CN115575635A - A diagnostic marker for cholangiocarcinoma and its screening method and application - Google Patents

Abstract

The invention relates to the field of clinical examination and diagnosis, in particular to a bile duct cancer diagnosis marker and a screening method and application thereof, wherein the diagnosis marker comprises the following 6 marker combinations: clusterin, indirect bilirubin, low density lipoprotein cholesterol, gamma glutamyltransferase, carbohydrate antigen 19-9, triglyceride; the invention adopts proteomics technology and artificial intelligence data analysis technology to obtain a plurality of diagnostic markers suitable for bile duct cancer diagnosis, the diagnostic markers comprise markers in bile and serum, the AUC of the marker combination is obviously higher than that of the independent CLU, the screening method of the diagnostic markers has strong operability, the model construction method is simple, the sensitivity of the obtained diagnostic markers is high, the specificity is good, the diagnostic markers are suitable for bile duct cancer diagnosis, the defects of the existing imaging diagnosis mode can be well made up, the diagnosis of the invention is simple and quick, the early diagnosis and early treatment of bile duct cancer are facilitated, and the invention has good clinical use and popularization values.

Description

A diagnostic marker for cholangiocarcinoma and its screening method and application

technical field

The invention relates to the field of clinical examination and diagnosis, in particular to a diagnostic marker for cholangiocarcinoma and a screening method and application thereof.

Background technique

Cholangiocarcinoma (CCA) is considered a highly aggressive malignancy. According to the anatomical location of the lesion, CCA can be divided into intrahepatic cholangiocarcinoma (iCCA), hilar cholangiocarcinoma (pCCA) and distal cholangiocarcinoma (dCCA) ^[1] . As the second most common malignant tumor of the hepatobiliary system, cholangiocarcinoma has a poor prognosis, a low 5-year survival rate (7%-20%), and a high mortality rate (about 2% of the annual global cancer-related deaths), mainly due to Early diagnosis is difficult ^[1] . Diagnosis of cholangiocarcinoma is difficult due to its indistinct clinical features and deep anatomical location. At the same time, the postoperative pathological results of several studies pointed out that 10%-25% of patients with suspected cholangiocarcinoma who received surgical treatment ended up with no cancer cells ^[4,5] , so researchers are urgently required to develop more accurate diagnostic tools for cholangiocarcinoma .

At present, cholangiocarcinoma is mainly detected by imaging methods, but its diagnostic effect is not satisfactory, the accuracy is not high, and the sensitivity is only 6%-71.9% ^[2] . Serum CA19-9 is commonly used in the clinical diagnosis of cholangiocarcinoma, but its sensitivity and specificity are poor ^[3] . Bile is mainly secreted by hepatocytes and bile duct epithelial cells, and abnormal changes in bile components often occur when biliary tract diseases occur. Cancer-associated proteins in cholangiocarcinoma tissue can be secreted into bile and may have the potential to be diagnostic markers ^[6] . For example, invention patent CN108982854B discloses the application of protein MUC13 in the preparation of reagents for diagnosing intrahepatic cholangiocarcinoma; invention patent CN111413447A discloses the application of chenodeoxycholic acid or/and taurine chenodeoxycholic acid in the diagnosis of cholangiocarcinoma.

In the inventor's previous research, it was first discovered and confirmed that the cancer-related protein clusterin CLU in bile can be used to diagnose cholangiocarcinoma. At the same time, the inventor also studied its combination with CA19-9 in the diagnosis of cholangiocarcinoma effect, and applied for the invention patent of CLU and its composition in the diagnosis of cholangiocarcinoma and the diagnostic kit for cholangiocarcinoma (2021103642823). The previous research results showed that the expression of cancer-related proteins clusterin CLU and CA19-9 in bile The expression level of CLU in the bile of patients with cholangiocarcinoma was significantly higher than that in the bile of patients with bile duct stones, indicating that CLU can be used as a tumor marker for the diagnosis of cholangiocarcinoma.

During follow-up research, the inventors found that many serum markers, such as alkaline phosphatase and bilirubin, may change in cholangiocarcinoma and benign biliary tract diseases, so they cannot be used alone for the diagnosis of cholangiocarcinoma ^[7] . Differential proteins in bile mainly reflect local changes, while serum markers mainly reflect systemic changes in the progression of cholangiocarcinoma ^[2] . On the basis of the previous research, in order to further improve the accuracy of diagnosis, the inventor added some serum indicators, combined with the machine learning model, and screened out the marker combination of bile and serum, which can improve the identification of cholangiocarcinoma and the accuracy of benign biliary strictures, and specifically disclosed a screening method for markers.

references:

[1] Banales J M, Marin J J G, Lamarca A, et al. Cholangiocarcinoma 2020: the next horizon in mechanisms and management [J]. Nat Rev Gastroenterol Hepatol, 2020, 17(9): 557-88.

T,Metzger J,Husi H,et al.Bile and urine peptide markerprofiles:access keys to molecular pathways and biological processes incholangiocarcinoma[J].J Biomed Sci,2020,27(1):13.[2]

T, Metzger J, Husi H, et al. Bile and urine peptide marker profiles: access keys to molecular pathways and biological processes incholangiocarcinoma[J].J Biomed Sci,2020,27(1):13.

[3] Blechacz B, Komuta M, Roskams T, et al. Clinical diagnosis and staging of cholangiocarcinoma [J]. Nat Rev Gastroenterol Hepatol, 2011, 8(9): 512-22.

[4] Nuzzo G, Giuliante F, Ardito F, et al. Improvement in perioperative and long-term outcome after surgical treatment of hilar cholangiocarcinoma: results of an Italian multicenter analysis of 440 patients [J]. Arch Surg, 2012, 147(1) :26-34.

[5] Banales J M, Cardinale V, Carpino G, et al. Expert consensus document: Cholangiocarcinoma: current knowledge and future perspectives consensus statement from the European Network for the Study of Cholangiocarcinoma (ENS-CCA) [J]. Nat Rev Gastroenterol Hepatol, 2016,13(5):261-80.

[6] Lankisch T O, Metzger J, Negm A A, et al. Bile proteomic profiles differentiate cholangiocarcinoma from primary sclerosing cholangitis and choledocholithiasis [J]. Hepatology, 2011, 53(3): 875-84.

[7] Rizvi S, Gores G J. Pathogenesis, diagnosis, and management of cholangiocarcinoma [J]. Gastroenterology, 2013, 145(6): 1215-29.

