Screening marker combination, screening prediction system and method for pancreatic ductal adenocarcinoma

Abstract

The invention relates to a screening marker combination for pancreatic ductal adenocarcinoma, a screening prediction system for pancreatic ductal adenocarcinoma and a screening prediction method for pancreatic ductal adenocarcinoma. The screening and predicting method comprises the following steps: detecting the expression levels of the characteristic genes TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G in a patient sample; inputting a detection result into the screening prediction system for pancreatic ductal adenocarcinoma provided by the invention to obtain a prediction value; when the predicted value reaches a preset value, the system outputs' PDAC 'which represents the risk of pancreatic duct adenocarcinoma; and when the predicted value reaches a preset value, outputting'one 'to indicate that the risk of pancreatic duct adenocarcinoma does not exist. According to the screening and predicting method for pancreatic ductal adenocarcinoma provided by the invention, the credibility and the accuracy of a detection result are improved, and prevention and early intervention of diseases can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of prognostic models, and in particular to a screening marker combination, a screening prediction system and a method for pancreatic ductal adenocarcinoma. Background Art

[0002] Pancreatic cancer (PC) is the most malignant digestive system tumor, and its mortality rate ranks sixth among cancers worldwide. It is expected to soar to second only to lung cancer in 2030. Among them, pancreatic ductal adenocarcinoma (PDAC) is the main histological subtype of PC, accounting for 90% of all cases. The overall 5-year survival rate of pancreatic ductal adenocarcinoma is about 10%, and the early diagnosis rate of pancreatic ductal adenocarcinoma is only 5%, but the 5-year survival rate after resection can be as high as 60%~80%. The main reason is that pancreatic ductal adenocarcinoma has an insidious onset, atypical early symptoms and is easy to be ignored. When more typical symptoms appear, it is usually in the middle and late stages. Existing screening technologies such as tumor markers, B-ultrasound, CT, MRI, etc. have low sensitivity and specificity for pancreatic ductal adenocarcinoma. Therefore, only by achieving early screening of such patients can the goal of improving the survival rate of pancreatic ductal adenocarcinoma be achieved.

[0003] Cell-free RNA (cfRNA) testing is a non-invasive test obtained through simple body fluid sampling. These cfRNA fragments are released from cells into body fluids such as blood, urine, and saliva, which can provide dynamic gene expression information reflecting real-time physiological and pathological changes. Finding biomarkers for pancreatic ductal adenocarcinoma through liquid biopsy technology has also become an emerging direction.

[0004] Among the existing technologies, the trend of using artificial intelligence in the field of life sciences is increasing year by year, especially machine learning methods are increasingly being used in cfRNA-based cancer prediction research. For example, Martin (Martin J Keiser et al., Development of a novel blood RNA gene expression platform based on early lung cancer diagnosis) successfully predicted the early onset of lung cancer through cfRNA data analysis using machine learning technology. Ning et al. (A comprehensive evaluation of full-spectrum cell-free RNA highlights the use of cell-free free RNA fragments for RNA detection in early hepatocellular carcinoma, Electronic Biomedicine, 2023) identified cfRNA biomarkers for early liver cancer and used a random forest machine learning model to accurately distinguish liver cancer patients from controls, with an overall AUC of more than 0.9. Similarly, in the classification of pancreatic ductal adenocarcinoma (PDAC), Yu et al. (Plasma extracellular vesicle long RNA profiles determine the diagnostic characteristics of pancreatic ductal adenocarcinoma, 2020) studied the extraction of long non-coding RNA (lncRNA) molecules from extracellular vesicles (EVs) and used EV-RNA profiles to identify PDAC with an AUC of 0.93. The above studies show that machine learning can be used to analyze large amounts of complex cfRNA data, extract key features, and build high-precision disease prediction models, especially early cancer screening models.

[0005] Therefore, solutions on how to further apply machine learning methods to cfRNA-based cancer prediction are needed. Summary of the invention

[0006] The present invention provides a screening marker combination, a screening prediction system and a method for pancreatic ductal adenocarcinoma to solve one or more of the above-mentioned problems and other problems, or to provide an alternative technical solution to the prior art.

