forest, and merging substances of the obtained optimal model to obtain the tumor radiotherapy sensitivity detection marker. According to the method, multiple samples from different sources can be screened to obtain multiple markers for detecting tumor radiotherapy sensitivity, radiotherapy sensitivity of cancer patients can be effectively distinguished independently or in a combined mode, and the accuracy of judging the radiotherapy sensitivity degree is improved.

Description

Technical Field

[0001] The present invention belongs to the field of medical diagnosis technology, and in particular relates to a tumor radiotherapy sensitivity marker detection method and application thereof. Background Art

[0002] Colorectal cancer accounts for approximately 10% of all cancers and cancer-related deaths diagnosed worldwide each year. It is the second most common cancer in women and the third most common cancer in men. Radiotherapy is an important part of the comprehensive treatment of colorectal cancer, and the preoperative chemoradiotherapy treatment strategy remains the standard treatment strategy for locally advanced rectal cancer (stage II and III). However, while radiotherapy brings therapeutic benefits to patients, it also increases its toxic side effects. The proportion of people who benefit is relatively small, and there are still many problems that need to be solved.

[0003] In the field of radiotherapy, clinical efficacy varies greatly between individuals. Even if the same regimen and dose of radiotherapy are received, about half of the tumors still do not achieve obvious regression, and even progress. The resistance of colorectal cancer (CRC) cells to radiotherapy largely leads to poor clinical prognosis of colorectal cancer patients. However, there are still obstacles in distinguishing colorectal cancer patients with high sensitivity (HS), moderate sensitivity (MS) and low sensitivity (LS) to radiotherapy. In the evaluation of radiotherapy response, tumor regression grading (TRG) is widely accepted. Previous studies have shown that AJCC-TRG grading accurately predicts the prognosis of rectal cancer after chemoradiotherapy. However, TRG grading fails to grade the sensitivity of radiation response of colorectal cancer patients before treatment. Therefore, it is of great significance to obtain a method for screening tumor radiotherapy sensitivity markers for the prediction of radiotherapy sensitivity. Summary of the invention

[0004] In view of this, the object of the present invention is to provide a marker for detecting tumor radiotherapy sensitivity and a screening method and application thereof, wherein the marker for detecting tumor radiotherapy sensitivity can be used to accurately detect tumor radiotherapy sensitivity.

[0005] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0006] The present invention provides a method for screening and detecting tumor radiotherapy sensitivity markers, comprising the following steps:

[0007] S1, normalizing the initial sample set to obtain a processed sample set;

[0008] S2. Processing the processed sample set using principal component analysis and orthogonal partial least squares method to obtain an initial difference substance sample set;

[0009] S3, using a rank sum test to preliminarily screen the initial difference sample set to obtain a second difference metabolite sample set;

[0010] S4. Use multiple comparison method and PerMANOVA multivariate analysis of variance method to screen the initial clinical indicators of the cohort samples of the prHS and prMS, prHS and prLS, prMS and prLS comparison experimental groups. If one or more clinical indicators in the comparison experimental group are not significant, the corresponding differential samples in the second differential metabolite sample set are differential substances after excluding the interference of clinical factors; if one or more clinical indicators in the comparison experimental group are significant, the differential samples in the corresponding second differential metabolite sample set are compared with the logistic regression model established with the significant clinical indicators and the P-value of the logistic regression model, and the differential substances with p-value < 0.05 are screened to obtain the differential substances after excluding the interference of clinical factors; the differential substances after excluding the interference of clinical factors are combined as the third differential metabolite sample set;

[0011] S5. Logistic regression, LASSO, support vector machine, and random forest model were used to screen and analyze the third differential metabolite sample set, and then the substances obtained by logistic regression, LASSO, support vector machine, and random forest model were combined to draw a joint ROC curve. The maximum AUC value was used as the optimal model to obtain a tumor radiotherapy sensitivity marker.

[0012] Preferably, in step S1, before normalization processing is performed, the following steps are further included:

[0013] Perform deviation value filtering on the initial sample set to obtain a first filtered sample set;

[0014] Performing missing value filtering on the first filtered sample set to obtain a second filtered sample set;

[0015] Filling missing values ​​in the sample set after the second filtering to obtain a filled sample set;

[0016] In step S1, the padded sample set is normalized to obtain a processed sample set;

[0017] In step S2, the principal component analysis method includes: using the processed sample set, performing principal component analysis on three different comparison strategies, namely, comparing prHS with prMS, comparing prHS with prLS, and comparing prMS with prLS, respectively, and reflecting the differences between groups and the distribution of samples through the principal component score scatter plot;

[0018] In step S3, the initial screening method is to use a rank sum test to calculate the hypothesized value p-value in the initial differential substance sample set, and use the differential substances with p-value < 0.05 as the screening differentially expressed substances to obtain the second differential metabolite sample set;

[0019] In step S4, the significant clinical indicators are screened and determined by using the multiple comparison method and PerMANOVA multivariate analysis of variance method to screen the initial clinical indicators; the initial clinical indicators include treatment response grouping, age, gender, body mass index (BMI), smoking amount, alcohol consumption, primary tumor T stage (T), regional lymph node N stage (N), tumor target volume (GTV), cm from the anus, circumferential resection margin (CRM), extramural vascular invasion (EMVI), neutrophil-lymphocyte ratio (NLR), concurrent chemotherapy regimen, hormone usage and antibiotic usage.

[0020] Preferably, the initial sample set is obtained by preprocessing samples, and the sample preprocessing method includes:

[0021] The obtained multi-cohort samples were classified into 6 categories according to the NAR (Neoadjuvant Rectal) score and before and after radiotherapy: sensitive data prHS before radiotherapy, moderately sensitive data prMS before radiotherapy, insensitive data prLS before radiotherapy, sensitive data poHS after radiotherapy, moderately sensitive data poMS after radiotherapy, and insensitive data poLS before and after radiotherapy;

[0022] All sample data were analyzed in positive and negative ion modes. For substances with duplicate names obtained by secondary mass spectrometry qualitative matching analysis, substances with high scores were retained, and substances with low scores were deleted, and the data in the samples were retained. When the scores of duplicate substances were the same, substances with high peak areas in the quality control samples were retained, and substances with low peak areas were deleted, and only the data in the samples were retained to obtain the initial sample set.

[0023] Preferably, the sample comprises a serum sample;

[0024] When the sample is a serum sample, the tumor radiotherapy sensitivity markers screened by the method are metabolic markers, and the metabolic markers include Oleamide (CAS: 3072-13-7), Dihydrofolicacid (CAS: 4033-27-6), H-PHE-PRO-OH (CAS: 7669-65-0), 2-Methylglutaricacid (CAS: 617-62-9), Phenylglyoxylicacid (CAS: 611-73-4), Thyroxine (CAS: 51-48- 9), 1-Methyl-L-Histidine (CAS: 332-80-9), Uridine (CAS: 58-96-8), D-Glutamine (CAS: 5959-95-5), Deoxycholicacid (CAS: 83-44 -3), Lauroyldiethanolamide (CAS:120-40-1), (R)-3-Hydroxy-tetradecanoicacid (CAS:28715-21-1) and Cholicacid (CAS:81-25-4).

[0025] Preferably, the metabolic markers are H-Phe-Pro-OH, 2-Methylglutaricacid, Levothyroxine, 1-Methyl-L-Histidine, Uridine, D-Gln-OH, Deoxycholicacid, (R)-3-Hydroxy-tetradecanoic acid, and Cholic acid.

[0026] The present invention also provides a computer-readable storage medium, which is used to store computer instructions, programs, code sets or instruction sets. When the computer-readable storage medium is run on a computer, the computer executes the method for screening and detecting tumor radiotherapy sensitivity markers as described above.

[0027] The present invention also provides an electronic device, comprising one or more processors and a storage device, storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for screening and detecting tumor radiotherapy sensitivity markers.

[0028] The present invention also provides a method for screening and detecting tumor radiotherapy sensitivity markers to obtain tumor radiotherapy sensitivity markers.

[0029] The present invention also provides an application of a tumor radiotherapy sensitivity marker in the preparation of a reagent for detecting tumor radiotherapy sensitivity.

[0030] Preferably, the tumor radiotherapy sensitivity marker includes a metabolic marker.