Contents of the invention

In view of the above technical problems, the present invention provides a diagnostic marker for cholangiocarcinoma, which comprises a combination of the following three substances: clusterin (CLU), indirect bilirubin (IBIL), low-density lipoprotein Protein cholesterol (LDLC).

Preferably, the diagnostic markers include a combination of the following four substances: clusterin (CLU), indirect bilirubin (IBIL), low-density lipoprotein cholesterol (LDLC), gamma-glutamyl transferase (GGT ).

Preferably, the diagnostic markers include a combination of the following five substances: clusterin (CLU), indirect bilirubin (IBIL), low-density lipoprotein cholesterol (LDLC), gamma-glutamyl transferase (GGT ), carbohydrate antigen 19-9 (CA19-9).

Preferably, the diagnostic markers include a combination of the following 6 substances: clusterin (CLU), indirect bilirubin (IBIL), low-density lipoprotein cholesterol (LDLC), gamma-glutamyl transferase (GGT ), carbohydrate antigen 19-9 (CA19-9), triglyceride (TG).

The present invention also provides a screening method for the diagnostic markers of cholangiocarcinoma, comprising the following steps:

(1) Screening of bile markers: Proteomic analysis of bile and cell supernatant was carried out by liquid chromatography-mass spectrometry, and the differentially expressed proteins between cholangiocarcinoma and control group were analyzed and identified; The abnormally highly expressed proteins in the cell supernatant were intersected to obtain bile markers;

(2) Mix the bile markers and serum indicators screened in step (1), use the random forest method to establish a classification prediction model, and rank each marker according to the importance of the prediction results in the cross-validation set, and use the R language The glment software package, based on the 10-fold cross-validation classification method, divides all index data of 287 patients into 10 non-overlapping parts, of which 2 groups are used for the test cohort and 8 groups are used for the training cohort;

(3) According to the Gini index ≥ 0.25, 12 markers are screened out, and the 12 markers are included in the initial input variables of the Lasso classifier training set. Only the variables that contribute to the classification are given non-zero weights. When adding markers The number of objects, when the accuracy, sensitivity and specificity of the Lasso classifier no longer increase, the performance of the Lasso classifier reaches the best accuracy, sensitivity and specificity on the test set;

(4) The receiver operating characteristic ROC curve was used to evaluate the best diagnostic performance of the Lasso model, and the best accuracy, sensitivity and specificity on the ROC curve were used as the cut-off point to obtain the relatively optimal number of features and their combination;

(5) Using the relatively optimal number of features and combination methods obtained in the ROC curve verification step (4) in the external verification set to obtain a marker combination suitable for the diagnosis of cholangiocarcinoma.

Preferably, the random forest and LASSO are performed in glment version 4.1-3.

Preferably, a total of 2000 decision trees are planted in step (2).

Preferably, the serum indicators described in step (2) include 37 blood biochemical indicators, 24 routine blood indicators and two tumor biomarkers.

The present invention also provides the application of the diagnostic markers in the preparation of diagnostic products for cholangiocarcinoma.

Preferably, the diagnostic products include kits, reagents or chips.

The present invention also provides a diagnostic kit for cholangiocarcinoma, comprising the diagnostic marker.

The beneficial effect of the present invention is: (1) the advantage of the present invention is to adopt proteomic technology and artificial intelligence data analysis technology to obtain the diagnostic marker that is suitable for the diagnosis of cholangiocarcinoma, and described diagnostic marker includes the marker in bile and serum The experimental results showed that Six-panel showed good predictive ability, with AUC of 0.926, sensitivity of 86.2%, and specificity of 85.3%, which were significantly higher than that of CLU alone (AUC of 0.840). (2) The diagnostic marker screening method of the present invention has strong operability, simple model construction method, and the obtained diagnostic marker has high sensitivity and good specificity, and is suitable for the diagnosis of cholangiocarcinoma. (3) The present invention combines bile and serum markers to further increase the reliability of diagnosis, and can well make up for the shortcomings of existing imaging diagnosis modes, and the present invention is simple and fast in diagnosis, which is conducive to the early diagnosis of cholangiocarcinoma Early treatment has good clinical application and promotion value.

Description of drawings

Figure 1 Flowchart of bile and cell supernatant proteomic screening for candidate markers of cholangiocarcinoma

Figure 2 Heat map of differentially expressed proteins in bile and string dendrogram of protein clustering

Note: B The heat map of differentially expressed proteins in bile analyzed by LC-MS/MS, N1-N4 represent the bile of 4 patients with benign biliary stricture, and T1-T5 represent the bile of 5 patients with cholangiocarcinoma. C Differential expression in bile obtained by KEGG analysis.

Fig. 3 Heat map of differentially expressed proteins in cell supernatant and chord tree of protein clustering Note: D Heat map of differentially expressed proteins in cell supernatant; E String tree of differentially expressed proteins in supernatant obtained by KEGG analysis picture.

Figure 4 Venn diagram of up-regulated proteins in cell supernatants analyzed by LC-MS/MS

NOTE: F Venn diagram of 54 upregulated proteins in supernatants of four cholangiocarcinoma cells analyzed using LC-MS/MS. G five upregulated proteins in bile and cell supernatant, and their gene names are listed on the right. H CLU proteins were also upregulated in proteomic data from another external bile group.

Figure 5 Western blot and immunohistochemical images of CLU in bile

Note: A Western blot analysis of CLU in the bile of 8 patients with cholangiocarcinoma and 8 patients with benign biliary strictures. B Representative immunohistochemical images, the expression level of CLU in cholangiocarcinoma tissue and interlobular bile duct tissue (Normal), the red arrow points to the interlobular bile duct.

Figure 6 Overall survival (OS) and recurrence-free survival curves of CLU in cholangiocarcinoma

Figure 7 Western blot analysis and mRNA levels of CLU in cells and cell supernatants of four cholangiocarcinoma cell lines and HIBEpiC cell line

NOTE: E and F Immunoblot analysis of CLU in cells and cell supernatants from four cholangiocarcinoma cell lines and the HIBEpiC cell line. G mRNA levels of CLU in four cholangiocarcinoma cell lines and HIBEpiC cell line.

Figure 8 Western blot analysis and mRNA levels of CLU in cells and cell supernatants of five primary cholangiocarcinoma cells and HIBEpiC cells

NOTE: H and I Immunoblot analysis of CLU in cells and cell supernatants from five primary cholangiocarcinoma cells and HIBEpiC cells. J CLU mRNA levels in 5 primary cholangiocarcinoma cells and HIBEpiC cells. *P<0.05, **P<0.01, ***P<0.001.