[0007] To achieve the above objectives, in a first aspect, the present invention provides a screening marker combination for pancreatic ductal adenocarcinoma, wherein the marker is a combination of the following characteristic genes: TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G.

[0008] In one implementation, each characteristic gene of the marker combination is obtained based on the cfRNA algorithm ReguGLM of GLM.

[0009] In a second aspect, the present invention further provides a screening and prediction system for pancreatic ductal adenocarcinoma, comprising: an acquisition module, used to obtain the expression level of a screening marker for pancreatic ductal adenocarcinoma; A prediction module, comprising a module for obtaining a prediction value according to the expression level obtained by the acquisition module, and outputting a disease risk assessment result according to the prediction value; Wherein, the prediction module includes a pancreatic ductal adenocarcinoma probability model, and the pancreatic ductal adenocarcinoma probability model includes a mathematical formula: (K=3,k=1,2,3), (1), in (k=1, 2, 3), (2), Among them, in formula (2), p is the number of characteristic genes, j represents the jth characteristic gene; Substituting specific values ​​into the above formula (2) yields: (3) In formula (1), P represents the probability that the screening result is of category k. When k = 1, P represents the probability that the result is a healthy control. When k = 2, P represents the probability that the result is a benign control. When k = 3, P represents the probability that the result is pancreatic ductal adenocarcinoma. β0 represents the probability that the kth category is z k The intercept, β 1-6 represents the weight of the expression value of each gene in the plasma of the sample in the kth category, X 1-6 It indicates the specific expression value of each gene in the plasma of the sample, and the order of each gene is TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G; Wherein, the acquisition module and the prediction module are connected wirelessly and / or wiredly.

[0010] In one implementation, the marker is the screening marker combination for pancreatic ductal adenocarcinoma described in the present invention.

[0011] In a third aspect, the present invention further provides a method for training a pancreatic ductal adenocarcinoma probability model for use in the screening prediction system of the present invention, comprising the following steps: S11. Sample data collection: plasma samples from multiple pancreatic ductal adenocarcinoma patients, healthy controls and benign controls were collected, and the expression levels of the characteristic genes TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G in the samples were determined; S12. Data labeling: label the samples into three categories: healthy control, benign control and pancreatic ductal adenocarcinoma, where the healthy control is the first category (k=1), the benign control is the second category (k=2), and the pancreatic ductal adenocarcinoma is the third category (k=3); S13, feature engineering: standardizing the expression levels of the feature genes to obtain standardized expression values ​​of the 7 feature genes in each sample; S14. Model construction: A probability model is established based on the generalized linear model, and the following mathematical formula is set: Construct the category probability formula: (k=1,2,3), (1), Where: In the above formula (1), k represents the type, which is divided into 1, 2, and 3 categories, P represents the probability of type k when x is input, K represents the number of categories, which is 3 here; z k is the linear prediction value of the kth class, expressed as: (j=1,2,...,p), (2), In the above formula (2), β is the regression coefficient to be estimated, k = 1, 2, 3, indicating that there are three categories, p is the number of characteristic genes, j represents the jth characteristic gene, and β k0 is the intercept of category k, β kj is the weight of the jth characteristic gene in category k, x j is the expression value of the jth characteristic gene; The gradient of the loss function is calculated as follows: (4) In the above formula (4), k represents the kth category, n is the number of samples, and β is approximated by using the quasi-Newton method to find its numerical solution; S15, parameter training: Through the maximum likelihood estimation method, based on the labeled sample data, the model parameters β are trained k0 and β kj ; S16. Model validation: The classification performance of the model was evaluated using the cross-validation method, and the accuracy, sensitivity, and specificity of the model were calculated; S17, model optimization: adjust the model parameters according to the verification results until the classification accuracy requirements of the screening prediction system are met; S18. Model output: deploying the finally optimized pancreatic ductal adenocarcinoma probability model into the screening prediction system as a prediction module of the system described in claim 3.