[0031] Preferably, the critical value of the logistic regression model constructed based on the tumor radiotherapy sensitivity marker is used to predict the radiotherapy sensitivity of colorectal cancer patients.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention provides a tumor radiotherapy sensitivity detection marker and a screening method and application thereof. The tumor radiotherapy sensitivity detection marker is obtained by the method for screening the tumor radiotherapy sensitivity detection marker. The method can screen samples from multiple different sources to obtain multiple tumor radiotherapy sensitivity detection markers, which can be used independently or in combination to effectively distinguish the radiotherapy sensitivity of cancer patients. The present invention uses the tumor radiotherapy sensitivity detection marker obtained by the above method to perform sensitivity grading of the radiation response of tumor patients before treating the tumor, thereby improving the accuracy of judging the degree of radiotherapy sensitivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is the AUC curve of prHS+prMSvsprLS non-targeted metabolomics;

[0035] Figure 2 is the AUC curve of prHSvsprMS non-targeted metabolomics;

[0036] Figure 3 Targeted metabolomics AUC curves of the Fuzhou cohort, where A is the prHS+prMSvsprLS targeted metabolomics AUC curve, and B is the prHSvsprMS targeted metabolomics AUC curve;

[0037] Figure 4 Figure 2 is the AUC curve of targeted metabolomics of the Guangzhou cohort, where A is the AUC curve of targeted metabolomics of prHS+prMSvsprLS, and B is the AUC curve of targeted metabolomics of prHSvsprMS. DETAILED DESCRIPTION

[0038] The present invention provides a method for screening and detecting tumor radiotherapy sensitivity markers, comprising the following steps:

[0039] S1, normalizing the initial sample set to obtain a processed sample set;

[0040] S2. Processing the processed sample set using principal component analysis and orthogonal partial least squares method to obtain an initial difference substance sample set;

[0041] S3, using a rank sum test to preliminarily screen the initial difference sample set to obtain a second difference metabolite sample set;

[0042] S4. Use multiple comparison method and PerMANOVA multivariate analysis of variance method to screen the initial clinical indicators of the cohort samples of the prHS and prMS, prHS and prLS, prMS and prLS comparison experimental groups. If one or more clinical indicators in the comparison experimental group are not significant, the corresponding differential samples in the second differential metabolite sample set are differential substances after excluding the interference of clinical factors; if one or more clinical indicators in the comparison experimental group are significant, the differential samples in the corresponding second differential metabolite sample set are compared with the logistic regression model established with the significant clinical indicators and the P-value of the logistic regression model, and the differential substances with p-value < 0.05 are screened to obtain the differential substances after excluding the interference of clinical factors; the differential substances after excluding the interference of clinical factors are combined as the third differential metabolite sample set;

[0043] S5. Logistic regression, LASSO, support vector machine, and random forest model were used to screen and analyze the third differential metabolite sample set, and then the substances obtained by logistic regression, LASSO, support vector machine, and random forest model were combined to draw a joint ROC curve. The maximum AUC value was used as the optimal model to obtain a tumor radiotherapy sensitivity marker.

[0044] In the present invention, the initial sample set is normalized to obtain a processed sample set. The initial sample set is obtained by preprocessing the samples, and the sample preprocessing method preferably includes:

[0045] The obtained multi-cohort samples were classified into 6 categories according to the NAR (Neoadjuvant Rectal) score and before and after radiotherapy: pre-radiotherapy sensitivity data (prHS), pre-radiotherapy moderate sensitivity data (prMS), pre-radiotherapy insensitive data (prLS), post-radiotherapy sensitivity data (poHS), post-radiotherapy moderate sensitivity data (poMS), and post-radiotherapy insensitive data (poLS); when the NAR score is <8, it means that the patient is highly sensitive to tumor radiotherapy, when the NAR score is 8-16, it means that the patient is moderately sensitive to tumor radiotherapy, and when the NAR score is >16, it means that the patient is insensitive to tumor radiotherapy.

[0046] All sample data were analyzed in positive and negative ion modes. For substances with duplicate names obtained by secondary mass spectrometry qualitative matching analysis, substances with high scores were retained, and substances with low scores were deleted, and the data in the samples were retained. When the scores of duplicate substances were the same, substances with high peak areas in the quality control samples were retained, and substances with low peak areas were deleted, and only the data in the samples were retained to obtain the initial sample set.

[0047] After obtaining the initial sample set, before normalization processing, it is also preferably included: filtering the initial sample set for deviation values ​​to obtain a first filtered sample set; filtering the first filtered sample set for missing values ​​to obtain a second filtered sample set; filling the second filtered sample set with missing values ​​to obtain a filled sample set. The deviation value filtering is preferably filtering the deviation value based on the relative standard deviation: the relative standard deviation is calculated for the quality control group, and the relative standard deviation in the quality control group is greater than 0.3. It is considered that the detection is unstable, and all detection data of the substance are deleted. The missing value filtering is preferably to retain only the peak area data with no more than 50% null values ​​in a single group or no more than 50% null values ​​in all groups; further, the group includes pre-radiotherapy sensitive data (referred to as prHS), pre-radiotherapy moderately sensitive data (referred to as prMS), pre-radiotherapy insensitive data (referred to as prLS), post-radiotherapy sensitive data (referred to as poHS), post-radiotherapy moderately sensitive data (referred to as poMS), and post-radiotherapy insensitive data (referred to as poLS). The missing value filling is to fill the missing values ​​in the second filtered sample set with the minimum half.

[0048] After obtaining the filled sample set, normalization processing is performed to obtain the processed sample set. The normalization processing is preferably performed using the total ion flow of each filled sample.

[0049] In the present invention, after obtaining the processed sample set, the processed sample set is processed using principal component analysis and orthogonal partial least squares to obtain an initial differential substance sample set. The differential substance is a substance (such as metabolites or flora) that causes different samples to produce different sensitivity levels. The principal component analysis method includes: through the processed sample set, three different comparison strategies for comparing prHS with prMS, comparing prHS with prLS, and comparing prMS with prLS, respectively, and the differences between groups and the distribution of samples are reflected by the principal component score scatter plot. When the sample point distribution is closer, the content and type of the substance in the sample are more similar, and vice versa, the sample point distribution is farther, and the overall material level difference is greater. The principal component analysis of the present invention is a statistical method that converts a set of observed possible related variables into linear unrelated variables (i.e., principal components) through orthogonal transformation, which can effectively highlight the overall distribution trend of metabolomics data and the degree of difference between samples in groups. Metabolomics data has the characteristics of high dimension (many types of metabolites are detected) and small sample (small sample size), and includes both difference variables related to categorical variables and a large number of indifferent variables that may be associated with each other in these variables. This results in that if the PCA model is used for analysis, due to the influence of related variables, the difference variables will be dispersed to more principal components, which is not conducive to finding the real difference between groups and screening effective difference metabolites. Therefore, the present invention further adopts the statistical method of orthogonal partial least squares discriminant analysis to analyze the processed sample set. By orthogonal partial least squares discriminant analysis, the orthogonal variables that are not related to the categorical variables in the metabolites can be filtered out, and the non-orthogonal variables and orthogonal variables are analyzed separately, so as to obtain more reliable intergroup differences of substances (such as metabolites) and the degree of correlation of the experimental group. The difference multiples between the two groups are compared by calculating the FoldChange value. The difference multiples of the substance between the two groups are greater than 1, which represents upregulation, and less than 1 represents downregulation. After the above analysis, combined with the unit variable and multivariate statistical analysis results, the difference substances are screened out. The present invention uses a multivariate statistical analysis method to analyze the data according to the inherent characteristics of the material (such as metabolite) data. Compared with traditional univariate statistical analysis methods such as the rank sum test, which pay more attention to the independent changes in metabolite levels, multivariate statistical analysis pays more attention to the relationship between metabolites and their promotion / antagonism in biological processes. Considering the results of two types of statistical analysis methods at the same time helps to observe the data from different angles and draw conclusions, and can also avoid false positive errors or model overfitting caused by using only one type of statistical analysis method.

[0050] In the present invention, after obtaining the initial differential sample set, the initial differential sample set is preliminarily screened using a rank sum test to obtain a second differential metabolite sample set. The rank sum test is also called the Mann-WhitneyU test, which is a non-parametric test used to test whether two independent samples are from the same distribution. The initial screening method preferably uses a rank sum test to calculate the hypothesized value p-value in the initial differential substance sample set, and uses the differential substances with p-value < 0.05 as the screening differentially expressed substances to obtain the second differential metabolite sample set. Further preferably, the null hypothesis is first established: the two groups of samples come from the same population, the two groups of data are mixed, and the ranks are uniformly arranged from small to large. When encountering the same data, the average rank is taken, and the statistic is calculated. The significance level a = 0.05 is set, and the probability of the current statistic value is obtained under the condition that the null hypothesis is established. If p-value < 0.05, the null hypothesis is rejected, indicating that there is a significant difference between the two samples.

