Marker combination, model and method for predicting prognosis risk of colorectal cancer patient

Abstract

The invention belongs to the technical field of microbiome clinical diagnosis, and particularly relates to a marker combination, a model and a method for predicting the prognosis risk of a colorectal cancer patient. The method comprises the steps of whole genome sequencing data analysis, human source sequence filtering, microorganism sequence extraction and annotation and machine learning diagnosis model construction. The method has the advantages of being high in accuracy and free of culture, tissue-specific microbiota still exist in low-biomass tumors, and promising diagnostic value is shown. A large amount of clinical patient sequencing data can be subjected to secondary development and utilization, the accuracy of tumor diagnosis is improved, and the method can be widely applied to prognosis survival evaluation of patients.

Description

Technical Field

[0001] The present invention belongs to the technical field of microbiome clinical diagnosis, and specifically relates to a marker combination, model and method for predicting the prognosis risk of colorectal cancer patients. Background Art

[0002] Existing studies have found that microorganisms promote tumor occurrence, development and treatment through a variety of mechanisms, such as inducing DNA mutations, activating carcinogenic pathways, promoting chronic inflammation, promoting tumor progression through the complement system, and initiating tumor metastasis. Researchers have found that most tumors and their adjacent normal tissues contain different types of strains, mainly in cancer cells and immune cells. Unlike the intestinal and blood microbiota, the microbiota in tumors is very close to tumor cells and can more accurately reflect the actual state of the tumor. There are significant differences in the composition of the microbiota between tumors and healthy tissues, and between different tumor stages, genetic mutations, and distant tumor metastases, making it a potential prognostic tool and a new target for clinical intervention. Exploring combined treatment strategies based on microbial intervention can also improve clinical treatment outcomes.

[0003] Intestinal microbes are a large and diverse group, including bacteria, fungi, viruses and other types of microorganisms. They live in the human intestine, interact and influence each other with the human body, and together constitute the intestinal microecological system. Under normal circumstances, intestinal microorganisms are in a relatively balanced state and participate in many physiological functions of the human body, such as assisting in the digestion of food, synthesizing vitamins (such as vitamin K, some B vitamins, etc.), and regulating intestinal immune function. However, an imbalance in intestinal microorganisms can cause intestinal inflammation, produce carcinogens, affect host gene expression and cell signaling pathways, and then cause colorectal cancer. Therefore, it is an extremely effective strategy to understand the occurrence, development, treatment and prognosis of cancer by studying the intestinal microbiome.

[0004] The present invention uses whole genome sequencing data analysis, human sequence filtering, microbial sequence extraction and annotation, and machine learning diagnostic model construction to screen a marker combination that predicts the prognosis risk of colorectal cancer patients, which can predict the prognosis survival time of patients. The method described has the advantages of high accuracy and no need for cultivation. Even in low-biomass tumors, there are tissue-specific microbiota, showing promising diagnostic value. A large amount of clinical patient sequencing data can be secondary developed and utilized to improve the accuracy of tumor diagnosis, and can be widely used in patient prognosis survival assessment. Summary of the invention

[0005] In view of the above technical problems, the object of the present invention is to provide a marker combination for predicting the prognosis risk of patients with colorectal cancer, wherein the marker combination includes Clostridium cadaveris, Prevotella sp. oral.taxon.473, Eubacterium sp. 3_1_31, Filifactor alocis, Porphyromonas catoniae, Slackia sp. CM382, and Citrobacter sp. AATXR.

[0006] The second object of the present invention is to provide a reagent for detecting the marker combination, wherein the reagent is used to detect the abundance of the marker combination.

[0007] The third object of the present invention is to provide the use of the marker combination or the reagent in the preparation of a reagent, a kit or a model for predicting the survival and prognosis risk of colorectal cancer patients.

[0008] The fourth object of the present invention is to provide a kit for predicting the prognosis of colorectal cancer patients, wherein the kit comprises a reagent or an instrument for detecting the marker combination.