[0012] In a fourth aspect, the present invention further provides a method for screening and predicting pancreatic ductal adenocarcinoma, comprising the following steps: S21, detection of the expression levels of TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2, and FCER1G in patient samples; S22, inputting the test result into the above-mentioned screening prediction system for pancreatic ductal adenocarcinoma to obtain a prediction value; S23. When the predicted value reaches a preset value, the system outputs "PDAC" to indicate that there is a risk of pancreatic ductal adenocarcinoma; when the predicted value reaches a preset value, the system outputs "none" to indicate that there is no risk of pancreatic ductal adenocarcinoma.

[0013] In a fifth aspect, the present invention further provides a method for obtaining the expression level of a screening marker combination for pancreatic ductal adenocarcinoma, characterized in that it comprises the following steps: S31. Sample collection: Collect plasma samples from patients; S32. Nucleic acid extraction: extracting circulating free RNA (cfRNA) from the plasma sample using a nucleic acid extraction kit; S33, reverse transcription: converting the extracted cfRNA into cDNA through reverse transcription reaction; S34, quantitative PCR: Real-time quantitative PCR technology was used to amplify and detect the expression levels of target genes in cDNA using specific primers and probes, including the following characteristic genes: TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G; S35. Data standardization: Standardize the results of quantitative PCR to obtain the relative expression level of each characteristic gene; S36. Result output: The standardized gene expression level data is output as input data of the screening prediction system for further analysis and prediction of the risk of pancreatic ductal adenocarcinoma.

[0014] In one implementation, the marker is the screening marker combination for pancreatic ductal adenocarcinoma described in the present invention.

[0015] Beneficial effects: The present invention provides a combination of seven new markers for pancreatic ductal adenocarcinoma in human plasma, as well as the weight of each new marker, and constructs an early auxiliary screening model based on machine learning that can be used to predict the probability of pancreatic ductal adenocarcinoma. The model can reduce the inaccuracy of traditional simple bioinformatics analysis plus human subjective judgment verification, reduce the interference of human factors to a certain extent, improve the credibility and accuracy of the test results, and achieve disease prevention and early intervention. The use of the model can be better and more conveniently applied to clinical practice; at the same time, the provided screening marker combination for pancreatic ductal adenocarcinoma can integrate the information provided by different markers, overcome the limitations of a single marker, and thus improve the sensitivity and specificity of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of the steps of a screening prediction method for pancreatic ductal adenocarcinoma; Figure 2It is a flow chart of the pancreatic ductal adenocarcinoma probability model training provided by the present invention; Figure 3 is a sample accuracy result diagram of the pancreatic ductal adenocarcinoma probability model provided by the present invention; Figure 4 is a confusion matrix diagram of the pancreatic ductal adenocarcinoma probability model provided by the present invention; Figure 5 It is a statistical graph of sample expression levels of the pancreatic ductal adenocarcinoma probability model provided by the present invention; Figure 6 is a sample accuracy result diagram of the model of comparative example 1 when different key factors are used in the present invention; Figure 7 It is a sample accuracy result diagram of the pancreatic ductal adenocarcinoma probability model of comparative example 2 when different key factors are used in the present invention.

[0017] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the description of the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" described below means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms is not necessarily for the same embodiment or example. Moreover, the technical features involved in each embodiment of the present invention can be combined with each other as long as they do not conflict with each other.

[0019] Example 1: Methods for screening markers, acquiring data and training models 1. Sample collection: Inclusion criteria: Patients with clinically confirmed pancreatic ductal adenocarcinoma (early stage pancreatic ductal adenocarcinoma is preferred. According to the guidelines, early stage pancreatic ductal adenocarcinoma is defined as tumor diameter ≤ 2 cm, lesions confined to the pancreas, without extrapancreatic infiltration and lymph node metastasis); benign control group (referring to patients with non-malignant diseases, such as benign tumors or other non-malignant diseases); healthy control group.

[0020] The age of the healthy group and the diseased group was relatively close, such as 65±20 years old.

[0021] The sample was randomly selected, with males and females accounting for approximately 50% each.

[0022] The diagnostic basis may include the following three: imaging examination, serum marker examination represented by carbohydrate antigen (CA) 19-9, histopathological examination, etc.

[0023] Exclusion criteria: patients with other serious diseases or complications; patients unable to comply with study requirements or compliance.