[0051] In the present invention, the initial clinical indicators of the cohort samples of the prHS and prMS, prHS and prLS, prMS and prLS, prHS+prMS and prLS comparative experimental groups are screened by using multiple comparison method and PerMANOVA multivariate analysis of variance method. If one or more clinical indicators in the comparative experimental group are not significant, the differential samples in the corresponding second differential metabolite sample set are the differential substances after excluding the interference of clinical factors; if one or more clinical indicators in the comparative experimental group are significant, the logistic regression model established between the differential samples in the corresponding second differential metabolite sample set and the significant clinical indicators and the P-value of the logistic regression model are used to screen the differential substances with p-value﹤0.05, and obtain the differential substances after excluding the interference of clinical factors; the differential substances after excluding the interference of clinical factors are combined as the third differential metabolite sample set. The significant clinical indicators are preferably determined by screening the initial clinical indicators using the multiple comparison method and the PerMANOVA multivariate analysis of variance method; the initial clinical indicators include treatment response grouping, age, gender, body mass index (BMI), smoking, drinking, primary tumor T stage (T), regional lymph node N stage (N), tumor target volume (GTV), cm from the anus, circumferential resection margin (CRM), extramural vascular invasion (EMVI), neutrophil-lymphocyte ratio (NLR), concurrent chemotherapy regimen, hormone usage, and antibiotic usage. Further preferably, when the initial clinical indicators are screened using the multiple comparison method and the PerMANOVA multivariate analysis of variance method, the screening method includes: obtaining the initial clinical indicators; using the multiple comparison method to retain the samples with p-value < 0.05 in the statistical test results of the initial clinical indicators, and determining the second clinical indicators; using the PerMANOVA multivariate analysis of variance method to screen the clinical indicator information that has a significant relationship with the second clinical indicator, and determine the significant clinical indicators. A further preferred method of the present invention for screening the initial clinical indicators using the multiple comparison method is: multiple comparisons are made on the clinical indicators of prHS and prMS, prHS and prLS, prMS and prLS, prHS+prMS and prLS, firstly the homogeneity is tested by Bartlett Test (corresponding to R language program function bartlett.test) (taking p value <0.05 as a non-null hypothesis), and then the p value is obtained by the oneway.test function of the R language program package. If p <0.05, it means that the clinical indicator is significant.After obtaining the significant clinical indicators, a logistic regression model is used to perform a final correction on the second differential metabolite sample set using the significant clinical indicator information as the response variable, and the differential substances whose p-value after regression is still less than 0.05 are screened out. In a further preferred embodiment, the significant clinical indicator information is used as the response variable, and the second differential metabolite sample set is used as the independent variable. A logistic regression model is established to obtain the p-value. If the p-value of the differentially expressed substance is less than 0.05, it means that these substances are not affected by the clinical indicators. The differential substances after excluding the interference of clinical factors are obtained by merging, that is, the third differential metabolite sample set is obtained.

[0052] In the present invention, after obtaining the third differential metabolite sample set, logistic regression, LASSO, support vector machine, and random forest model are used to screen and analyze the third differential metabolite sample set, and then the substances obtained by logistic regression, LASSO, support vector machine, and random forest model are combined to draw a joint ROC curve. The maximum AUC value is used as the optimal model to obtain a tumor radiotherapy sensitivity marker.

[0053] In the present invention, the sample preferably includes a serum sample; the serum sample is preferably collected by centrifuging the blood to obtain the supernatant. The centrifugation is preferably a double centrifugation, that is, the blood is allowed to stand at room temperature for 0.5 to 1.5 hours for coagulation and stratification, then centrifuged at 2500 to 3500 rpm at room temperature for 5 to 15 minutes, the supernatant is transferred to a clean centrifuge tube, and then centrifuged at 11000 to 12000 rpm at 4°C for 5 to 15 minutes to obtain the supernatant.

[0054] When the sample is a serum sample, the tumor radiotherapy sensitivity marker screened by the method is preferably a metabolic marker, and the metabolic marker includes oleamide (CAS: 3072-13-7), dihydrofolic acid (CAS: 4033-27-6), H-PHE-PRO-OH (CAS: 7669-65-0), 2-Methylglutaric acid (CAS: 617-62-9), phenylglyoxylic acid (CAS: 611-73-4), Thyroxine (CAS: 51-48-9), 1-Methyl-L-Histidine (CAS: 332-80-9), Uridine (CAS: 58-96-8), D-Glutamine (CAS: 5959-95-5), Deoxycholic acid (CAS: 83-44-3), Lauroyl diethanolamide (CAS: 120-40-1), (R)-3-Hydroxy-tetradecanoic acid (CAS: 28715-21-1) and Cholic acid (CAS: 81-25-4);

[0055] Furthermore, the metabolic markers are H-Phe-Pro-OH, 2-Methylglutaric acid, Levothyroxine, 1-Methyl-L-Histidine, Uridine, D-Gln-OH, Deoxycholic acid, (R)-3-Hydroxy-tetradecanoic acid, and Cholic acid.

[0056] The present invention also provides a computer-readable storage medium, which is used to store computer instructions, programs, code sets or instruction sets. When the computer-readable storage medium is run on a computer, the computer executes the method for screening and detecting tumor radiotherapy sensitivity markers as described above.

[0057] The present invention also provides an electronic device, comprising one or more processors and a storage device, storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for screening and detecting tumor radiotherapy sensitivity markers.

[0058] The present invention also provides a method for screening and detecting tumor radiotherapy sensitivity markers to obtain tumor radiotherapy sensitivity markers.

[0059] The present invention also provides an application of a tumor radiotherapy sensitivity marker in the preparation of a reagent for detecting tumor radiotherapy sensitivity.

[0060] In the present invention, the tumor radiotherapy sensitivity marker is a metabolic marker. The present invention detects tumor radiotherapy sensitivity through metabolic markers. The present invention predicts the radiotherapy sensitivity of colorectal cancer patients through the critical value of the logistic regression model constructed by the tumor radiotherapy sensitivity marker, and the determination method is: the logistic regression model constructed by the serum sample to be tested vs the serum sample of the colorectal cancer patient who is not sensitive to radiotherapy predicts the optimal threshold value of P, when P < 0.6, it indicates that the tumor patient is insensitive to radiotherapy, when P ≥ 0.6, it indicates that the tumor patient is moderately sensitive to radiotherapy or highly sensitive to radiotherapy;

[0061] In order to further determine whether the tumor patient is moderately sensitive or highly sensitive to radiotherapy, the logistic regression model constructed by comparing the serum samples to be tested with P≥0.6 with the serum samples of colorectal cancer patients with moderate sensitivity to radiotherapy predicts the optimal threshold of P. When P<0.44, it indicates that the tumor patient is moderately sensitive to radiotherapy, and when P≥0.44, it indicates that the tumor patient is highly sensitive to radiotherapy. At the same time, the AUC curve was drawn according to the logistic regression model. The results showed that the metabolic markers: prHS+prMS vs prLS AUC reached 0.914, and prHS vs prMS AUC was 0.817. It can be seen that the tumor radiotherapy sensitivity marker of the present invention has high accuracy in detecting tumor radiotherapy sensitivity.

[0062] The tumor is preferably colorectal cancer.

[0063] The technical solutions provided by the present invention are described in detail below in conjunction with the embodiments, but they should not be construed as limiting the protection scope of the present invention.

[0064] Example 1

[0065] A method for screening and detecting metabolic markers of tumor radiotherapy sensitivity, comprising the following steps:

[0066] (1) Metabolomics sample collection plan

[0067] The blood was collected in a centrifuge tube and allowed to stand at 37°C (or room temperature) for 1 hour for coagulation and stratification. Then, the blood was centrifuged at 3000 rpm for 10 minutes at room temperature, and the supernatant was transferred to a clean centrifuge tube. The blood was centrifuged at 12000 rpm for 10 minutes at 4°C, and the supernatant was dispensed into 1.5 mL centrifuge tubes to obtain serum, which was stored at -80°C.

[0068] (2) Metabolomics serum sample pre-treatment plan

[0069] The serum sample of step (1) was transferred to an EP tube, and an extract (Baiqu Biotechnology, Shanghai) (containing an isotope-labeled internal standard mixture) was added, vortexed for 30 seconds, ultrasonicated in an ice-water bath for 10 minutes, and allowed to stand at -40°C for 1 hour; the sample was centrifuged at 4°C for 15 minutes to obtain the supernatant, which was placed in a sample injection bottle for detection;

[0070] (3) Metabolomics serum sample experimental detection plan

[0071] Ultra-high performance liquid chromatography can efficiently separate various mixtures and perform qualitative analysis on each component in the sample by measuring peak area or peak height.

[0072] The supernatant obtained in step (2) was chromatographed using a Vanquish (Thermo Fisher Scientific) ultra-high performance liquid chromatograph with a Waters ACQUITY UPLCHSS T3 (2.1 mm × 100 mm, 1.8 μm) liquid chromatography column. Phase A of the liquid chromatography was an aqueous phase containing 5 mmol / L ammonium acetate and 5 mmol / L acetic acid, and phase B was acetonitrile. Sample plate temperature: 4°C, injection volume: 2 μL. The Orbitrap Exploris 120 mass spectrometer was capable of primary and secondary mass spectrometry data acquisition under the control of the control software (Xcalibur, version: 4.4, Thermo).