[0009] The fifth object of the present invention is to provide a method for screening markers for predicting the prognosis of patients with colorectal cancer, the method comprising:

[0010] (1) A total of 208 samples of colorectal cancer patients were collected. The patient survival time was used as the prediction index and the abundance of 1145 species was used as the prediction data. The samples were divided into a training set and a validation set in a ratio of 7:3.

[0011] (2) Prediction data screening: using R packages as machine learning algorithms, including machine learning gradient boosting machine, random forest, stepwise Cox, likelihood enhanced Cox, elastic network regression algorithm, Lasso regression, partial least squares regression of Cox model and support vector machine, and combining the above machine learning algorithms in pairs for prediction. After comparing the diagnostic AUC values ​​of different models, the model with AUC greater than 0.65 on both the training set and the validation set was selected. Finally, random forest combined with stepwise Cox was selected for prediction data screening, and the marker combination of claim 1 was screened out;

[0012] (3) External validation: The marker combination obtained in step (2) is validated on the validation set;

[0013] (4) Pearson correlation test was performed on the marker combination, and markers with p-value less than 0.05 were selected. The optimized marker combination contained 24 markers in total;

[0014] (5) Linear fitting was performed using the 24 microbial abundance and patient survival time data, and stepwise regression was performed using the R package MASS. The Akaike information criterion (AIC) was used as the screening criterion (the smaller the AIC, the better). There were 7 marker combinations with the smallest AIC values;

[0015] (6) The abundance of seven microbial markers and the patient's delivery time were used to fit the model, and the glm() function of R software was used for logistic regression calculation to calculate the prediction accuracy (AUC).

[0016] The sixth object of the present invention is to provide a model for predicting the prognosis of colorectal cancer patients established by the described method.

[0017] The seventh object of the present invention is to provide a model system for predicting the prognosis of colorectal cancer patients, comprising:

[0018] (1) a pre-input module, at least for inputting marker combination data;

[0019] (2) a prediction module, at least for predicting the marker combination data, which is performed by the model for predicting the prognosis of colorectal cancer patients;

[0020] (3) A display module, at least used to display the evaluation results.

[0021] An eighth object of the present invention is to provide a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the system implements the following steps, including: (1) inputting marker combination data for the prognosis of a colorectal cancer patient;

[0022] (2) Substituting the marker combination data into the model for calculation, the prognosis prediction conclusion of colorectal cancer patients was obtained;

[0023] (3) Output the predicted conclusion of the prognosis of patients with colorectal cancer.

[0024] A ninth object of the present invention is to provide a computer-readable storage medium on which the computer device is stored.

[0025] The beneficial effects of the present invention are:

[0026] ① The present invention provides a marker combination for predicting the prognosis risk of colorectal cancer patients using microorganisms, which is used to predict the prognosis survival time of patients, and the prediction accuracy is 0.7734;

[0027] ② The machine learning algorithm provided by the present invention has the advantages of high accuracy and no need for training;

[0028] ③ The prognostic markers provided by the present invention can guide clinical drug use, can conduct secondary development and utilization of a large amount of clinical patient sequencing data, improve the accuracy of tumor diagnosis, and can be widely used in the prognosis and survival assessment of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flow chart of the construction of microbial marker prognostic model;

[0030] Figure 2 Shannon diversity index of the intratumor microbiome;

[0031] Figure 3 Species composition of the tumor microbiome at the phylum level;

[0032] Figure 4 Sparcc clustering heat map of top 40 genera in the intratumor microbiome;

[0033] Figure 5 Phylogenetic relationships of the top 40 genera in the tumor microbiome;

[0034] Figure 6 Linear discriminant analysis Effect Size (LEfSe) analyzed the microbial differences between groups;

[0035] Figure 7 Use intratumoral microorganisms and multiple machine learning algorithms to build a prognostic survival model;

[0036] Figure 8 Based on the phylogenetic relationship and abundance display of 80 microbial markers;

[0037] Fig. 9 Diagnostic AUC plot based on 7 microbial markers. DETAILED DESCRIPTION

[0038] The following examples are provided to facilitate a better understanding of the present invention, but are not intended to limit the present invention.