Figure 9 ELISA determination of CLU or CA19-9 in bile and serum.

Note: A Patient cohort used for the CLU pilot study. B ELISA analysis of CLU in serum from 40 patients with cholangiocarcinoma, 40 patients with benign biliary stricture and 40 healthy volunteers. C ELISA analysis of CLU in bile of 40 patients with cholangiocarcinoma and 40 patients with benign biliary stricture. D Patient cohort used for the CA19-9 pilot study. E and F ELISA analysis of CA19-9 in serum and bile of 40 patients with cholangiocarcinoma and 40 patients with benign biliary stricture. G Patient cohort in the cross-validation set. H and I ELISA analysis of bile CLU and serum CA19-9 from 123 patients with cholangiocarcinoma and 164 patients with benign biliary strictures.

Figure 10 ROC curves of CLU or CA19-9 in bile and serum Note: ROC curves of J CLU, CA19-9 and CLU&CA19-9, the blue curve represents CLU, the green curve represents CA19-9, and the yellow curve represents CLU&CA19-9. tSNE analysis of K CLU&CA19-9, orange represents benign bile duct stricture, and light blue represents cholangiocarcinoma. DCA analysis of L CLU, CA19-9 and CLU&CA19-9, green represents CLU&CA19-9, blue represents CLU, red represents CA19-9. *P<0.05, **P<0.01, ***P<0.001.

Figure 11 The random forest model screened out the top 30 markers

Note: According to the accuracy rate (left) and Gini index (right), the features closer to the upper right are more important.

Figure 12 The accuracy of the multi-marker joint diagnostic model screened by the LASSO model

Note: B The AUC and accuracy (ACC) values of the multi-marker combined diagnostic model of different biomarker combinations screened by the LASSO model; C The sensitivity and specificity of the multi-marker combined diagnostic model.

Figure 13 Accuracy analysis of Six-panel joint diagnosis model

Note: ROC curve of D Six-panel and its members; E Correlation matrix between six markers of Six-panel, including CLU, IBIL, LDL-C, CA19-9, GGT and TG, numbers represent two features The correlation coefficient (r) between; F Six-panel tSNE analysis; G Six-panel, CLU and CA19-9 DCA analysis, AUC is the area under the curve. r≥0.8 means high correlation, 0.5≤r<0.8 means strong correlation, 0.3≤r<0.5 means weak correlation, and r<0.3 means no correlation.

detailed description

The present invention will be further explained below in conjunction with specific implementation examples. The embodiments of the present invention are only used to explain the present invention, and are not meant to limit the protection scope of the present invention.

Carbohydrate antigen 19-9 (CA19-9) described in the following examples refers to a tumor marker related to pancreatic cancer, gallbladder cancer, colon cancer and gastric cancer, also known as gastrointestinal tract-related antigen. Carbohydrate antigen 19-9 has higher sensitivity and better specificity for pancreatic cancer, and its positive rate is between 85% and 95%.

The AUC (Area Under Curve) described in the following embodiments is defined as the area enclosed by the ROC curve and the coordinate axis, and obviously the value of this area will not be greater than 1. And because the ROC curve is generally above the straight line y=x, the value range of AUC is between 0.5 and 1. The closer the AUC is to 1.0, the higher the authenticity of the detection method; when it is equal to 0.5, the authenticity is the lowest and has no application value.

"Benign biliary stricture" as used herein refers to cicatricial narrowing of the bile duct lumen due to bile duct injury and recurrent cholangitis or congenital. Benign biliary strictures can be stimulated by iatrogenic injury, abdominal trauma, gallbladder stones, bile duct stones, and bile duct inflammation, leading to the proliferation of fibrous tissue in the bile duct wall, thickening of the tube wall, and gradual narrowing of the lumen of the bile duct. Bile duct stones and cholangitis are clinically The most common benign biliary stricture. Clinical manifestations include abdominal pain, chills, high fever, and intermittent jaundice. Antibiotic treatment is feasible in the early stage, but surgical treatment is the fundamental treatment method for this disease.

The "cholangiocarcinoma" mentioned herein refers to malignant tumors that occur in the extrahepatic bile duct, that is, the left and right hepatic ducts to the lower part of the common bile duct. The etiology of this disease is still not clear, mostly occurs in 50-70 years old, men are slightly more than women, this disease may be related to hepatolithiasis and primary sclerosing cholangitis, most patients with cholangiocarcinoma will have jaundice, usually without Painful, progressive jaundice, abnormal stools, grayish-white stools like kaolin, darkened urine like strong tea, and gallbladder enlargement due to cancer in the middle and lower bile ducts.

Decision tree is a relatively classic machine learning algorithm. It usually selects the optimal feature recursively, and divides the training data according to the feature, so that each sub-data set has the best classification process. The random forest (RF) is actually multiple decision trees, and the relative importance of the included features to the prediction of the target variable can be evaluated by the relative order (ie depth) of feature usage.

The Gini index is a commonly used feature importance evaluation method and a measure of node journals. The greater the decline in the Gini coefficient, the greater the contribution of this variable to the high purity in node splitting.

LASSO regression is a model in which the constraint term of the L1 norm is added after the cost function of the linear regression model. Variable screening and complexity adjustment can be performed by controlling the parameter lambda. In this paper, various combinations of different numbers of markers are verified. ,

The ROC (Receiver Operating Characteristic) curve described in the following examples is also called the receiver operating characteristic curve.

Example 1. Screening of bile markers and their application in the diagnosis of cholangiocarcinoma

1. Case and sample collection

This study was approved by the Human Research Ethics Committee of the First Hospital of Lanzhou University (LDYYLL2022-381), exempted from informed consent, and conducted in accordance with the Declaration of Helsinki Principles. Clinical specimens were obtained from two centers.

A total of 514 patients were divided into discovery set, cross-validation set and external validation set for study. In the discovery set, nine cases of bile from the First Hospital of Lanzhou University were collected for proteomic analysis.

In the discovery set, there were 9 patients.