[0073] The raw data were converted into mzXML format using ProteoWizard software, and then processed using the R package (with XCMS as the kernel) for peak identification, peak extraction, peak alignment, and integration. They were then matched with the BiotreeDB (V2.1) mass spectrometry database for substance annotation, and the cutoff value for algorithm scoring was set to 0.3.

[0074] In practical applications, ① multiple cohorts of serum samples were obtained and classified into 6 categories, including prHS (pre-radiotherapy high sensitivity), prMS (pre-radiotherapy medium sensitivity), prLS (pre-radiotherapy low sensitivity), poHS (post-radiotherapy high sensitivity), poMS (post-radiotherapy medium sensitivity), and poLS (post-radiotherapy low sensitivity). Specifically, prHS stands for: sensitive data before radiotherapy (a subset of highly sensitive metabolites before radiotherapy), prMS stands for: moderately sensitive data before radiotherapy (a subset of moderately sensitive metabolites before radiotherapy), prLS stands for: insensitive data before radiotherapy (a subset of insensitive metabolites before radiotherapy), poHS stands for: sensitive data after radiotherapy (a subset of highly sensitive metabolites after radiotherapy), poMS stands for: moderately sensitive data after radiotherapy (a subset of moderately sensitive metabolites after radiotherapy), and poLS stands for: insensitive data after radiotherapy (a subset of insensitive metabolites after radiotherapy).

[0075] ② Combined analysis of positive and negative ion modes.

[0076] The data of 185 serum samples were divided into positive ion mode and negative ion mode, including 19 samples in poHS group, 5 samples in poLS group, 21 samples in poMS group, 56 samples in prHS group, 30 samples in prLS group, and 54 samples in prMS group. Among them, there were 958 substances in positive ion mode and 475 substances in negative ion mode.

[0077] The serum sample data results include MS2name (the substance name obtained by the secondary mass spectrometry qualitative matching analysis) and MS2score (the score of the secondary mass spectrometry qualitative matching, with a value of [0, 1]). The data are merged in the positive and negative ion modes. For substances with duplicate names, i.e., repeated MS2names, substances with high MS2score secondary mass spectrometry qualitative matching scores are retained, and those with low MS2scores are deleted, and the data in the sample is retained; when the scores of substances with duplicate names are the same, substances with high peak areas in the QC (quality control) samples, i.e., substances with high QC sample data, are retained, and those with low peak areas are deleted. MS2name and MS2score are retained, and only the data in the sample is retained. The positive and negative ions are merged to obtain the merged data set, i.e., the initial sample set. There are 1,328 substances after the positive and negative ions are merged.

[0078] Step S1: normalizing the initial sample set to obtain a processed sample set, specifically including:

[0079] A. Filter the deviation value of the initial sample set to obtain a first filtered sample set:

[0080] The deviation values ​​were filtered based on the relative standard deviation, and the relative standard deviation (rsd) was calculated for the QC group, i.e., the standard deviation divided by the corresponding mean value multiplied by 100%. The detection in the QC group with rsd greater than 0.3 was considered unstable, and all detection data of the metabolite were deleted to obtain the first filtered sample set.

[0081] B. Filter missing values ​​on the first filtered sample set to obtain a second filtered sample set:

[0082] Missing value filtering: only retain a single group of null values, i.e., peak area data with null values ​​of no more than 50% in each group of prHS, prMS, prLS, poHS, poMS, and poLS or null values ​​of no more than 50% in all groups, to obtain the second filtered sample set;

[0083] C. Fill missing values ​​in the sample set after the second filtering to obtain a filled sample set:

[0084] Performing minimum half filling on the missing values ​​in the second filtered sample set to obtain a filled sample set;

[0085] D. Perform normalization processing on the padded sample set to obtain a processed sample set:

[0086] The total ion current of each sample in the filled sample set was used for normalization to obtain the processed sample set. After normalization, there were 1192 substances in total.

[0087] Then, the processed sample set is used to train the initial model, specifically including:

[0088] Step S2: using principal component analysis and orthogonal partial least squares method to process the processed sample set to obtain an initial differential metabolite sample set;

[0089] Through the processed sample set, principal component analysis was performed for three different comparison strategies: prHS vs. prMS, prHS vs. prLS, and prMS vs. prLS. The differences between groups and the distribution of samples were reflected by the scatter plot of the principal component scores. The closer the sample points were, the more similar the content and type of metabolites in the samples were. Conversely, the farther the samples were, the greater the difference in their overall metabolic levels. The results were further analyzed by the statistical method of orthogonal partial least squares discriminant analysis. Through orthogonal partial least squares discriminant analysis, orthogonal variables that were not related to the categorical variables in the metabolites can be filtered out, and non-orthogonal variables and orthogonal variables can be analyzed separately, so as to obtain more reliable information on the correlation between the inter-group differences of metabolites and the experimental groups. The difference multiples between the two groups were compared by calculating the FoldChange value. The difference multiples between the two groups of substances greater than 1 represent upregulation, and less than 1 represent downregulation. After the above analysis, combined with the statistical analysis results of unit variables and multivariate variables, differential metabolites were screened out, and the initial differential metabolite sample set was obtained.

[0090] Step S3: using a rank sum test to preliminarily screen the initial differential metabolite sample set to obtain a second differential metabolite sample set.

[0091] First, establish the null hypothesis: the two groups of samples come from the same population. Mix the two groups of data and rank them uniformly from small to large. When encountering the same data, take the average rank and calculate the statistic. Set the significance level a=0.05, and obtain the probability of the current statistic value under the condition that the null hypothesis is true. If p-value<0.05, reject the null hypothesis, indicating that there is a significant difference between the two samples.

[0092] Specifically, the rank sum test was performed for the four comparisons of prHS and prMS, prHS and prLS, prMS and prLS, and prHS+prMS and prLS. If p-value < 0.05, it means that there is a significant difference between the two samples. After screening, 47, 34, 40, and 36 differential metabolites were obtained between prHS and prMS, prHS and prLS, prMS and prLS, and prHS+prMS and prLS, respectively, as shown in Table 1.

[0093] Table 1 The second differential metabolite sample set after screening between prHS and prMS, prHS and prLS, prMS and prLS, prHS+prMS and prLS

[0094]

[0095]

[0096]

[0097] Step S4: Based on the logistic regression model and P-value established between the differential samples in the second differential metabolite sample set and the significant clinical indicators, the differential substances after eliminating the interference of clinical factors are obtained, that is, the third differential metabolite sample set is obtained.

[0098] The initial clinical indicators were screened using multiple comparisons and PerMANOVA multivariate analysis of variance, and significant clinical indicators were obtained, including:

[0099] a. Obtaining initial clinical indicators:

[0100] A total of 185 clinical samples were collected, and the clinical indicator information was extracted, including treatment response group, age, gender, BMI, smoking, drinking, T, N, GTV, distance from the anus in cm (MR / CT), CRM (MR), EMVI (MR), NLR, concurrent chemotherapy regimen, hormone use, and antibiotic use, a total of 16 items.

[0101] b. Screening the initial clinical indicators using the multiple comparison method to determine the second clinical indicator:

[0102] The multiple comparison method was used to retain the samples with p-value < 0.05 in the statistical test results of the above 16 clinical indicators. The p-value was obtained by performing multiple comparisons on the clinical indicators among multiple experimental groups, firstly testing their homogeneity by BartlettTest (corresponding to the R language program function bartlett.test) (taking p-value < 0.05 as a non-null hypothesis), and then calculating their p-values ​​by the oneway.test function of the R language program package. If p < 0.05, it means that the clinical indicator is significant.

[0103] c. Use the PerMANOVA multivariate analysis of variance method to screen the second clinical indicator and determine the significant clinical indicator.

[0104] The permaonva analysis method was used to further screen out clinical indicator information that had a significant relationship with the metabolic profile expression of human serum samples (i.e., the second clinical indicator);

[0105] d. Using the logistic regression model, the significant clinical index information obtained in step c is used as the response variable to perform a final correction on the second differential metabolite sample set obtained in step S3, and the differential substances whose p-value after regression is still less than 0.05 are screened out to obtain the differential substances (third differential metabolite sample set) after eliminating the interference of clinical factors, specifically including:

[0106] The clinically significant index information obtained in step c is used as the response variable, and the second differential metabolite sample set obtained in step S3 is used as the independent variable. A logistic regression model is established to obtain the p-value. If the p-value of the differentially expressed substance is less than 0.05, it means that these substances are not affected by the clinical indicators, that is, they are differential substances after eliminating the interference of clinical factors.