[0039] The experimental methods in the following examples are conventional methods unless otherwise specified. The reagents in the following examples are conventional reagents and can be purchased from the market unless otherwise specified.

[0040] TCGA is a landmark cancer genomics project jointly conducted by the National Cancer Institute (NCI) and the National Human Genome Research Institute (NHGRI). It began in 2006 and has molecularly characterized more than 20,000 primary cancers and matched normal samples, covering 33 cancer types and generating more than 2.5 petabytes of genomic, epigenomic, transcriptomic and proteomic data. TCGA data can be accessed through the Genomic Data Commons (GDC) data portal, and some web-based analysis and visualization tools are also provided. The official website is https: / / www.cancer.gov / tcga / .

[0041] The TNM staging system is currently the most commonly used tumor staging system in the world. It was jointly developed and continuously updated by the International Union Against Cancer (UICC) and the American Joint Committee on Cancer (AJCC). It is mainly used to assess the severity of cancer, guide treatment options, and predict patient prognosis.

[0042] PathSeq is a computational analysis pipeline developed by the Broad Institute in the United States that detects the presence of microorganisms (such as bacteria, viruses, fungi, etc.) in host tissues and analyzes their composition based on high-throughput sequencing data. It is mainly used in many research fields such as tumors and immune-related diseases, helping researchers explore the relationship between microorganisms and the occurrence and development of diseases.

[0043] BAM file (Binary Alignment / Map format) is a binary file that is widely used in the field of bioinformatics. It is mainly used to store the result information of high-throughput sequencing data after sequence alignment. It is a very critical file type after processing the original sequencing data.

[0044] PathSeq is a computational pipeline developed by the Broad Institute in the United States for analyzing microbial sequences in high-throughput sequencing data. It aims to identify DNA or RNA sequences of microbial origin from the host genome background, helping researchers to deeply explore the relationship between microorganisms and hosts (such as humans, animals and plants) in health and disease states, especially playing an important role in many research fields such as tumors, infectious diseases, and immune-related diseases.

[0045] The flagstats function is a functional module in the SAMtools software package. It is mainly used to quickly count the alignment statistics in the alignment result files (usually SAM or BAM format files. BAM files are binary compressed forms of SAM files. Both store relevant information after the sequencing reads are aligned with the reference genome). It can help researchers quickly understand some key features and overall situation of the sequencing data after alignment to the reference genome.

[0046] SparCC (Sparse Correlations for Compositional data) correlation analysis is a method specifically used to analyze the correlation of compositional data, and is widely used in fields such as microbiome. Compositional data refers to the situation where the sum of each part of the data is a fixed value (usually 1 or 100%), such as the relative abundance data of each species in a microbial community, the sum of the relative abundance of all species is 100%. Traditional correlation analysis methods (such as Pearson correlation analysis, etc.) often have limitations when processing such compositional data, while SparCC can more effectively mine the correlation information therein.

[0047] RPM normalization aims to eliminate the impact of different sequencing depths on analysis results such as gene expression, so that data between different samples are more comparable. In high-throughput sequencing experiments, the amount of sequencing data obtained from different samples (that is, the total number of sequencing reads) is often inconsistent, and our focus is usually on the relative expression level of genes, rather than simply the number of reads. Through RPM normalization, the data of each sample can be unified to a relatively standard scale, which is convenient for subsequent accurate comparison of gene expression levels in different samples and analysis of differences in gene expression patterns.

[0048] Pearson correlation test, also known as Pearson product-moment correlation coefficient test, is an analytical method widely used in the field of statistics to measure the strength and direction of the linear relationship between two continuous variables.

[0049] Gradient Boosting Machine (GBM) is an ensemble learning algorithm based on the Boosting concept. Its core is to build multiple weak learners (usually decision trees) in sequence, and let each new weak learner focus on the error part of the prediction results of the previous round of learners, continuously iterate and improve, and gradually improve the prediction ability of the overall model. In each round of iteration, the negative gradient of the loss function with respect to the current model prediction value is calculated, which is used as the target value to be fitted by the new weak learner, and finally the prediction results of these weak learners are weighted and summed to obtain the final prediction result.