[0024] The PDAC group excluded patients with clear imaging evidence showing a history of tumor metastasis to the brain or other organs; the benign control group and the healthy control group excluded patients with a history of malignant tumors (regardless of whether they were cured or in remission).

[0025] Finally, a total of 299 subjects were included in the study. The proportion of each group is shown in Table 1.

[0026] Table 1. Sample plasma sample information table 2. RNA extraction: 1. Sample preparation: Blood samples were centrifuged to separate plasma and transferred to new storage tubes.

[0027] 2.cfRNA extraction: cfRNA extraction was processed using the Tiangen Virus Genomic DNA / RNA Extraction Kit (DP315). Prepare 200ul of plasma and add 200ul of Carrier RNA working solution.

[0028] 3. Add 250ul of anhydrous ethanol, mix well, and leave at room temperature for 5 min.

[0029] 4. Transfer the solution and flocculent precipitate in the centrifuge tube to adsorption column CR2, centrifuge and discard the waste liquid.

[0030] 5. Add 500ul of buffer GD, centrifuge at 8000rpm for 1min, wash and collect RNA. Add 100ul of DNase I treatment solution, including 5ul of DNase I (full gold, GD201-01), 10ul of 10×DNase I Reaction Buffer and 85ul of enzyme-free water; incubate at 25-30℃ for 15 minutes.

[0031] 6. Add 600ul of rinse solution PW, let stand for 2min, centrifuge at 8000rpm for 1min, discard the waste solution. Repeat once.

[0032] 7. Add 500ul of anhydrous ethanol, centrifuge at 8000rpm for 1min, and discard the waste liquid.

[0033] 8. Place the adsorption column back into the collection tube and centrifuge at 12,000 rpm for 3 minutes to completely dry the adsorption film and discard the waste liquid.

[0034] 9. Add 30ul of enzyme-free water to the adsorption column, let it stand at room temperature for 5 minutes, and centrifuge at 12000rpm for 1 minute.

[0035] 3. Library preparation: Library preparation was performed using the Yisheng library preparation kit (Cat. No. 12308) 5.3.1 Fragmentation and removal of ribosomal rRNA: Take 8 ul of RNA, 8.5 ul of 2×Frag / Prime Buffer, 0.5 ul of One-Step rRNA Removal Mix, and make up to 17 ul with water.

[0036] After mixing, run the following reaction program: 94°C, 7 min; 75°C, 1 min; 50°C, 2 min; 5.3.2 First-strand cDNA synthesis: Take 17ul of the above reaction, add 6ul of Strand Specificity Reagent and 2ul of 1st Strand EnzymeMix, and run the following reaction program: 25℃, 10min; 42℃, 15min; 70℃, 15min.

[0037] 5.3.3 Second-strand cDNA synthesis, end repair and dA addition 1. Take the above reaction system, add 2nd Strand Buffer (dNTP) 30ul and 2nd Strand Enzyme Master Mix 5ul; run the following reaction program: 16℃, 30min; 72℃, 15min.

[0038] 2. End repair connector connection To the above reaction system (60ul), add 30ul of Ligation enhancer, 5ul of T4 DNA ligase, and 5ul of MGIDNA Adaptor (0.5uM) at 20°C for 15min. Then, AMPure XP magnetic beads (Beckman, A63880) were used to purify the ligation product.

[0039] 5.3.4 Library Amplification Take 20ul of the above ligation product, add 25ul of 2×Super Canace® II High-Fidelity Mix and 5ul of MGI Primer mix (Yisheng, 13350) to carry out PCR reaction: 105℃, hot cover; 98℃, 1min; (98℃, 10sec; 60℃, 30sec; 72℃, 30sec), 15 cycles; 72℃, 5min; 4℃, Hold.

[0040] The ligation product was subsequently purified using AMPure XP magnetic beads (Beckman, A63880).

[0041] 5.3.5 Library quality control The library concentration was tested by the Qubit 4.0 fluorescence quantification instrument, and the samples that passed the quality inspection were sequenced.