[0107] In the prHS+prMS and prLS groups, the clinical indicator smoking was found to be significant using the permaonva analysis method, and the 36 differentially expressed substances screened were respectively used to establish logistic regression with the clinical indicator smoking. Among these 36 differentially expressed substances, 18 had p values ​​less than 0.05, namely ACar (11:1), H-PHE-PRO-OH, Thyroxine, ACar (13:0), ACar (9:0), ACar (13:1), SM (d14:0 / 18:2), 2-Methylglutaricacid, N1-Methyl-2-pyridone-5-carboxamide, N1-Methyl-4-pyridone-3-carboxamide, Dethiobiotin, 3-Hydroxyisovalericacid, [7]-Paradol, I-Urobilin, (-)-Catechin3-O-gallate, Isopentenyl acetate, Desglucocoroloside, and Spirotaccagenin.

[0108] All clinical indicators obtained by permaonva analysis method for prHS and prMS, prHS and prLS, and prMS and prLS groups were not significant. The 47, 34, and 40 differentially expressed substances in the initial screening did not need to establish logistic regression with all clinical indicators respectively. Therefore, the 47, 34, and 40 differentially expressed substances in the second differential metabolite sample set were the differential substances after excluding the interference of clinical factors.

[0109] e. For the above four comparison strategies of prHS and prMS, prHS and prLS, prMS and prLS, and prHS+prMS and prLS, the differential substances after eliminating the interference of clinical factors were merged to form the third differential metabolite sample set, and three different machine learning algorithms, random forest method, support vector machine method, and LASSO, were used for further screening and analysis. Based on the key candidate differential substance combinations finally screened out by these three different algorithms, joint ROC analysis was carried out, and finally the final differential substance combination panel was determined according to the obtained AUC value, specifically:

[0110] First, the third differential metabolite sample set was divided into a training set and a test set (the ratio of training set to test set was 8:2), and different algorithms were used to screen differential substances, as follows:

[0111] Random forest method: Use the R package randomForest to fit the random forest model to the training set, and finally determine the treeNum in the fitting model through parameter tuning. Fit the training set with the optimal parameters, and test the test set to obtain the confusion matrix of the prediction results. Further, use the rfcv function in the R package randomForest to perform 10-fold cross validation to obtain the relationship between variable selection and the average error rate of model classification, and obtain the optimal differential material based on MeanDecreaseAccuracy. The larger the value, the greater the importance of the variable.

[0112] Taking prHS and prMS as an example, there are 110 samples. According to the 47 differentially expressed substances obtained above, the training set (88) and the test set (22) are distinguished according to 8:2. The treeNum in the fitting model is finally determined to be 500 through parameter tuning. The rfcv function performs 10-fold cross validation to obtain the lowest average error rate of 34.1% when the number of variables is 47, which are Ovalicin, 3-Hydroxydecanoic acid, Glutamylthreonine, Diethylfumarate, Tetracosahexaenoic acid, 3 -Hydroxymethylglutaricacid, Estrone, SM(d14:0 / 28:2), (3beta,5alpha,6beta,9alpha,22E,24R)-23-Methylergosta-7,22-die ne-3,5,6,9-tetrol, (10E,12Z)-9-HODE, FAHFA(18:0 / 22:3), Pisatoside, 2-Isopropyl-3,5-dimethoxy-6-methylpyrazine, PC(16 :3 / 24:4), Hydroxyisocaproicacid, Apo-12'-violaxanthal, Fexofenadine, Gossypetin8-glucuronide3-sulfate, Cholicacid, 4- Acetylbutyrate, 3-(2-Furanylmethylene)pyrrolidine, Lauroyldiethanolamide, Cuscohygrine, PC(14:0e / 18:0), PE(22:6(4Z,7 Z,10Z,13Z,16Z,19Z) / 16:0), Riboflavin, (-)-(E)-1-(4-Hydroxyphenyl)-7-phenyl-6-hepten-3-ol, (R)-3-Hydroxy-tetradecan oicacid, Vulgarole, Isoliquiritigenin, Pyroglutamicacid, PC(22:2(13Z,16Z) / 16:1(9Z)), Serylphenylalanine, Prostaglandin F1a, PC(16:1e / 20:4), Cholesterol, Mizolastine, (4E)-1,7-bis(4-hydroxyphenyl)hept-4-en-3-one, 4-Pentenoicacid, Nortriptyline, 8-IsoprostaglandinE1, Pipernonaline, 4-Deoxyerythronicacid, Piperolein B, Triethylcitrate, 15-Keto-prostaglandinE2, PC(16:0 / 20:3). ,

[0113] Taking prHS and prLS as an example, there are 86 samples. According to the 34 differentially expressed substances obtained above, the training set (70) and the test set (17) are distinguished according to 8:2. The treeNum in the fitting model is finally determined to be 500 through parameter tuning. The rfcv function performs 10-fold cross validation to obtain the lowest average error rate of 22.1% when the number of variables is 34, namely ACar (11:1), D-Glutamine, Thyroxine, 1-Methyl-L-Histidine, Uridine, ACar (13:0), PC (14:0e / 18:0), ACar (9:0), ACar (13:1), SM (d14:0 / 18:2), Deoxycholicacid, Aspartame, Tetracosahexaenoicacid, PC (22:6 (4Z, 7Z, 10Z, 13Z, 16Z, 19Z) / 20:3 (5Z, 8Z, 11Z)) , ACar(9:1), 2-Methylglutaricacid, PC(22:2(13Z,16Z) / 16:1(9Z)), Butylparaben, Riboflavin, Uraci l. PC(22:6(4Z,7Z,10Z,13Z,16Z,19Z) / 18:3(6Z,9Z,12Z)), ACar(15:0), 3-Hydroxyisovalerylcarnitin e. Nicotinamide, Dehydrophytosphingosine, Ricinoleicacid, S-(2-carboxypropyl)-Cysteamine, 3-Hydroxyisovalericacid, beta-Citraurinepoxide, Selenomethionine, (-)-Catechin3-O-gallate, Butyl Levulinate, Desglucocoroloside, Spirotaccagenin.

[0114] Taking prMS and prLS as an example, there are 84 samples. According to the 40 differentially expressed substances obtained above, the training set (67) and the test set (17) are distinguished according to 8:2. The treeNum in the fitting model is finally determined to be 500 through parameter tuning. The rfcv function performs 10-fold cross validation to obtain the lowest average error rate of 29.9% when the number of variables is 40, which are Isopentenylacetate, 2-Methylglutaric acid acid, Nicotinamide, ACar(9:0), Vulgarole, Spirotaccagenin, Dhelwangin, Phenylglyoxylicacid, 7-Ketocholesterol, (3beta,5alpha,6beta,9alpha,22E,2 4R)-23-Methylergosta-7,22-diene-3,5,6,9-tetrol,Carbendazim,Oleamide,H-PHE-PRO-OH,I-Urobilin,MethylIndole-3-acetate,15-Keto-prostaglandin E2, Desglucocoroloside, PC(9:0 / 26:1), Aspartame, ACar(11:1), 3-Methyl-2-oxopentanoate, 1-Hydroxy-6-methoxypyrene, Dihy drofolicacid, ACar(13:0), LysoPC(P-16:0), Dethiobiotin, beta-Citraurinepoxide, FAHFA(18:3 / 22:3), SM(d14:0 / 22:2), N-Acet ylglutamicacid, Xanthurenicacid, [7]-Paradol, 8,15-DiHETE, 2-Isopropyl-3,5-dimethoxy-6-methylpyrazine, 7-Methylguano sine, N1-Methyl-2-pyridone-5-carboxamide, ACar(13:1), PC(16:0e / 22:6), N1-Methyl-4-pyridone-3-carboxamide, L-Urobilin.

[0115] Taking prHS+prMS and prLS as examples, there are 140 samples. According to the 18 differentially expressed substances obtained above, the training set (112) and the test set (28) were distinguished according to 8:2. The treeNum in the fitting model was finally determined to be 500 through parameter tuning. The rfcv function performed 10-fold cross validation to obtain the lowest average error rate of 17.9% when the number of variables was 18, which were ACar (11:1), H-PHE-PRO-OH, Thyroxine, ACar (13:0), ACar (9:0), ACar (13:1), SM (d14:0 / 18:2), 2-Methylglutaric acid acid, N1-Methyl-2-pyridone-5-carboxamide, N1-Methyl-4-pyridone-3-carboxamide, Dethiobiotin, 3-Hydroxyisovalericacid, [7]-Paradol, I-Urobilin, (-)-Catechin3-O-gallate, Isopentenylacetate, Desglucocoroloside, Spirotaccagenin.