[0050] Random forest is a bagging algorithm in ensemble learning. It extracts multiple sub-datasets from the original training data set with replacement through bootstrap sampling, and then builds a decision tree for each sub-dataset, thus generating a "forest" of multiple decision trees. When making predictions (using a voting mechanism in classification problems and an averaging mechanism in regression problems), the prediction results of these decision trees are combined to obtain the final output. In addition, in the process of building each decision tree, the feature selection for each node is also random, which further increases the diversity and generalization ability of the model.

[0051] Stepwise Cox regression is a variable screening method developed from the traditional Cox proportional hazard model. The Cox proportional hazard model is used to analyze survival data (including information such as the time of event occurrence and whether the event occurred). It assumes that there is a certain proportional relationship between the hazard rate and the covariate. Stepwise Cox gradually introduces or removes covariates, based on certain statistical criteria (such as likelihood ratio test, AIC criterion, etc.) to screen out variables that have a significant impact on survival time, and construct a relatively optimal Cox model to better explain and predict survival.

[0052] Likelihood Boosting Cox (CoxBoost) combines the ideas of the Cox proportional hazard model and the Boosting algorithm. It is based on the Cox proportional hazard model. In each iteration, a weak learner (usually based on a certain function form or a simple model) is constructed through Boosting, focusing on fitting the residuals predicted by the current model (similar to the idea of ​​fitting negative gradients in the gradient boosting machine), and then continuously updating the model to enhance the ability to fit survival data, and finally obtain a comprehensive model that can better handle survival data.

[0053] Elastic Net Regression (Enet, Elastic Net) is a regularization method for linear regression, which combines the characteristics of L1 regularization (the main method in Lasso regression) and L2 regularization (similar to the method in ridge regression). By adding a penalty term consisting of L1 and L2 regularization terms to the loss function, the coefficients of the model are constrained to prevent overfitting, while achieving the functions of variable selection and coefficient shrinkage. Its goal is to meet the regularization conditions and find a set of optimal regression coefficients while minimizing the loss function (such as the mean square error loss function).

[0054] Lasso regression is short for Least Absolute Shrinkage and Selection Operator regression, which is a regularization method for linear regression. Its core is to add an L1 regularization term to the loss function of traditional linear regression (such as the mean square error loss function), that is, to penalize the sum of the absolute values ​​of the regression coefficients. In the process of minimizing the loss function, this regularization term will cause some unimportant regression coefficients to shrink to 0, thereby realizing the function of automatic variable selection and preventing the model from overfitting.

[0055] Partial least squares regression (plsRcox) of the Cox model is a technique that applies the partial least squares regression (PLS) method to the Cox proportional hazard model. Partial least squares regression is a multivariate statistical analysis method that can handle complex situations such as multicollinearity between independent variables and the number of independent variables being greater than the number of samples. It constructs a regression model by extracting the principal components (comprehensive variables) of the independent variables and making these principal components have the largest covariance with the dependent variable (in survival analysis, related information such as survival time). When applied to the Cox model, the principal components extracted by PLS are used to better explain and predict survival time. At the same time, taking into account the characteristics of the Cox model such as the assumption of proportional hazards, a model that better meets the requirements of survival data analysis is constructed.

[0056] Support vector machines were originally machine learning models for binary classification problems. The core idea is to find an optimal hyperplane (a straight line in two-dimensional space, and a similar dividing plane in high-dimensional space) to separate data points of different categories as much as possible, and to maximize the distance from this hyperplane to the closest points in the two categories of data (these points are called support vectors). This distance is called margin. For linearly separable data, such a hyperplane can be found directly; for nonlinearly separable data, the original data is mapped to a high-dimensional space through a kernel function (such as the commonly used Gaussian kernel function, polynomial kernel function, etc.), so that it becomes linearly separable in the high-dimensional space, and then the optimal hyperplane is found for classification. In regression problems (support vector regression, SVR), the goal is to find a hyperplane so that most data points fall within a certain range around this hyperplane (this range is controlled by setting a parameter) to minimize prediction errors.