In the cross-validation set, 287 patients were recruited from January 2019 to March 2022 in the Department of General Surgery of Lanzhou University First Hospital, including 123 patients with cholangiocarcinoma and 164 patients with benign biliary strictures. Among them, benign biliary strictures are mainly divided into three categories, one is common bile duct (CBD) stones with cholangitis, one is intrahepatic bile duct (IBD) stones with cholangitis, and the other is CBD and IBD stones with cholangitis. The diagnosis of patients with cholangiocarcinoma is mainly based on pathological results, and the specific information of the patients is shown in Table 1

In the external validation set, 87 patients with cholangiocarcinoma were recruited from the Cancer Hospital of the Chinese Academy of Medical Sciences from January 2021 to May 2022, and 131 patients with benign cholangiocarcinoma were recruited from the First Hospital of Lanzhou University from January 2022 to May 2022. Patients with biliary strictures.

Bile is mainly collected during ERCP or PTCD or surgery. Bile and serum samples were transported on ice immediately after acquisition, then centrifuged at 3000 × g for 15 min at 4 °C, and the supernatants were collected and stored at −80 °C until experimentation.

Table 1 Basic information of patients

Note: BBS: benign biliary obstruction; CBD: common bile duct stones; IBD: intrahepatic bile duct stones; TBIL: total bilirubin; GGT: γ-glutamyl transferase

2.LC-MS/MS

(1) Protein sample preparation

① Take out the preserved bile and place it on ice for operation. The acetone was pre-cooled, and an appropriate amount of acetone was added to the bile for shock, to make it fully mixed, and placed in a -20°C refrigerator overnight for precipitation.

②The next day, centrifuge the above mixture at 4°C and 12000g for 10 minutes, discard the supernatant with a pipette, and pay attention to avoid washing away the sediment at the bottom during the process, and keep the sediment at the bottom of the tube.

③ Take an appropriate amount of pre-cooled acetone and add the above precipitate, place the mixture in a vortex apparatus to shake and mix thoroughly, then centrifuge again under the same centrifugation conditions as above for 15 minutes, discard the supernatant to collect the precipitate. Repeat this step one more time.

④Put the precipitate obtained above in a fume hood to dry at room temperature, add an appropriate amount of tissue lysate to it, and use a vortex instrument to oscillate and mix to fully lyse the protein.

⑤ Centrifuge the above mixture at 12000g for 15 minutes at room temperature, take the supernatant and place it in a new EP tube, and centrifuge the supernatant again to ensure that insoluble impurities are fully removed.

⑥ The supernatant obtained at last is the total protein solution of bile, and then the protein concentration of each sample is measured and stored in aliquots.

(2) Detection of protein concentration

① Take out the bile protein sample and the ultrafiltered cell supernatant and place them on ice for operation.

② Prepare the BCA working solution according to the instructions of the BCA kit, and prepare according to 200ul per well (200-300ul can be added when preparing to ensure that the working solution is sufficient).

③Prepare an appropriate amount of 0.5mg/ml protein standard solution according to the BSA instructions, and then add the protein standard product and diluent to the standard wells of the 96-well plate according to the corresponding dilution factor. Two sets of duplicate wells can be set up to improve accuracy.

④ Add 2ul of the sample to be tested in each sample well. In order to reduce the error, 2 or 3 replicate wells can be set up for each sample, and 18ul of standard dilution solution is added to each well, for a total of 20ul.

⑤ After adding the above reagents, add the BCA working solution, 200ul per well (to avoid air bubbles during the addition process), and then place the 96-well plate after the addition of the samples at 37°C for 30 minutes.

⑥ Use a microplate reader or an ultraviolet spectrophotometer to measure the absorbance value at A562nm of each well, use the concentration of the standard well and the corresponding OD value to make a standard curve, and then use the OD value in each sample well to calculate the protein concentration.

(3) SDS-PAGE electrophoresis

① Put the sample on ice for the whole process, add a certain amount of 3× loading buffer according to the protein concentration and volume of the sample, mix well and place it in a metal bath at 100°C for 5 minutes to completely denature it.

② Prepare SDS-PAGE gel with a concentration of 12%, take 10ug of each sample for electrophoresis detection experiment, set the electrophoresis conditions as stacking gel 70V, separation gel 120V, and then electrophoresis under the set constant voltage conditions.

③ After electrophoresis, open the glass plate, gently remove the gel and cut off the upper and lower redundant gels, place the remaining gel in Coomassie brilliant blue staining solution for staining, and continue washing with water until the background is clear.

④ After washing with water, scan the gel with an ImageScanner scanner.

(4) Protein reductive alkylation and enzymatic hydrolysis

① After protein quantification, take 100ug of each sample, add 5 times the volume of pre-cooled acetone to each sample, and place it in a -20°C refrigerator for 1 hour to fully precipitate the protein.

②Centrifuge the above mixture at 4°C and 12,000 rpm for 10 minutes, suck up the supernatant with a pipette gun and discard it, and vacuum freeze the bottom sediment to make it dry.

③In order to dissolve the dried precipitate, add Dissolution Buffer to it, after the protein is fully dissolved, add 4ul Reducing Reagent to it, and then place the mixture at 60°C for 1 hour.

④ Add Cysteine-Blocking Reagent to the above reaction mixture, let it undergo reductive alkylation reaction at room temperature, transfer the protein solution to an ultrafiltration tube after the reaction, and centrifuge at 12000rpm for 25 minutes.

⑤ Add 100 μl Dissolution Buffer to the ultrafiltrate collected above, and centrifuge at 12,000 rpm for 20 minutes after it is fully mixed, and repeat this step 3 times.

⑥ Add sequencing-grade trypsin solvent to the collected solution to fully react, then add Dissolution Buffer to the collected peptide, centrifuge and collect the solution at the bottom of the tube.

(5) Protein labeling

①Take out the iTRAQ reagent and add a certain amount of isopropanol to the bile sample according to the instructions, and perform the following operations after it is fully dissolved.

② Take 50 μl of protein samples for reductive alkylation and enzymatic hydrolysis, add iTRAQ reagent and mix thoroughly, place the mixture on a shaker at room temperature to fully react for 2 hours, then add 100ul distilled water to the reaction mixture to terminate the reaction.

③Mix all labeled bile samples according to the above experimental operation, and vortex to mix thoroughly, and then centrifuge to the bottom of the tube.

④ Vacuum freeze-dry the above-treated samples and save them for future use.

(6) LC-MS/MS analysis

① Dissolve the dried histone samples in the mobile phase A solution and mix thoroughly with a vortex shaker.

②Agilent 1200HPLC instrument was used to conduct peptide separation experiments on the above mixture.

③ After the separation of the peptides, combined with TripleTOF5600 system (AB SCIEX) and nanospray III ion source (AB SCIEX) for mass spectrometric identification and analysis.