[0116] Support vector machine (SVM): Support vector machine is a two-class classification model. Its basic model is a linear classifier that finds the separating hyperplane with the maximum margin in the feature space. SVM is an algorithm for the post-selection of a sequence based on the maximum margin principle. In the first iteration, the SVM model is optimized for all feature sets in the data set. Then the score of each feature is calculated and sorted in descending order. The feature set with the smallest score is recorded, and then the feature with the smallest score is deleted. It is iterated again until only one feature is left. The training set is trained using the svm function of the R language to obtain the relationship between variable selection and the average error rate of model classification, and the feature set with the lowest error rate is selected.

[0117] Taking prHS and prMS as an example, there were 110 samples. According to the 47 differentially expressed substances obtained above, the training set (88) and the test set (22) were distinguished according to 8:2. The training set was trained using the svm function of R language, and the lowest error rate was 35.2% when the variables were Diethylfumarate, Pisatoside, Pyroglutamic acid, SM (d14:0 / 28:2), 4-Pentenoic acid, Pipernonaline, 4-Deoxyerythronic acid, FAHFA (18:0 / 22:3), PC (16:1e / 20:4), 4-Acetylbutyrate, Fexofenadine, PC (16:0 / 20:3), 3-(2-Furanylmethylene)pyrrolidine, PC (16:3 / 24:4), Cholesterol, Mizolastine, Glutamylthreonine, and Apo-12'-violaxanthal.

[0118] Taking prHS and prLS as an example, there were 86 samples in total. According to the 34 differentially expressed substances obtained above, the training set (70) and the test set (17) were distinguished according to 8:2. The training set was trained using the svm function of R language, and the lowest error rate was 20.6% when the variables were Spirotaccagenin, Selenomethionine, Uracil, Uridine, Desglucocoroloside, (-)-Catechin3-O-gallate, PC (22:6 (4Z, 7Z, 10Z, 13Z, 16Z, 19Z) / 18:3 (6Z, 9Z, 12Z)), Butylparaben, beta-Citraurinepoxide, 3-Hydroxyisovalerylcarnitine, 1-Methyl-L-Histidine, Thyroxine, 2-Methylglutaricacid, Nicotinamide, ACar (9:1), and ACar (13:0).

[0119] Taking prMS and prLS as an example, there are 84 samples. According to the 40 differentially expressed substances obtained above, the training set (67) and the test set (17) were distinguished according to 8:2. The training set was trained using the R language svm function, and the variables Spirotaccagenin, Desglucocoroloside, Dihydrofolicacid, Dhelwangin, Isopentenylacetate, 1-Hydroxy-6-methoxypyrene, I-Urobilin, and Carben dazim, [7]-Paradol, beta-Citraurinepoxide, 2-Isopropyl-3,5-dimethoxy-6-methylpyrazine, Dethiobiotin, N1-Methyl- 4-pyridone-3-carboxamide, (3beta,5alpha,6beta,9alpha,22E,24R)-23-Methylergosta-7,22-diene-3,5,6,9-tetrol, N-A cetylglutamicacid, Nicotinamide, 15-Keto-prostaglandinE2, 7-Methylguanosine, 7-Ketocholesterol, Xanthurenicacid, Vulgarole, N1-Methyl-2-pyridone-5-carboxamide, 3-Methyl-2-oxopentanoate, LysoPC(P-16:0), 2-Methylglutaricacid, The lowest error rate was 16.4% when using L-Urobilin, Oleamide, Aspartame, Phenylglyoxylicacid, MethylIndole-3-acetate, PC(9:0 / 26:1), PC(16:0e / 22:6), ACar(13:1), ACar(9:0), 8,15-DiHETE, ACar(13:0), FAHFA(18:3 / 22:3), SM(d14:0 / 22:2), H-PHE-PRO-OH, and ACar(11:1).

[0120] Taking prHS+prMS and prLS as examples, there were 140 samples. According to the 18 differentially expressed substances obtained above, the training set (112) and the test set (28) were distinguished according to 8:2. The training set was trained using the svm function of R language, and the lowest error rate was 16.1% when the variables were Isopentenyl acetate, (-)-Catechin3-O-gallate, 2-Methylglutaricacid, Desglucocoroloside, ACar (13:1), H-PHE-PRO-OH, Thyroxine, [7]-Paradol, and N1-Methyl-2-pyridone-5-carboxamide.

[0121] LASSO: LASSO imposes regression penalties on all variables, making the coefficients of relatively unimportant variables become 0, thus being excluded from modeling, and then selecting independent variables that have a greater impact on the dependent variable and calculating the corresponding regression coefficients, and finally obtaining a prediction model. The model is trained on the training set using the R language glmnet. The lambda value with the smallest mean square error is selected through the mean square error under different parameters lambda. A vertical line is drawn through the position of the corresponding penalty value of the change trajectory diagram of each independent variable coefficient. Each curve represents a variable, and the variable that intersects with the penalty value is the variable finally included in the model. The ordinate corresponding to the variable is the regression coefficient of the variable, and the regression coefficient is also the contribution of the variable.

[0122] Taking prHS and prMS as an example, there are 110 samples. According to the 47 differentially expressed substances obtained above, the training set (88) and the test set (22) were distinguished according to 8:2. The training set was trained using the R language glmnet, and 28 differential metabolites were obtained: (R)-3-Hydroxy-tetradecanoic acid, Lauroyldiethanolamide, FAHFA (18:0 / 22:3), Hydroxyisocaproic acid, PC (16:3 / 24:4), PC (14:0e / 18:0), Cholic acid, 3 -(2-Furanylmethylene)pyrrolidine, SM(d14:0 / 28:2), Serylphenylalanine, PC(16:0 / 20:3), Tetracosahexaenoicacid, PC(22:2(13Z, 16Z) / 16:1(9Z)), Vulgarole, Riboflavin, 4-Acetylbutyrate, Gossypetin8-glucuronide3-sulfate, Estrone, 3-Hydroxymethylglutaric acid, (4E)-1,7-bis(4-hydroxyphenyl)hept-4-en-3-one, Pisatoside, (3beta,5alpha,6beta,9alpha,22E,24R)-23-Methylergosta-7,22-diene-3,5, 6,9-tetrol, Diethylfumarate, 2-Isopropyl-3,5-dimethoxy-6-methylpyrazine, PE(22:6(4Z,7Z,10Z,13Z,16Z,19Z) / 16:0), Ovalicin, Glutamylthreo Nine, Isoliquiritigenin, the specific regression equation is: Y = 0.176630112-0.481373480×(R)-3-Hydroxy-tetradecanoic acid+0.189970030×Lauroyldiethanolamide+0.294612153×FAHFA(18:0 / 22:3)-0.114201224×Hydroxyisocaproic acid-0.329202341×PC(16:3 / 24:4)+0.063337880×PC(14:0e / 18:0)-0.202945506×Cholic acid

[0123] acid+0.402645810×3-(2-Furanylmethylene)pyrrolidine+0.068413260×SM(d14:0 / 28:2)-0.453305087×Ser ylphenylalanine+0.245675693×PC(16:0 / 20:3)-0.547027447×Tetracosahexaenoicacid+0.020912311×PC(22:2(13Z,16Z) / 16:1(9Z))+0.220321552×Vulgarole+0.049755862×Riboflavin+0.325673143×4-Acetylbutyrate+0.112288678×Gossypetin8-glucuronide3-sulfate-0.014118167×Estrone-0.621134893×3-Hydroxymethylglutaricacid+0.158254558×(4E)-1,7-bis(4-hydroxyphenyl)hept-4-en-3-one+0.237417433×Pisatoside-0.052592560×(3beta,5alpha,6beta,9alpha,22E,24R)-23-Methylergosta-7,22-diene-3,5,6,9-tetrol+0.265102112×Diethy lfumarate-0.303018854×2-Isopropyl-3,5-dimethoxy-6-methylpyrazine+0.131173867×

[0124] PE(22:6(4Z,7Z,10Z,13Z,16Z,19Z) / 16:0)+0.196723739×Ovalicin-0.581576171×Glutamylthreonine-0.371153252×Isoliquiritigenin。

[0125] Taking prHS and prLS as an example, there are 86 samples. According to the 34 differentially expressed substances obtained above, the training set (70) and the test set (17) were distinguished according to 8:2. The training set was trained by R language glmnet, and 21 differential metabolites were obtained: Thyroxine, Uridine, ACar (13:0), PC (14:0e / 18:0), SM (d14:0 / 18:2), Deoxycholicacid, Aspartame, Tetracosahexaenoic acid, PC (22:6 (4Z, 7Z, 10Z, 13Z, 16Z, 19Z) / 20:3 (5Z, 8Z, 11Z)), ACar (9:1), 2-Methylglutaricacid, Butylparaben, Riboflavin, Nicotinamide, Dehydrophytosphingosine, Ricinoleic acid acid, beta-Citraurinepoxide, (-)-Catechin3-O-gallate, Butyllevulinate, Desglucocoroloside, Spirotaccagenin, the specific regression equation is: Y = 0.945592741 + 0.425529358 × Thyroxine + 0.434334101 × Uridine + 0.1416733 × ACar (13: 0) + 0.122174358 × PC (14: 0e / 18: 0) + 0.023755087 × SM (d14: 0 / 18: 2) - 0.433318642 × Deoxycholic

[0126] acid+0.139953171×Aspartame-0.042631334×Tetracosahexaenoicacid+0.01596018×PC(22:6(4Z,7 Z,10Z,13Z,16Z,19Z) / 20:3(5Z,8Z,11Z))+0.040489289×ACar(9:1)-0.472540206×2-Methylglutaric

[0127] acid-0.331792329×Butylparaben+0.019591005×Riboflavin+0.014836255×Nicotinamide-0.002452764×Dehydrophytosphingosine-0.475405577×Ricinoleicacid-0.025951223×beta-Citraurin epoxide+0.489435332×(-)-Catechin 3-O-gallate-0.058483768×Butyl levulinate+0.003892541×Desglucocoroloside+0.056270747×Spirotaccagenin.