[0042] 4. High-throughput sequencing: Inject the sample into the sample well of the kit (kit combined with probe-anchored polymerase sequencing method G99 SM FCL PE150) 5. Sequencing data preprocessing: 5.1 Sequencing data quality control: Fastp v0.23.3 was used for data quality control and to clean up low-quality data. The quality of the original sequencing data was assessed and low-quality sequences were removed. Then, the adapter sequences were detected and removed to finally generate cleaned high-quality sequencing data.

[0043] 5.2 Sequencing data comparison: The sequencing data were aligned with the human genome data using STAR v2.7.10b.

[0044] First, a genome index is constructed using the human genome reference sequence and annotation files.

[0045] Secondly, the cleaned sequencing data (FASTQ file) was aligned with the constructed genome index to generate an alignment file (BAM format).

[0046] 5.3 Gene Counting: Gene counting was performed using feature Counts v 2.0.6. First, the BAM file generated by the alignment was aligned with the gene annotation file (GTF format). Second, the number of sequencing reads for each gene was counted to generate a gene count matrix.

[0047] 5.4 Normalization: In order to enhance counting stability and reduce the deviation caused by library size, each gene was normalized using the following method: in, is the normalized gene expression, xi is the original expression, λ is the amplification constant to prevent the value from being too small, and n is the number of genes in the corresponding sample.

[0048] 6. Artificial Intelligence Analysis The present invention provides a cfRNA algorithm ReguGLM based on GLM, and the algorithm process is as follows: for a three-classification problem, a feature matrix X and a response variable Y are set, where Y can take the value of 0, 1 or 2.

[0049] Use the ReguGLM model to describe the relationship between the response variable and the features: (k=1,2,3), (1), Where k represents the type, which is divided into 1, 2, and 3 categories, P represents the probability of type k when x is input, and K represents the number of categories, which is 3 here. k is the linear prediction value of the kth class, expressed as: (2), In formula (2), β is the regression coefficient to be estimated, k = 1, 2, 3, indicating that there are three categories, p is the number of characteristic genes, and j represents the jth characteristic gene; Using the log-likelihood function as the loss function, its negative log-likelihood function is: (5) In formula (5), y im Represents the true value (label) of the mth category of the ith sample.

[0050] In order to solve the problems of multicollinearity and overfitting, the regularization term is introduced as follows: (6) In equations (5) and (6), λ is the regularization parameter that controls the regularization strength, n is the number of samples, m represents the mth category, K represents the number of categories, j is the number of characteristic genes, and βmj represents the regression coefficient of the jth gene in the mth category.

[0051] The present invention needs to minimize the loss function Loss. By calculating the gradient of the loss function with respect to β and setting it to zero, the optimal solution can be obtained. The gradient of the loss function is calculated as follows: (4) In formula (4), k represents the kth category and n is the number of samples.

[0052] Usually, it is impossible to directly obtain an analytical solution because the loss function is not quadratically differentiable. The present invention uses the quasi-Newton method to find its numerical solution for approximation.

[0053] Regularization reduces the variance of the regression coefficient and improves the robustness of the model. Finally, the coefficient β is obtained, which can be used to predict new samples.

[0054] In this process, the present invention divides the samples into corresponding training sets (64%), validation sets (16%), and test sets (20%). The model input is the key characteristic gene of cfRNA, and a five-fold cross-validation is performed to predict the probability of suffering from pancreatic ductal adenocarcinoma. Finally, the corresponding weight of each characteristic gene in the prediction of pancreatic ductal adenocarcinoma disease and the formula for predicting the probability of pancreatic ductal adenocarcinoma are obtained. The training model is shown in the figure below: Figure 2 shown.

[0055] Example 2: Reliability verification of the probability model of pancreatic ductal adenocarcinoma 6.1 Using labeled sample gene expression data: The expression data of the seven key characteristic genes in the two groups were used as input features to more specifically predict the risk of pancreatic ductal adenocarcinoma. This supervised learning may improve the prediction accuracy of the model.

[0056] 6.2 Training set: The training set data is used to train the model. The model learns the relationship between sample features and labels in the training set and adjusts its own parameters so that it can accurately predict new samples. The training set of this model accounts for 80% of the total data set.