(7) Protein Qualitative, Quantitative and Identification

After obtaining the LC-MS/MS raw data, use MaxQuant (version 1.6) to analyze the raw file, and compare it with the Swiss-Prot human protein sequence database (updated in February 2019, 20,413 protein sequences) Identify the corresponding protein. At the protein and peptide levels, the false discovery rate of proteins was less than 1% (FDR<1%), and at least two peptides were identified for further data processing.

3. Western blot:

(1) Protein sample preparation and concentration measurement

Add 200ul of cell lysate to the cell culture dish, fully lyse the cells at room temperature for 15 minutes, transfer them to an EP tube and place them on ice for the whole process, then put the lysed protein solution at 4°C at a speed of 12000rpm After centrifugation, the mixture in the EP tube is divided into 3 layers, the lower layer is sediment, the middle layer is cell protein, and the upper layer is lipid. The middle layer is absorbed as cell protein and stored in separate packages; the bile is extracted by acetone method Protein in (same as LC-MS/MS part); cell supernatant protein was obtained by ultrafiltration of cell supernatant. The concentrations of the above three protein solutions were measured using the BCA method.

(2) SDS-PAGE electrophoresis

① Glue making: Scrub the glass plate, check whether the front and back of the glass plate are clean (strictly ensure the cleanliness), adjust the front and back of the glass plate, clamp the glass plate tightly with clips, and then firmly clamp the clips vertically on the horizontal frame . Prepare an appropriate amount of 12% separation gel, and then use a pipette gun to absorb the separation gel for glue filling. When the glue surface reaches the position of the concentrated gel, stop the glue filling. The whole process of glue filling should be done gently to prevent air bubbles. Then use a 200ul pipette gun to gently add a layer of distilled water on the glue, when the liquid level is higher than the top of the glass plate. After 30 minutes, the separating gel had coagulated, and then the upper layer of distilled water was discarded. Prepare an appropriate amount of 5% stacking gel according to the instructions, fill the stacking gel with a pipette until it overflows, and then insert a clean comb, and the stacking gel has agglutinated after about 20 minutes.

②Sample loading: remove the glass plate, wash it, put it into the electrophoresis tank, inject the electrophoretic liquid into the electrophoretic tank, first fill up the inner tank, and then add the electrophoretic liquid to the outer tank, pull out the electrophoretic liquid after injection comb. According to the protein concentration, the total volume of each group of samples to be loaded is calculated in advance, and after the samples are shaken and mixed, the marker and each group of samples are added to the wells in turn.

③ Electrophoresis: Cover the electrophoresis tank, turn on the power, and run electrophoresis under constant voltage conditions. At the beginning, set the voltage to 80V. When the protein sample is electrophoresed to the separation gel, set the condition to constant voltage 120V. Always observe the position of the bromophenol blue band. If the band has reached the lower edge of the glass plate, the electrophoresis can be terminated. Turn off the power, open the cover of the electrophoresis tank and prepare to remove the glass plate.

(3) transfer film

① After the electrophoresis, use a cutting board to gently pry up the small glass plate, and then cut off the useless parts of the stacking gel and separating gel. Prepare a sponge and filter paper on the black side of the transfer clip, then move the glue onto the filter paper, cut a PVDF membrane of appropriate size and place it on the glue, gently roll out the air bubbles on the membrane with a glass rod to ensure that the filter paper and PVDF membrane Close the transfer clamp when there are no air bubbles.

②Put the transfer holder in the transfer tank, pour the prepared 1× transfer solution into the transfer tank, and stop adding liquid when the transfer solution completely submerges the highest point of the filter paper. According to the molecular weight, the membrane transfer conditions were set as a constant current of 200 mA for 90 minutes (different molecular weights correspond to different current values and membrane transfer times).

(4) closed

Remove the transfer clip, mark the front and back sides on the PVDF membrane, then place the membrane face up in an incubation box filled with skimmed milk powder, and place it on a shaker at room temperature to seal for 1 hour.

(5) immune response

The primary and secondary antibodies were diluted to appropriate concentrations with 5% skimmed milk powder, and GAPDH was set as an internal reference. Then incubated overnight at 4°C with rabbit anti-CLU antibody (1:1000, Cell Signaling) or mouse anti-GAPDH (1:2000, Proteintech). The next day, the membrane was removed and washed three times with TBST buffer on a shaker, each time for 5 minutes, and then the membrane was incubated with goat anti-mouse or anti-rabbit IgG (1:2000, Cell Signaling) secondary antibody for 1 hour at room temperature .

(6) Chemiluminescence development

After the secondary antibody incubation, the membrane was washed to remove residual antibody and skimmed milk powder on the membrane, and then soaked in TBST buffer for development. Prepare an appropriate amount of chemiluminescence liquid, and use a chemiluminescence analyzer to collect and analyze images.

4. Immunohistochemical (IHC) staining:

(1) Baking slices: In order to prevent tissue detachment and better dewaxing, the cholangiocarcinoma tissue slices need to be baked in a 75°C oven for 2 hours before operation (the baking time should not be too short).

(2) Dewaxing and hydration of paraffin sections: Immediately after the baking, place the slides in xylene reagent for dewaxing, 10 minutes each time, 3 times in total; then put the slides into the concentration from high to low Aligned ethanol solution was hydrated, and finally the slides were continuously rinsed with distilled water for 2 minutes.

(3) High-temperature and high-pressure antigen repair: Prepare about 1000ml of citrate repair solution, place it in a pressure cooker and heat it to boiling with an induction cooker, then put the plastic frame with slices into the pot vertically (to prevent the plastic frame from falling), and wait for the pressure of the pressure cooker After the valve starts to spray air for 2 minutes, stop heating. After 10 minutes, open the lid and rinse the pressure cooker with continuous running water to cool down (the cooling process should not be too fast).

(4) Endogenous peroxidase blocking: take the slice out of the pressure cooker, rinse with distilled water to remove the remaining citrate repair solution, then place the slice in PBS buffer and wash it on a shaker for 5 minutes. The steps are repeated 3 times. After drying the slices, add an endogenous peroxidase blocking agent dropwise on top of the tissue (the blocking agent should completely cover the tissue), cover the wet box and allow it to react at room temperature for 10 minutes.

(5) Serum blocking: After the blocking, the slides were placed in a cleaning tank filled with PBS buffer, and placed on a shaker to wash 3 times. Dry the slices and drop an appropriate amount of goat serum on top of the tissue, place the slices flat in a wet box, cover the lid and seal at room temperature for 10 minutes.