[0057] Example 1: Construction of a prognostic survival model based on microbial markers

[0058] This study analyzed the whole genome data of colorectal cancer tissues in the TCGA public database, excluded human sequences and contamination, performed microbial comparison and annotation on the remaining sequences, obtained the corresponding species abundance table, analyzed the diversity and community composition of intratumoral microorganisms, and compared them with adjacent tissues to find differential markers. Taking patient survival as the prediction object, different machine learning combination algorithms were constructed to obtain the best diagnostic performance.

[0059] like Figure 1 Shown is the design and validation route of the entire microbial prognostic model.

[0060] 1. TCGA whole genome sequencing data analysis and microbial annotation

[0061] (1) Download the whole genome sequencing data of patients from the TCGA website, including primary tumors and solid normal tissues, as well as clinical data (TNM stage, gender, survival, etc.).

[0062] (2) The PathSeq pipeline was used to screen the microbial content in all WGS data of TCGA samples. For TCGA input bam files and PathSeq output bam files, RPM normalization calculations were performed using the flagstats function of SAMtools. The bacterial abundance values ​​were then normalized to the total read counts (in millions) of the input bam file. The sequence denoising method refers to Dohlman AB, et al. 2021, 10; 29(2): 281-298.

[0063] (3) Using R analysis software, Shannon diversity was calculated and species composition at the phylum level was analyzed. Clustering between cancer and normal tissues was analyzed based on the TOP40 genera.

[0064] The results showed that the microbial diversity in colorectal cancer was higher than that in normal tissues ( Figure 1 ), Bacteroides and Firmicutes were the main phyla in normal tissues, while the microorganisms in tumors were mainly Bacteroides, Proteobacteria and Fusobacteria ( Figure 2 ). Through SparCC correlation analysis, the heat map showed that the intratumoral microorganisms formed a cluster centered on Fusobacterium, while the healthy tissue formed a cluster centered on Bacteroides, indicating that the intratumoral microorganisms were significantly different from normal tissues ( Figure 3 The dendrogram further shows that normal tissues contain beneficial bacteria such as Alistipes and Butyricimonas from the Bacteroides cluster, while the Fusobacterium cluster contains harmful bacteria such as Fusobacterium and Leptotrichia ( Figure 4 The above species differences were further confirmed by LDAEFfect Size analysis ( Figure 5 and Figure 6 ).

[0065] 2. Use machine learning algorithms to build a prognostic survival model

[0066] (1) A total of 208 colorectal cancer patient samples were collected. The patient survival time was used as the prediction indicator, and the abundance of 1145 species was used as the prediction data. The samples were divided into a training set and a validation set in a ratio of 7:3 to select the best marker combination.