[0057] 6.3 Validation set: During the training process, the validation set is used to adjust the model's hyperparameters (such as regularization strength, etc.). The model calculates the optimal hyperparameters based on the validation set data. The model sets the optimal parameters for itself and then retrains. After adjusting the parameters, retraining is performed to effectively avoid overfitting or underfitting, ensuring that the model performs better in practical applications.

[0058] 6.4 Test Set: The test set is not involved in the training and validation process and is only used in the final evaluation of model performance, ensuring the fairness and reliability of the evaluation results. The training set of this model accounts for 20% of the total data set.

[0059] 6.5 Prediction of pancreatic ductal adenocarcinoma Prediction principle: These 7 key factors are the union genes of the three groups of differentially compared genes analyzed by bioinformatics. The gene expression data corresponding to the phenotypes of the 7 key factors are used as the input for model training. After the model is trained, the output results include prediction probability and prediction performance. The prediction probability is used as an evaluation indicator to predict the probability of pancreatic ductal adenocarcinoma. The higher the prediction value, the higher the probability of the patient suffering from pancreatic ductal adenocarcinoma. The prediction performance is the accuracy of the prediction, that is, the area under the receiver operating characteristic curve (ROC curve), also known as AUC. The higher the AUC, the better the model prediction performance.

[0060] Prediction results: Combined with reference Figure 3~Figure 5 , Figure 3 is a sample accuracy result diagram of the pancreatic ductal adenocarcinoma probability model provided by the present invention; Figure 4 is a confusion matrix diagram of the pancreatic ductal adenocarcinoma probability model provided by the present invention; Figure 5 It is a statistical graph of sample expression levels of the pancreatic ductal adenocarcinoma probability model provided by the present invention. Figure 4 In the confusion matrix, the horizontal axis represents the predicted value, the vertical axis represents the actual value, and the square represents the percentage of the predicted value to the actual value in each category. The darker the color, the higher the proportion, and vice versa.

[0061] from Figure 3~Figure 5 The results show that the overall AUC reaches 80%, revealing the high accuracy and effectiveness of the model in making decisions for given data; the model prediction results include 7 key factors and weights: TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2, FCER1G, as shown in Table 2; the sequences of the 7 key factors are shown in the attached nucleotide sequence table. It can be seen that the model obtained by using the key factors and parameter adjustment method according to the present invention has a better effect than the corresponding model using other key factors.

[0062] Table 2. Prediction results of the probability model for pancreatic ductal adenocarcinoma The sample expression levels corresponding to the 7 key factors were plotted. The horizontal axis represents the disease grouping, which is divided into pancreatic ductal adenocarcinoma (PDAC), benign control (Benign), and healthy control (Control). The vertical axis represents the expression level of the gene in the sample. The gene expression values ​​have been standardized, and the data are expressed as mean ± standard error. The P value above the bar graph is the corrected P value in the difference analysis results between the two groups.

[0063] The phenotypes of the seven key factors were used to predict pancreatic ductal adenocarcinoma, and the model prediction accuracy was 0.81. The phenotypes of the seven key factors were used to distinguish the three types of subjects, and the model prediction accuracy was 0.80. The importance scores, i.e., weights, of the seven key factors were extracted from the trained model.

[0064] The specific relationship between weight and prediction result: The weight indicates the contribution of the feature to the model prediction. The weight is related to the correlation between the factor itself and the disease. The higher the weight, the greater the correlation with the disease, and vice versa. The lower the weight, the smaller the correlation with the disease. Among the key factors used to predict PDAC, CTSS is the most important and has the highest weight for the prediction result. The final prediction result is the judgment of whether it is pancreatic ductal adenocarcinoma or not, which is determined by the prediction value. When the prediction value reaches a significant value, after inputting new data into the trained model, the model outputs "PDAC" to indicate the risk of pancreatic ductal adenocarcinoma. If the value is not reached, the output is "none" to indicate that there is no risk of pancreatic ductal adenocarcinoma. The output statement is the final evaluation result.