(6) Primary antibody incubation: Gently shake off the blocking serum on the slices (no need to wash with PBS), place the slices in a wet box and drop the primary antibody (GRP78 dilution concentration is 1:1000), about 150ul per slice Antibody, and then place it in a 4°C refrigerator for overnight incubation (this process should be handled gently to prevent the antibody from flowing out from the edge of the section).

(7) Secondary antibody incubation: the next day, take out the wet box and place it at room temperature. When the temperature of the slices in the wet box returns to room temperature, shake off the primary antibody and wash the slices with PBS buffer. After rinsing and drying the sections, add 150ul horseradish enzyme-labeled goat anti-mouse/rabbit IgG polymer dropwise on top of the tissue on each slide.

(8) DAB color development: After the secondary antibody was incubated at room temperature for 40 minutes, pour off the residual secondary antibody on the slide, wash the slices with PBS buffer on a shaker, dry the slices, and quickly drop enough DAB chromogenic agent (dropping process should be fast, and do a good job of timing), and after judging the end of tissue staining, rinse the slice under tap water immediately to stop the color development.

(9) Hematoxylin counterstaining: Put tissue sections into hematoxylin reagent for hematoxylin staining (different hematoxylin staining time is different), then place in 1% hydrochloric acid alcohol to differentiate and rinse with tap water continuously to turn blue.

(10) Dehydration: Dehydrate the slices in ethanol solutions with concentrations ranging from low to high, and soak for 2 minutes at each concentration. Sections were then placed in a xylene solution in a fume hood (this step was performed in a fume hood).

(11) Sealing: absorb an appropriate amount of neutral gum to seal the slices dried by the hair dryer (select slides of different specifications according to the size of the tissue in the slices, and prevent air bubbles when sealing the slices), and place them in a well-ventilated place after finishing. place.

(12) Image reading: Interpret the results after the slices are dried. Commonly used methods mainly include direct observation of images with a microscope or observation of images after scanning slices with a digital pathology scanner.

(13) Result analysis: The images were analyzed by Image pro plus 6.0 software. The expression intensity of CLU was independently judged by two senior pathologists without knowledge of any clinical and pathological data. The staining intensity is divided into 4 grades, 0 represents negative expression (negative), 1 represents weak expression (weak), 2 represents medium expression (moderate), and 3 represents strong expression (strong). Finally, in order to facilitate statistical analysis of data, we defined negative, moderate and weak expression (0-2 points) as low expression, and strong expression (3 points) as high expression.

5. Enzyme-linked immunosorbent assay-ELISA

ELISA kits were used to detect the levels of CLU (E-TSEL-H0014, Elabscience) and CA19-9 (E-EL-H0637c, Elabscience) in bile or serum. ELISA experiments were performed according to the instructions provided by the manufacturer. Bile and serum need to be diluted before testing to measure CLU level, and the bile and serum are diluted 100 times and 5000 times respectively; in order to measure the level of CA19-9, the bile and serum are diluted 10000 times and 5 times respectively.

6. Statistical Analysis

Continuous variables were presented as median (interquartile range) or mean ± SD (standard deviation) and compared using the Mann-Whitney U test or Student's t-test. Categorical variables were expressed as ratios and compared with each other by chi-square test. The ROC curve was used to evaluate the diagnostic performance of each marker or marker combination, and the Youden index was used to calculate the threshold. The area under the curve (AUC) was calculated by the trapezoidal method, and the larger the value, the better the diagnostic performance. Sensitivity, specificity, and accuracy (ACC) were calculated by standard 2 × 2 contingency tables, which are the main indicators for assessing diagnostic performance. DCA (Decision Curve Analysis) was used to compare the diagnostic value of different clinical diagnostic models or markers. The t-distributed stochastic neighborhood embedding (tSNE) algorithm was used to visually evaluate the effectiveness of the diagnostic model. A P value of less than 0.05 was considered statistically significant. All analyzes were performed using SPSS Statistics20, GraphPad Prism version 7.0 and R version 4.1.0 (R Foundation for Statistical Computing; http://www.R-project.org)

7. Result Analysis

(1) Proteomics of bile and cell supernatant of cholangiocarcinoma

As shown in Figures 1-4, bile and cell supernatant proteomics were used to screen candidate biomarkers for the diagnosis of cholangiocarcinoma (Figure 1A). In bile proteomics, comparing cholangiocarcinoma with benign bile duct strictures, a total of 1585 proteins were identified based on the standard of Fold Change ≥ 5.0 or Fold Change ≤ 0.2, and 167 differentially expressed proteins were screened out, including 130 up-regulated proteins and down-regulated proteins. 37 proteins (Fig. 1A and 2B). The biological functions and key pathways of differentially expressed proteins were annotated by GO analysis and KEGG analysis (Fig. 2C). The results showed that these differentially expressed proteins were mainly related to tumorigenesis and intercellular interactions, including chemokine signaling pathways, inflammation and immunity, endocytosis and lysosomes, etc.

Cell supernatants collected from four cholangiocarcinoma cell lines (TFK-1, HuCCT-1, RBE, and HCCC-9810) and one normal human intrahepatic biliary epithelial cell line (HIBEpiC) were used for label-free quantitative analysis. 932 proteins were identified, including 273 up-regulated proteins and 659 down-regulated proteins (Fig. 1A and 3D). GO and KEGG analyzes revealed that these differentially expressed proteins were associated with signal transduction and immune regulation during tumor progression, including ECM-receptor interactions, endocytic vesicles, and endocytosis (Fig. 3E). Fifty-four proteins were elevated in the cholangiocarcinoma cell line compared with the HIBEpiC cell line (Fig. 4F). A total of five proteins were screened when intersecting upregulated proteins in bile and supernatant (Fig. 4G), including CLU, COL6A1, GOLM1, QSOX1, and IGFBP1. Considering the limited number of our bile samples, another bile proteome dataset (External bile 1) ^[8] from the study by Marut Laohaviroj et al. was cited. Comparing cholangiocarcinoma and controls based on the criterion of fold ≥ 1.5, 63 upregulated proteins were identified in External bilis 1, but only CLU was elevated in External bilis 1 among the 5 candidate proteins (Fig. 4H). Finally, CLU is selected for further research.