[0067] (2) Use R packages (randomForestSRC, glmnet, plsRcox, superpc, gbm, CoxBoost, survivalsvm, survival) to perform predictions using machine learning algorithms, including: machine learning gradient boosting machine (GBM), random forest (RSF), stepwise Cox (StepCox), likelihood boosted Cox (CoxBoost), elastic network regression algorithm (Enet), Lasso regression (Lasso), partial least squares regression of Cox model (plsRcox), support vector machine (SVM), and their pairwise combinations. Finally, the machine forest combined with stepCOX model was selected. The marker combination was screened on the training set and verified on the validation set; a total of 80 marker combinations were screened, including Actinomyces sp. HMSC035G02, Actinomyces sp. ph3, Aggregatibacter.segnis, Agrobacterium sp. SUL3, Akkermansia muciniphila, Anaerotignum propionicum, Asaccharospora irregularis, Bacteroides clarus, Bacteroides dorei, Bacteroides massiliensis, Bacteroides nordii, Bacteroides spectinophilus, Bacteroides salieri, and Bacteroides clarus. salyersiae, Bacteroides sp. 14A, Bacteroides timonensis, Bittarella massiliensis, Butyricimonas synergistica, Citrobacter freundii., Citrobacter rodentium, Citrobacter sp. MGH103, Citrobacter sp. MGH104, Clostridiales bacterium VE202.15, Clostridioides difficile, Clostridium paraputrificum, Clostridium saudiense, Clostridium sp. KLE.1755, Comamonas testosteroni, Dialister pneumosintes, Eisenbergiella tayi, Enterobacter sp. K66-74, Enterococcus sp. 255_ESPC, Enterobacter sp. TW09276, Enterobacter sp. TW09308, Faecalibaculum rodentium, Fournierella massiliensis, Fusobacterium hwasookii, Fusobacterium sp. CM22, Haemophilus haemolyticus, Haemophilus sp. CCUG.66565, Holdemanella biformis, Intestinimonas butyriciproducens, Lachnospirapectinoschiza, Lachnospiraceae bacterium.)1_4_56FAA, Levyella massiliensis, Massilioclostridium coli, Negativibacillus massiliensis, Neglecta timonensis, Neofamilia massiliensis, Oribacterium sinus, Oscillibacter valericigenes, Parabacteroides johnsonii, Peptoniphilus harei, Peptoniphilus senegalensis, Porphyromonas endodontalis, Porphyromonas levii, Porphyromonas gingivalis sp.) HMSC065F10, Prevotella baroniae, Prevotella buccae, Prevotella falsenii, Prevotella multisaccharivorax, Prevotella pleuritidis, Prevotella shahii, Prevotella sp. P4.76, Pyramidobacter piscolens, Pyramidobacter sp. C12.8, Roseburia hominis, Selenomonas artemidis, Selenomonas infelix, Selenomonas sputigena, Senegalimassilia anaerobia), Shigella sp. FC2045, Shigella sp. FC2928, Solobacterium moorei, Streptococcus mitis, Streptococcus sp. CCUG.49591, Streptococcus sp.)HMSC065C01, Streptococcus sp. HMSC074F05, Streptococcus sp. HMSC076C09, Subdoligranulum variabile and Tepidiphilus thermophilus.

[0068] (3) Using the patients’ overall survival time and marker data, we performed a Pearson correlation test and selected markers with a p-value less than 0.05. The optimized markers included a total of 24 microorganisms.

[0069] (4) The 24 markers and the patients' overall survival time were linearly fitted, and stepwise regression screening was performed using the R package MASS. The Akaike information criterion (AIC) was used as the screening criterion (the smaller the value, the better). There were 7 marker combinations with the smallest AIC values: Clostridium cadaveris, Prevotella sp. oral.taxon.473, Eubacterium sp. 3_1_31, Filifactor alocis, Porphyromonas catoniae, Slackia sp. CM382, and Citrobacter sp. AATXR.

[0070] (5) The glm() function of R software was used to perform logistic regression fitting, and the abundance of the seven markers and the patient's overall survival time were fitted into a logistic regression model to calculate the accuracy of predicting the patient's survival time (AUC value).

[0071] The results show that the random forest combined with the stepCOX model has the best prediction effect, with an AUC value of 0.830 in the training set and an AUC value of 0.673 in the training set. Figure 7 ). 80 major markers, of which harmful bacteria accounted for a large proportion, and some bacteria came from oral microorganisms ( Figure 8 ).

[0072] The accuracy of survival time prediction using the 7 microbial markers finally screened was 0.7734 ( Fig. 9 ).

[0073] Example 2: A model system for predicting the prognosis of colorectal cancer patients

[0074] include:

[0075] (1) a pre-input module, at least for inputting marker combination data;

[0076] (2) a prediction module, at least for predicting the marker combination data, which is performed by the model for predicting the prognosis of colorectal cancer patients;

[0077] (3) A display module, at least used to display the evaluation results.