[0065] The final calculation formula for the probability of pancreatic ductal adenocarcinoma in this model is: (1) (K=3, k=1, 2, 3, indicating there are three categories) in (2) (k = 1, 2, 3, indicating that there are three categories, p is the number of genes, and j represents the jth characteristic gene) Substituting specific values ​​into the above formula (2) yields: (3) In formula (1), x represents the characteristic gene expression, and z represents the predicted output value of the model, which is proportional to the probability. P represents the probability that the screening result is the k-th category. When k=1, P represents the probability that the result is a healthy control, when k=2, it represents the probability that the result is a benign control, and when k=3, it represents the probability that the result is pancreatic ductal adenocarcinoma. β0 represents the probability that Z is the k-th category. k The intercept (bias), β 1-6 represents the weight of the expression value of each gene in the plasma of the sample in the kth category (Table 2), X 1-6 It indicates the specific expression value of each gene in the sample in plasma. The gene order is TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2, and FCER1G.

[0066] In addition, the inventors conducted experiments on other characteristic gene combinations different from the characteristic genes according to the present invention, and obtained the following results: Figure 6 and Figure 7Sample accuracy results of the pancreatic ductal adenocarcinoma probability model. Figure 6 and Figure 7 Two results with better AUC are shown, where Figure 6 The results obtained after targeting the key factors: RUNX3, TMEM222, CTTNBP2NL, S100A4 and MPZL1 and adjusting the hyperparameters using the method described above, the overall AUC was about 70%; Figure 7 The results obtained after adjusting the hyperparameters using the method described above for the key factors: E2F2, RUNX3, TMEM222, DMAP1, PDZK1IP1 and PDE4B, the overall AUC was also about 70%.

[0067] Compared with Comparative Examples 1 and 2, the overall AUC of the model provided by the present invention reaches 80%, indicating the high accuracy and effectiveness of the model in making decisions on given data, and has excellent effects in the early screening of pancreatic ductal adenocarcinoma.

[0068] In summary, the present invention provides a combination of 7 new markers for pancreatic ductal adenocarcinoma in human plasma, as well as the weight of each new marker, and constructs an early auxiliary screening model based on machine learning that can be used to predict the probability of pancreatic ductal adenocarcinoma. The model can reduce the inaccuracy of traditional simple bioinformatics analysis plus human subjective judgment verification, reduce the interference of human factors to a certain extent, improve the credibility and accuracy of the test results, and achieve disease prevention and early intervention. The use of the model can be better and more conveniently applied to clinical practice; at the same time, the screening marker combination for pancreatic ductal adenocarcinoma provided can integrate the information provided by different markers, overcome the limitations of a single marker, and thus improve the sensitivity and specificity of diagnosis. The pancreatic ductal adenocarcinoma probability model established by the present invention has better performance, low cost, and short cycle, and can be widely used in the early screening of pancreatic ductal adenocarcinoma.

[0069] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A screening marker combination for pancreatic ductal adenocarcinoma, characterized in that: The marker is a combination of the following characteristic genes: TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G.

2. The screening marker combination for pancreatic ductal adenocarcinoma according to claim 1, characterized in that: Each characteristic gene of the marker combination is obtained based on the cfRNA algorithm ReguGLM of GLM.

3. A screening and prediction system for pancreatic ductal adenocarcinoma, characterized in that: include: an acquisition module, used to obtain the expression level of a screening marker for pancreatic ductal adenocarcinoma; A prediction module, comprising a module for obtaining a prediction value according to the expression level obtained by the acquisition module, and outputting a disease risk assessment result according to the prediction value; Wherein, the prediction module includes a pancreatic ductal adenocarcinoma probability model, and the pancreatic ductal adenocarcinoma probability model includes a mathematical formula: (K=3,k=1,2,3),(1),, in (k=1, 2, 3), (2), Among them, in formula (2), p is the number of characteristic genes, j represents the jth characteristic gene; Substituting specific values ​​into the above formula (2) yields: (3)， In formula (1), P represents the probability that the screening result is of category k. When k = 1, P represents the probability that the result is a healthy control. When k = 2, P represents the probability that the result is a benign control. When k = 3, P represents the probability that the result is pancreatic ductal adenocarcinoma. β0 represents the probability that the kth category is z k The intercept, β 1-6 represents the weight of the expression value of each gene in the plasma of the sample in the kth category, X 1-6 It indicates the specific expression value of each gene in the plasma of the sample, and the order of each gene is TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G; Wherein, the acquisition module and the prediction module are connected wirelessly and / or wiredly.