(2) High expression of CLU in cholangiocarcinoma

As shown in Figures 5-8, the levels of CLU protein and mRNA in cholangiocarcinoma were verified in clinical specimens and cells. Sixteen bile samples (8 from cholangiocarcinoma and 8 from benign biliary strictures) were collected for validation of CLU protein levels. As shown in Figure 5A, CLU protein levels were higher in cholangiocarcinoma, but little or no expression was expressed in bile with benign biliary strictures. Tissue microarrays (TMA) containing 90 cholangiocarcinoma tissues and 31 interlobular bile duct tissues were used for immunohistochemical staining. CLU was mainly located in the cytoplasm (Fig. 5B). Among 90 cases of cholangiocarcinoma tissues, 89 cases were positive for CLU (98.9%). Among the positively stained cases, the numbers of cases with weak, medium and strong expression were 4 cases (4.5%), 36 cases (40.4%) and 49 cases (55.1%), respectively. Of the 31 cases of interlobular bile duct tissue, 12 cases (38.7%) were negatively stained. Immunohistochemical image analysis showed that CLU protein was significantly increased in cholangiocarcinoma (P<0.001) (Fig. 5B). Kaplan-Meier survival analysis showed that cholangiocarcinoma patients with high CLU levels had shorter overall survival (OS) time (p<0.0001) and recurrence-free survival (RFS) time (p<0.001) (Fig. 6C and 5D). In conclusion, high expression of CLU can promote the progression of CCA. As shown in Figures 7E, 7F and 7G, CLU protein and mRNA were highly expressed in the four cholangiocarcinoma cell lines (P<0.05). We have successfully isolated five primary cholangiocarcinoma cells from postoperative tissues, and CLU was also highly expressed in them (Fig. 8H, 8I and 8J).

(3) Diagnostic value of bile CLU and serum CA19-9 for CCA

In order to verify which of bile CLU and serum CLU is more suitable for the diagnosis of cholangiocarcinoma, 40 cases of bile and blood of patients with cholangiocarcinoma or benign biliary stricture, and blood of 40 healthy volunteers were collected for preliminary research, as shown in Figures 9 and 10 shown. As shown in Figures 9B and 9C, CLU was highly expressed in serum and bile of patients with cholangiocarcinoma compared with controls. The expression level of CLU in serum of cholangiocarcinoma is particularly high, even in healthy people the average expression is 102,028.5±36,784.1ng/ml. However, the mean levels of CLU in bile from cholangiocarcinoma or benign biliary stricture were only 2,458.1±3,366.0 ng/ml and 302.5±283.2 ng/ml, respectively. Highly abundant proteins are not suitable as diagnostic markers due to their low sensitivity. Therefore, bile CLU was identified as a candidate biomarker for CCA. CA19-9 was expressed in both serum and bile (Fig. 9D). As shown in Figures 9E and 9F, the level of CA19-9 in bile or serum was higher in cholangiocarcinoma. The average levels of bile in patients with cholangiocarcinoma and benign biliary stricture were 688307.0±803859.0IU/ml and 293463.6±321862.1IU/ml respectively, and the serum levels were 270.6±361.8IU/ml and 42.5±50.0IU/ml respectively. Similarly, CA19-9 in serum is more suitable as a diagnostic biomarker for cholangiocarcinoma. 287 patients were then included in the cross-validation set for further study (Fig. 9G). As shown in Figures 9H and 10J, bile CLU is highly expressed in cholangiocarcinoma, and the results of ROC analysis showed a good diagnostic ability, with an AUC of 0.857 (sensitivity 73.98%, specificity 93.29%). Serum CA19-9 was highly expressed in cholangiocarcinoma, with an AUC value of 0.809 (84.55% sensitivity, 67.68% specificity) (Fig. 9I and 10J). Since bile CLU has high specificity and low sensitivity, while CA19-9 is just the opposite, we considered building a model including bile CLU and serum CA19-9 for better accuracy. As shown in Figure 10J, the diagnostic value of the CLU&CA19-9 model increased significantly, with an AUC of 0.917, much higher than its two members. Its sensitivity and specificity increased to 88.6% and 82.9%, respectively, indicating better diagnostic performance. The tSNE algorithm can be used to simplify complex confusion matrices, which can help us visualize the distribution of diseases. As shown in Figure 10K, the cholangiocarcinoma group and the control group formed different clusters using tSNE, indicating that CLU & CA19-9 can well differentiate cholangiocarcinoma. Decision curve analysis (DCA) was used to observe the clinical performance of CLU, CA19-9, and CLU&CA19-9, and the results showed that the CLU&CA19-9 model added more clinical combined benefits than CLU or CA19-9 models alone to predict CCA (Fig. 10L) .

Example 2. Screening biomarker combinations by machine learning

1. Screening method

Contains the following steps:

(1) Screening of bile markers: Proteomic analysis of bile and cell supernatant was carried out by liquid chromatography-mass spectrometry, and the differentially expressed proteins between cholangiocarcinoma and control group were analyzed and identified; The abnormally highly expressed proteins in the cell supernatant were intersected to obtain bile markers; the specific screening steps were the same as in Example 1.

(2) Mix the bile markers and serum indicators screened in step (1), use the random forest method to establish a classification prediction model, and rank each marker according to the importance of the prediction results in the cross-validation set, and use the R language The glment software package, based on the 10-fold cross-validation classification method, divides all indicator data of 287 patients into 10 non-overlapping parts, of which 2 groups are used for the test cohort and 8 groups are used for the training cohort; a total of 2000 decision trees were planted tree; the serum indicators include 37 blood biochemical indicators, 24 routine blood indicators and two tumor biomarkers.

(3) According to the Gini index ≥ 0.25, 12 markers are screened out, and the 12 markers are included in the initial input variables of the Lasso classifier training set. Only the variables that contribute to the classification are given non-zero weights. When adding markers The number of objects, when the accuracy, sensitivity and specificity of the Lasso classifier no longer increase, the performance of the Lasso classifier reaches the best accuracy, sensitivity and specificity on the test set; the random forest and LASSO described in glment Made in version 4.1-3.

(4) The receiver operating characteristic ROC curve was used to evaluate the best diagnostic performance of the Lasso model, and the best accuracy, sensitivity and specificity on the ROC curve were used as the cut-off point to obtain the relatively optimal number of features and their combination;

(5) Using the relatively optimal number of features and combination methods obtained in the ROC curve verification step (4) in the external verification set to obtain a marker combination suitable for the diagnosis of cholangiocarcinoma.