[0078] Embodiment 3: A computer device

[0079] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the system implements the following steps, including: (1) inputting marker combination data for the prognosis of a colorectal cancer patient;

[0080] (2) Substituting the marker combination data into the model for calculation, the prognosis prediction conclusion of colorectal cancer patients was obtained;

[0081] (3) Output the predicted conclusion of the prognosis of patients with colorectal cancer.

[0082] In summary, the present invention provides a marker combination for predicting the prognostic risk of colorectal cancer patients using microorganisms for predicting the prognosis and survival of patients; it has the advantages of high accuracy and no need for cultivation, and even in low-biomass tumors, there are tissue-specific microbiota, showing promising diagnostic value. It can conduct secondary development and utilization of a large amount of clinical patient sequencing data, improve the accuracy of tumor diagnosis, and can be widely used in the prognosis and survival assessment of patients.

[0083] The above description is a detailed description of the preferred feasible embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modified changes completed under the technical spirit suggested by the present invention should fall within the patent scope covered by the present invention.

Claims

1. A marker combination for predicting the prognosis risk of colorectal cancer patients, characterized in that: The marker combination includes Clostridium cadaveris, Prevotella sp. oral.taxon.473, Eubacterium sp. 3_1_31, Filifactor alocis, Porphyromonas catoniae, Slackia sp. CM382, and Citrobacter sp. AATXR.

2. A reagent for detecting the marker combination as claimed in claim 1, characterized in that: The reagent is used to detect the abundance of the marker combination.

3. Use of the marker combination according to claim 1 or the reagent according to claim 2 in the preparation of a reagent, a kit or a model for predicting the survival and prognostic risk of colorectal cancer patients.

4. A kit for predicting the prognosis of colorectal cancer patients, characterized in that: The kit comprises reagents or instruments for detecting the marker combination according to claim 1.

5. A method for screening markers for predicting the prognosis of patients with colorectal cancer, characterized in that: The method includes: (1) A total of 208 samples of colorectal cancer patients were collected. The patient survival time was used as the prediction index and the abundance of 1145 species was used as the prediction data. The samples were divided into a training set and a validation set in a ratio of 7:

3. (2) Prediction data screening: using R packages as machine learning algorithms, including machine learning gradient boosting machine, random forest, stepwise Cox, likelihood enhanced Cox, elastic network regression algorithm, Lasso regression, partial least squares regression of Cox model and support vector machine, and combining the above machine learning algorithms in pairs for prediction. After comparing the diagnostic AUC values ​​of different models, the model with AUC greater than 0.65 on both the training set and the validation set was selected. Finally, random forest combined with stepwise Cox was selected for prediction data screening, and the marker combination of claim 1 was screened out; (3) External validation: The marker combination obtained in step (2) is validated on the validation set; (4) Pearson correlation test was performed on the marker combination, and markers with p-value less than 0.05 were selected. The optimized marker combination contained 24 markers in total; (5) Linear fitting was performed using the 24 microbial abundance and patient survival time data, and stepwise regression was performed using the R package MASS. The Akaike information criterion (AIC) was used as the screening criterion (the smaller the AIC, the better). There were 7 marker combinations with the smallest AIC values; (6) The abundance of seven microbial markers and the patient's delivery time were used to fit the model, and the glm() function of R software was used for logistic regression calculation to calculate the prediction accuracy (AUC).

6. A model for predicting the prognosis of colorectal cancer patients established by the method described in claim 5.

7. A model system for predicting the prognosis of colorectal cancer patients, characterized in that: include: (1) a pre-input module, at least for inputting marker combination data; (2) a prediction module, at least for predicting the marker combination data, which is executed by the model for predicting the prognosis of colorectal cancer patients as described in claim 7; (3) A display module, at least used to display the evaluation results.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the system of claim 8 implements the following steps, including: (1) inputting marker combination data for the prognosis of colorectal cancer patients; (2) Substituting the marker combination data into the model for calculation, the prognosis prediction conclusion of colorectal cancer patients was obtained; (3) Output the predicted conclusion of the prognosis of patients with colorectal cancer.

9. A computer-readable storage medium having stored thereon the computer device according to claim 8.

Images