[0128] Taking prMS and prLS as an example, there are 84 samples. According to the 40 differentially expressed substances obtained above, the training set (67) and the test set (17) are distinguished according to 8:2. The training set is trained by using R language glmnet, and the 24 differential metabolites are obtained: The specific regression equation is: Y = 2.99887369337435 + 1.10976468382751 × H-PHE-PRO- OH+0.920014445806788×ACar(13:0)+0.459808787978451×8,15-DiHETE+1.1340621781099 1×ACar(9:0)+1.17299463874812×PC(16:0e / 22:6)-0.959473151493842×Phenylglyoxylic

[0129] acid+2.92897693190931×Oleamide-1.83958237987782×L-Urobilin-3.65659024396626×2-Methylglutari cacid+0.0473198495987288×N1-Methyl-2-pyridone-5-carboxamide-1.2340219781403×Vulgarole+0.704837527183466×15-Keto-prostaglandin

[0130] E2+0.894224083216816×Nicotinamide-0.354283446737299×(3beta,5alpha,6beta,9alpha,22E,24R)-23-Methylergosta-7,22-diene-3,5,6,9-tetrol+2.69118679210089×Dethiobiotin-0.300262274515251×2-Isopropy l-3,5-dimethoxy-6-methylpyrazine-0.816645076823208×beta-Citraurinepoxide-0.84713756840258×I-Urobilin+1.62670013495577×1-Hydroxy-6-methoxypyrene+1.58421576632725×Isopentenylacetate-2.5391338060763×Dhelwangin+1.36756431423767×Dihydrofolicacid+0.055139707565556×Spirotaccagenin。

[0131] Taking prHS+prMS and prLS as examples, there are 140 samples. According to the 18 differentially expressed substances obtained above, the training set (112) and the test set (28) were distinguished according to 8:2. The training set was trained using the R language glmnet model, and 15 differential metabolites were obtained: ACar (11:1), H-PHE-PRO-OH, Thyroxine, ACar (13:0), ACar (13:1), 2-Methylglutaricacid, N1-Methyl-2-pyridone-5-carboxamide, Dethiobiotin, 3-Hydroxyisovalericacid, [7]-Paradol, I-U robilin, (-)-Catechin3-O-gallate, Isopentenylacetate, Desglucocoroloside, Spirotaccagenin, the specific regression equation is: Y = 2.408973956 + 0.059529776 × ACar (11: 1) + 0.173874078 × H-PHE-PRO-OH + 0.581385965 × Thyroxine + 0.343568892 × ACar (13: 0) + 0.144166563 × ACar (13: 1) - 0.979911099 × 2-Methylglutaricacid + 0.112680502

[0132] ×N1-Methyl-2-pyridone-5-carboxamide+0.20221917×Dethiobiotin+0.551895565×3-Hydroxyisovaleric acid-0.710781395×[7]-Paradol-0.298712382×I-Urobilin+1.325241453×(-)-Catechin3-O-gallate+

[0133] 0.673871343×Isopentenylacetate+0.038857388×Desglucocoroloside+0.446201899×Spirotaccagenin.

[0134] Step S5: The substances obtained by logistic regression, LASSO, support vector machine, and random forest models are combined to draw a joint ROC curve. The larger the AUC value, the better the effect of the model. The optimal model is determined to obtain metabolic markers for screening and detecting tumor radiotherapy sensitivity.

[0135] The present invention screened and detected 13 kinds of metabolic markers for tumor radiotherapy sensitivity: Oleamide (CAS: 3072-13-7), Dihydrofolic acid(CAS:4033-27-6), H-PHE-PRO-OH(CAS:7669-65-0), 2-Methylglutaricacid(CAS:617-62-9), Phenylg lyoxylicacid(CAS:611-73-4), Thyroxine(CAS:51-48-9), 1-Methyl-L-Histidine(CAS:332-80-9), Uridi ne(CAS:58-96-8), D-Glutamine(CAS:5959-95-5), Deoxycholicacid(CAS:83-44-3), Lauroyldiethanolam ide (CAS: 120-40-1), (R)-3-Hydroxy-tetradecanoicacid (CAS: 28715-21-1) and Cholicacid (CAS: 81-25-4). The CAS number in brackets is the internationally used CAS number.

[0136] The panel composed of the above 13 metabolic markers had the largest AUC (area under the curve) value, with prHSvsprLSAUC reaching 0.932, prHSvsprMSAUC being 0.817, prMSvsprLSAUC being 0.931, and prHS+prMSvsprLSAUC being 0.914.

[0137] The above 13 serum metabolite markers obtained by the present invention are expected to become fixed markers for determining the sensitivity of colorectal cancer to radiotherapy, and the judgment is made by collecting patient serum and detecting the above metabolites.

[0138] Example 2

[0139] Application of the tumor radiotherapy sensitivity metabolic marker screened in Example 1 in the preparation of a reagent for predicting the radiotherapy sensitivity of colorectal cancer patients

[0140] The steps of the method for predicting the radiotherapy sensitivity of colorectal cancer patients using the tumor radiotherapy sensitivity metabolic markers screened in Example 1 are as follows:

[0141] The tumor radiotherapy sensitivity metabolic markers screened in Example 1: Oleamide (CAS: 3072-13-7), Dihydrofolic acid (CAS: 4033-27-6), H-PHE-PRO-OH (CAS: 7669-65-0), 2-Methylglutaricacid (CAS: 617-62-9), Phenylglyoxylicacid (CAS: 611-73-4), Thyroxine (CAS: 51-48-9), 1-Methyl-L-histidine (CAS: 332-80-9), Uridine (CAS: 58-96-8), D-Glutamine (CAS: 5959-95-5), Deoxychol The invention discloses a method for predicting the radiotherapy sensitivity of colorectal cancer patients by using a logistic regression model constructed based on the critical value of the 13 tumor radiotherapy sensitivity markers, wherein the reagent is divided into serum samples to be tested, serum samples of colorectal cancer patients who are insensitive to radiotherapy, and serum samples of colorectal cancer patients who are moderately sensitive to radiotherapy. The serum samples of colorectal cancer patients who are insensitive to radiotherapy and serum samples of colorectal cancer patients who are moderately sensitive to radiotherapy are obtained by NAR (Neoadjuvant Rectal) scoring classification.

[0142] Judgment method: The logistic regression model constructed by comparing the serum samples to be tested with the serum samples of colorectal cancer patients who are insensitive to radiotherapy predicts the optimal threshold of P. When P < 0.6, it indicates that the tumor patient is insensitive to radiotherapy. When P ≥ 0.6, it indicates that the tumor patient is moderately sensitive to radiotherapy or highly sensitive to radiotherapy.

[0143] In order to further determine whether the tumor patients are moderately sensitive or highly sensitive to radiotherapy, a logistic regression model was constructed by comparing the serum samples with P ≥ 0.6 with the serum samples of colorectal cancer patients with moderate sensitivity to radiotherapy to predict the optimal threshold of P. When P < 0.44, it indicates that the tumor patients are moderately sensitive to radiotherapy, and when P ≥ 0.44, it indicates that the tumor patients are highly sensitive to radiotherapy.

[0144] In this example, the serum of colorectal cancer patients who are highly sensitive to radiotherapy is used as the serum sample to be tested, and the above method is used to predict the radiotherapy sensitivity of colorectal cancer patients. The results are as follows:

[0145] The logistic regression model of the serum samples to be tested vs the serum samples of colorectal cancer patients who are not sensitive to radiotherapy is logit(P)=log(P / (1-P))=2.289+0.6681-Methyl-L-Histidine+0.438Dihydrofolicacid+3.151H-PHE-PRO-OH+0.664Oleamide+4.197Thyroxine+1.631Uridine-0.354Deoxycholicacid+0.14 9Cholicacid-0.171(R)-3-Hydroxy-tetradecanoicacid+1.157D-Glutamine+0.772Lauroyldiethanolamide-6.964Phenylglyoxylicacid-2.8532-Methylglutaricacid, the optimal threshold (or cutoff value) for predicting P is 0.6, indicating that the colorectal cancer patients verified in this example are moderately sensitive to radiotherapy or highly sensitive to radiotherapy.