2. Result analysis

Table 2 Diagnostic value of markers and multi-index combinations

By combining biomarkers with different types of circulating biomarkers, their diagnostic performance can be improved. The data of each patient used for machine learning contains bile CLU and common 63 blood indicators, including 37 blood biochemical indicators, 24 routine blood indicators and two tumor biomarkers. In the cross-validation set, the above features are classified using a random forest (RF) model. As shown in Figure 11A, the top 30 markers were displayed according to their accuracy (left) and Gini index (right). To select the most suitable features, 12 markers with Gini index ≥0.25 were selected for further study, including CLU, DBIL, TBIL, CA19-9, IBIL, LDLC, GGT, ALP, TG, AST, CL, and ALT. And their AUC values were about equal to or higher than 0.7 (Table 2).

Selecting the most appropriate number of markers is crucial for building the final classification model. Based on the above 12 selected markers, the LASSO method was used for screening. LASSO screened out the best combinations when the combinations contained different numbers of markers. ROC analysis showed that five-panel had the highest AUC value (0.958). The AUC value of the Six-ten panel is the same, both are 0.954, but the ACC value is the highest (Figure 12B and Table 2). As shown in FIG. 12C , the sensitivity and specificity of the Six-ten panel are relatively ideal (sensitivity: 90.2%, specificity: 89.0%). In summary, the Six-panel model exhibited a good balance between complexity and accuracy, and thus was identified as the best model for diagnosing cholangiocarcinoma.

The AUC value of the Six-panel model was significantly higher than that of its six members (Fig. 13D). And there was almost no correlation (r<0.5) among the six biomarkers, indicating that they could form a good diagnostic model (Fig. 13E). The tSNE results of Six-panel showed that the cholangiocarcinoma group and the control group formed different clusters (Fig. 13F), indicating that Six-panel could well distinguish cholangiocarcinoma even under the visualization condition. The results of DCA showed that Six-panel improved more overall clinical benefits in distinguishing cholangiocarcinoma from benign biliary strictures (Fig. 13G). To further evaluate the stability and reliability of Six-panel, we apply it to an independent external validation set. In the external validation set, Six-panel showed good predictive ability, with AUC of 0.926, sensitivity of 86.2%, and specificity of 85.3%, significantly higher than that of CLU alone (AUC of 0.840). tSNE and DCA analyzes also showed good diagnostic capabilities.

In summary, the present invention uses proteomics technology and artificial intelligence data analysis technology to obtain a combination of diagnostic markers suitable for the diagnosis of cholangiocarcinoma. The diagnostic markers include markers in bile and serum. The experimental results show that Six -panel showed good predictive ability, with AUC of 0.926, sensitivity of 86.2%, and specificity of 85.3%, significantly higher than that of CLU alone (AUC of 0.840). The diagnostic marker screening method of the invention has strong operability, simple model construction method, high sensitivity and good specificity of the obtained diagnostic marker, and is suitable for the diagnosis of cholangiocarcinoma. The present invention combines bile and serum markers, further increases the reliability of diagnosis, can well replace the existing imaging diagnosis mode, and the present invention is simple and fast in diagnosis, which is beneficial to the early diagnosis and early treatment of cholangiocarcinoma, and has great advantages. Good clinical use and promotion value.

Claims (10)

1. A diagnostic marker for cholangiocarcinoma, which comprises the following combination of 3 substances: clusterin (CLU), indirect Bilirubin (IBIL), low Density Lipoprotein Cholesterol (LDLC).

2. The diagnostic marker of claim 1, wherein said diagnostic marker comprises a combination of 4 of: clusterin (CLU), indirect Bilirubin (IBIL), low Density Lipoprotein Cholesterol (LDLC), gamma-glutamyl transferase (GGT).

3. The diagnostic marker of claim 2, wherein said diagnostic marker comprises a combination of5 of: clusterin (CLU), indirect Bilirubin (IBIL), low Density Lipoprotein Cholesterol (LDLC), gamma-glutamyltransferase (GGT), and carbohydrate antigen 19-9 (CA 19-9).

4. The diagnostic marker of claim 3, wherein said diagnostic marker comprises a combination of 6 of: clusterin (CLU), indirect Bilirubin (IBIL), low Density Lipoprotein Cholesterol (LDLC), gamma-glutamyl transferase (GGT), carbohydrate antigen 19-9 (CA 19-9), triglycerides (TG).

5. A screening method for the biliary duct cancer diagnostic marker according to any one of claims 1 to 4, comprising the steps of:

(1) Screening of bile markers: performing proteomics analysis on the bile and cell supernatant by adopting a liquid chromatography-mass spectrometry technology, and analyzing and identifying differentially expressed proteins between the bile duct cancer and a control group; taking intersection of bile of the bile duct cancer and protein with abnormally high expression in cell supernatant to obtain a bile marker;

(2) Mixing the bile markers obtained by screening in the step (1) with serum indexes, establishing a classification prediction model by using a random forest method, sequencing each marker according to the importance of a cross validation centralized prediction result, and dividing all index data of 287 patients into 10 groups of non-overlapping parts by using a comment software package of an R language and based on a 10-fold cross validation classification method, wherein 2 groups are used for testing queues, and 8 groups are used for training queues;

(3) Screening 12 markers according to the Gini index of more than or equal to 0.25, bringing the 12 markers into an initial input variable of a Lasso classifier training set, and only giving non-zero weight to the variables contributing to classification; when the number of the markers is increased and the accuracy, the sensitivity and the specificity in the Lasso classifier are not increased any more, the performance of the Lasso classifier achieves the optimal accuracy, sensitivity and specificity on a test set;

(4) The optimal diagnostic performance of the Lasso model is evaluated by referring to an ROC curve of the working characteristics of a testee, and the optimal accuracy, sensitivity and specificity on the ROC curve are taken as cut-off points to obtain the relatively optimal characteristic number and the combination mode;

(5) And (3) obtaining a marker combination suitable for diagnosing the bile duct cancer by using the relatively optimal characteristic number and combination mode obtained in the ROC curve verification step (4) in an external verification set.

6. The screening method of claim 5, wherein said random forest and LASSO are performed in comment version 4.1-3.

7. The screening method according to claim 5, wherein 2000 decision trees are co-planted in step (2), and the serum markers comprise 37 blood biochemical markers, 24 conventional blood markers and two tumor biomarkers.

8. Use of a diagnostic marker as defined in any one of claims 1 to 4 in the manufacture of a product for the diagnosis of biliary duct cancer.

9. The use of claim 8, wherein the diagnostic product comprises a kit, reagent or chip.

10. A kit for early diagnosis of cholangiocarcinoma, comprising the diagnostic marker according to any one of claims 1 to 4.

Images