4. The screening prediction system according to claim 3, characterized in that: The marker is the screening marker combination for pancreatic ductal adenocarcinoma according to claim 1.

5. A method for training a pancreatic ductal adenocarcinoma probability model for use in the screening prediction system of claim 3, characterized in that: The following steps are involved: S11. Sample data collection: plasma samples from multiple pancreatic ductal adenocarcinoma patients, healthy controls and benign controls were collected, and the expression levels of the characteristic genes TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G in the samples were determined; S12. Data labeling: label the samples into three categories: healthy control, benign control and pancreatic ductal adenocarcinoma, where the healthy control is the first category (k=1), the benign control is the second category (k=2), and the pancreatic ductal adenocarcinoma is the third category (k=3); S13, feature engineering: standardizing the expression levels of the feature genes to obtain standardized expression values ​​of the 7 feature genes in each sample; S14. Model construction: A probability model is established based on the generalized linear model, and the following mathematical formula is set: Construct the category probability formula: (k=1,2,3)，(1)， Where: In the above formula (1), k represents the type, which is divided into 1, 2, and 3 categories, P represents the probability of type k when x is input, K represents the number of categories, which is 3 here; z k is the linear prediction value of the kth class, expressed as: (j=1,2,...,p), (2), In the above formula (2), β is the regression coefficient to be estimated, k = 1, 2, 3, indicating that there are three categories, p is the number of characteristic genes, j represents the jth characteristic gene, and β k0 is the intercept of category k, β kj is the weight of the jth characteristic gene in category k, x j is the expression value of the jth characteristic gene; The gradient of the loss function is calculated as follows: (4)， In the above formula (4), k represents the kth category, n is the number of samples, and β is approximated by using the quasi-Newton method to find its numerical solution; S15, parameter training: Through the maximum likelihood estimation method, based on the labeled sample data, the model parameters β are trained k0 and β kj ; S16. Model validation: The classification performance of the model was evaluated using the cross-validation method, and the accuracy, sensitivity, and specificity of the model were calculated; S17, model optimization: adjust the model parameters according to the verification results until the classification accuracy requirements of the screening prediction system are met; S18. Model output: deploying the finally optimized pancreatic ductal adenocarcinoma probability model into the screening prediction system as a prediction module of the system described in claim 3.

6. A method for screening and predicting pancreatic ductal adenocarcinoma, characterized in that: The following steps are involved: S21, detecting the expression levels of characteristic genes TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2, and FCER1G in patient samples; S22, inputting the test result into the screening and prediction system for pancreatic ductal adenocarcinoma according to claim 3 to obtain a predicted value; S23. When the predicted value reaches a preset value, the system outputs "PDAC" to indicate that there is a risk of pancreatic ductal adenocarcinoma; when the predicted value reaches a preset value, the system outputs "none" to indicate that there is no risk of pancreatic ductal adenocarcinoma.

7. A method for obtaining the expression level of a screening marker combination for pancreatic ductal adenocarcinoma, characterized in that: The following steps are involved: S31. Sample collection: Collect plasma samples from patients; S32. Nucleic acid extraction: extracting circulating free RNA (cfRNA) from the plasma sample using a nucleic acid extraction kit; S33, reverse transcription: converting the extracted cfRNA into cDNA through reverse transcription reaction; S34, quantitative PCR: Real-time quantitative PCR technology was used to amplify and detect the expression levels of target genes in cDNA using specific primers and probes, including the following characteristic genes: TMEM222, PDZK1IP1, PDE4B, CTTNBP2NL, CTSS, TAGLN2 and FCER1G; S35. Data standardization: Standardize the results of quantitative PCR to obtain the relative expression level of each characteristic gene; S36. Result output: The standardized gene expression level data is output as input data of the screening prediction system for further analysis and prediction of the risk of pancreatic ductal adenocarcinoma.

8. The method according to claim 7, characterized in that The marker is the screening marker combination for pancreatic ductal adenocarcinoma according to claim 1.

Images