[0146] The serum of the above colorectal cancer patients was further used as the serum sample to be tested vs the serum sample of the colorectal cancer patients with moderate sensitivity to radiotherapy. The logistic regression model was logit(P)=log(P / (1-P))=0.119+1.9Lauroyldiethanolamide+0.9541-Methyl-L-Histidine-0.263Cholicacid+2.368Uridine-0.77H-PHE-PRO-OH+0.945Phenylglyoxylicac id+1.632-Methylglutaricacid-2.618(R)-3-Hydroxy-tetradecanoicacid+0.462Thyroxine-0.145D-Glutamine+2.474Deoxycholicacid-0.237Oleamide-0.705Dihydrofolicacid, the optimal threshold (or cutoff value) for predicting P is 0.44, indicating that the colorectal cancer patients verified in this example are highly sensitive to radiotherapy.

[0147] At the same time, the AUC curve is drawn according to the logistic regression model. Figure 1-2 The results showed that the metabolic markers: prHS+prMSvsprLS AUC reached 0.914, and prHSvsprMS AUC was 0.817.

[0148] Example 3

[0149] The inclusion criteria of the Fujian validation cohort and the Guangzhou validation cohort included: (1) patients who received surgical treatment after neoadjuvant radiotherapy or chemoradiotherapy; (2) the radiotherapy regimen was long-term radiotherapy; the exclusion criteria included: (1) any other type of malignant tumor or serious immune, nervous, digestive, or blood system diseases; (2) patients who could not obtain plasma samples; (3) patients with colorectal adenocarcinoma that had metastasized; and (4) patients who received short-term radiotherapy.

[0150] According to the above inclusion and exclusion criteria, the number of people collected in the Fujian validation cohort was 91, and the number of people collected in the Guangzhou validation cohort was 62. Then, by performing targeted metabolomics to detect metabolite levels of 9 metabolites (i.e., H-Phe-Pro-OH, 2-Methylglutaric acid, Levothyroxine, 1-Methyl-L-Histidine, Uridine, D-Gln-OH, Deoxycholicacid, (R)-3-Hydroxy-tetradecanoicacid, Cholicacid) among the above 13 serum metabolite markers, the ROC curve was drawn according to the logistic regression model constructed by the method of Example 2, and the Fujian validation cohort: Figure 3 The results showed that the AUCs of the serum samples to be tested vs the serum samples of colorectal cancer patients who were not sensitive to radiotherapy and the serum samples to be tested vs the serum samples of colorectal cancer patients who were moderately sensitive to radiotherapy were 0.749 and 0.784, respectively; Guangzhou validation cohort: Figure 4 The results showed that the AUCs of the serum samples to be tested vs the serum sample group of colorectal cancer patients who were insensitive to radiotherapy and the serum sample group of colorectal cancer patients who were moderately sensitive to radiotherapy were 0.772 and 0.803, respectively.

[0151] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for screening and detecting tumor radiotherapy sensitivity markers, characterized in that: The following steps are involved: S1, normalize the initial sample set to obtain a processed sample set; S2. Processing the processed sample set using principal component analysis and orthogonal partial least squares method to obtain an initial difference substance sample set; S3, using a rank sum test to preliminarily screen the initial difference sample set to obtain a second difference metabolite sample set; S4. Use multiple comparison method and PerMANOVA multivariate analysis of variance method to screen the initial clinical indicators of the cohort samples of the prHS and prMS, prHS and prLS, prMS and prLS comparison experimental groups. If one or more clinical indicators in the comparison experimental group are not significant, the corresponding differential samples in the second differential metabolite sample set are differential substances after excluding the interference of clinical factors; if one or more clinical indicators in the comparison experimental group are significant, the differential samples in the corresponding second differential metabolite sample set are compared with the logistic regression model established with the significant clinical indicators and the P-value of the logistic regression model, and the differential substances with p-value < 0.05 are screened to obtain the differential substances after excluding the interference of clinical factors; the differential substances after excluding the interference of clinical factors are combined as the third differential metabolite sample set; S5. Logistic regression, LASSO, support vector machine, and random forest model were used to screen and analyze the third differential metabolite sample set, and then the substances obtained by logistic regression, LASSO, support vector machine, and random forest model were combined to draw a joint ROC curve. The maximum AUC value was used as the optimal model to obtain a tumor radiotherapy sensitivity marker.

2. The method according to claim 1, characterized in that In step S1, before normalization, the following steps are also performed: Perform deviation value filtering on the initial sample set to obtain a first filtered sample set; Perform missing value filtering on the first filtered sample set to obtain a second filtered sample set; Filling missing values ​​in the sample set after the second filtering to obtain a filled sample set; In step S1, the padded sample set is normalized to obtain a processed sample set; In step S2, the principal component analysis method includes: using the processed sample set, performing principal component analysis on three different comparison strategies, namely, comparing prHS with prMS, comparing prHS with prLS, and comparing prMS with prLS, respectively, and reflecting the differences between groups and the distribution of samples through the principal component score scatter plot; In step S3, the initial screening method is to use a rank sum test to calculate the hypothesized value p-value in the initial differential substance sample set, and use the differential substances with p-value < 0.05 as the screening differentially expressed substances to obtain the second differential metabolite sample set; In step S4, the significant clinical indicators are screened and determined by using the multiple comparison method and PerMANOVA multivariate analysis of variance method to screen the initial clinical indicators; the initial clinical indicators include treatment response grouping, age, gender, body mass index, smoking amount, alcohol consumption, primary tumor T stage, regional lymph node N stage, tumor target area, cm from the anus, circumferential resection margin, extraintestinal vascular invasion, neutrophil-lymphocyte ratio, concurrent chemotherapy regimen, hormone usage and antibiotic usage.

3. The method according to claim 1 or 2, characterized in that: The initial sample set is obtained by preprocessing the samples, and the sample preprocessing method includes: The obtained multi-cohort samples were classified into 6 categories according to the NAR score and before and after radiotherapy: sensitive data prHS before radiotherapy, moderately sensitive data prMS before radiotherapy, insensitive data prLS before radiotherapy, sensitive data poHS after radiotherapy, moderately sensitive data poMS after radiotherapy, and insensitive data poLS before and after radiotherapy; All sample data were analyzed in positive and negative ion modes. For substances with duplicate names obtained by secondary mass spectrometry qualitative matching analysis, substances with high scores were retained, and substances with low scores were deleted, and the data in the samples were retained. When the scores of duplicate substances were the same, substances with high peak areas in the quality control samples were retained, and substances with low peak areas were deleted, and only the data in the samples were retained to obtain the initial sample set.

4. The method according to claim 3, characterized in that The samples include serum samples; When the sample is a serum sample, the tumor radiotherapy sensitivity markers screened by the method are metabolic markers, and the metabolic markers include Oleamide, Dihydrofolic acid, H-PHE-PRO-OH, 2-Methylglutaric acid, Phenylglyoxylic acid, Thyroxine, 1-Methyl-L-Histidine, Uridine, D-Glutamine, Deoxycholic acid, Lauroyl diethanolamide, (R)-3-Hydroxy-tetradecanoic acid and Cholic acid.

5. The method according to claim 4, characterized in that The metabolic markers are H-Phe-Pro-OH, 2-Methylglutaric acid, Levothyroxine, 1-Methyl-L-Histidine, Uridine, D-Gln-OH, Deoxycholic acid, (R)-3-Hydroxy-tetradecanoic acid, and Cholic acid.

6. A computer-readable storage medium, which is used to store computer instructions, programs, code sets or instruction sets. When the computer-readable storage medium is run on a computer, the computer executes the method for screening and detecting tumor radiotherapy sensitivity markers as described in any one of claims 1 to 5.

7. An electronic device comprising one or more processors and a storage device, storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method for screening and detecting tumor radiotherapy sensitivity markers as described in any one of claims 1 to 5.

8. Obtaining a tumor radiotherapy sensitivity marker according to the method for screening and detecting tumor radiotherapy sensitivity markers according to any one of claims 1 to 5.

9. Use of the tumor radiotherapy sensitivity marker according to claim 8 in the preparation of a reagent for detecting tumor radiotherapy sensitivity.

10. The use according to claim 9, characterized in that: The critical value of the logistic regression model constructed according to tumor radiosensitivity markers predicts the radiosensitivity of colorectal cancer patients.

Images