CN114509510A

CN114509510A - Blood marker for identifying malignant mesothelioma and application thereof

Info

Publication number: CN114509510A
Application number: CN202110622861.3A
Authority: CN
Inventors: 陈忠坚; 毛伟敏; 高赟
Original assignee: Zhejiang Cancer Hospital
Current assignee: Zhejiang Cancer Hospital
Priority date: 2021-06-04
Filing date: 2021-06-04
Publication date: 2022-05-17

Abstract

The invention discloses a blood marker for identifying malignant mesothelioma and application thereof. The blood marker is a combination of the following nine metabolites: uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, and histidine. The invention combines non-targeted metabonomics and machine learning algorithms to research the differential metabolites between malignant mesothelioma patients and healthy individual control plasma samples, excavates and verifies an optimal classification model consisting of nine metabolites, has good sensitivity and specificity, can realize high-sensitivity and specificity detection on malignant mesothelioma as a blood marker, can be applied to the preparation of diagnostic reagents for identifying malignant mesothelioma, is beneficial to the diagnosis of malignant mesothelioma, improves the screening efficiency, reduces the screening cost, provides direction for individualized treatment, and improves the survival rate and the survival quality of patients.

Description

Blood marker for identifying malignant mesothelioma and application thereof

Technical Field

The invention belongs to the field of biomedical analysis, and particularly relates to a blood marker for identifying malignant mesothelioma and application thereof.

Background

Malignant Mesothelioma (MM) is a rare invasive malignancy, closely associated with asbestos exposure. The median overall survival for malignant mesothelioma is only 7 months, epithelial-like mesothelioma being the most common histological subtype with a median survival of 14 months, and the median survival for sarcoma-like mesothelioma and biphasic mesothelioma of 4 months and 10 months, respectively. The poor prognosis of malignant mesothelioma is due to the difficulty of diagnosis and high resistance to chemotherapy. Malignant mesothelioma has long incubation period, and the peak period of the disease is 45 years after the contact. Malignant mesothelioma is clinically insignificant in early stage, and most patients are already in advanced stage at the time of diagnosis. The treatment method of malignant mesothelioma mainly comprises operations, chemotherapy, radiotherapy and the like, but the curative effect is poor. Newly developed molecular targeted drugs and immune checkpoint inhibitors are expected to improve patient prognosis, but clinical trial results are not satisfactory. Despite the expansion of treatment options in recent years, malignant mesothelioma remains an incurable disease. Early diagnosis of malignant mesothelioma, timely intervention, if possible, is an important approach to improving prognosis.

Clinical diagnosis of mesothelioma includes imaging diagnosis (X-ray and CT) and pathological diagnosis. However, the misdiagnosis rate of image diagnosis is high and is easily confused with lung adenocarcinoma and ovarian cancer. The accuracy of pathological diagnosis after puncture is relatively high, but the diagnosis error rate can still be 50%, and the defects of invasiveness, long time consumption, high price and the like exist. Therefore, the search for novel diagnostic markers can not only improve the clinical diagnosis accuracy of mesothelioma, but also provide a potentially suitable therapeutic target so as to improve the prognostic survival rate of patients.

Metabolomics enables robust quantitative measurements of small molecule metabolites. Metabonomics is based on platforms such as nuclear magnetic resonance spectroscopy (NMR) or Mass Spectrometry (MS), and multivariate statistics are combined to clarify metabolic changes of biological samples after pathophysiological stimulation or gene modification, so that the metabonomics are widely applied to discovery of new biological blood markers and mechanisms of various cancers. Metabolomics can explain the metabolic changes of biological samples under pathophysiological stimuli or genetic modifications, and is therefore widely used for discovering new biological blood markers and mechanisms of various cancers. However, mesothelioma, a rare cancer species, has a problem of difficulty in collecting samples, and limits the establishment of methods based on metabonomics of samples, and therefore, has not been widely used for clinical diagnosis.

Disclosure of Invention

In view of the above, the present invention aims to provide a blood marker for identifying malignant mesothelioma and applications thereof, which are helpful for diagnosing malignant mesothelioma and providing potential therapeutic targets for clinical application.

A blood marker for identifying malignant mesothelioma, which is a combination of nine metabolites: uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, and histidine.

The screening method for identifying the blood marker of the malignant mesothelioma comprises the following steps:

(1) plasma is collected from a malignant mesothelioma patient and a healthy individual, after anticoagulation treatment, a plasma sample of the malignant mesothelioma patient and a healthy control plasma sample are obtained, and all the plasma samples are randomly set into 2 queues: training and testing sets;

respectively taking plasma samples with the same volume from each plasma sample of a malignant mesothelioma patient and each plasma sample of a healthy control, mixing to obtain a mixed plasma sample, and evenly subpackaging the mixed plasma sample into a plurality of parts, wherein each part is a quality control plasma sample;

(2) performing LC-MS analysis pretreatment on each plasma sample of the patient with malignant mesothelioma, each healthy control plasma sample and each quality control plasma sample to respectively obtain a plasma pretreatment sample of the patient with malignant mesothelioma, a healthy control plasma sample and a quality control plasma pretreatment sample; the plasma pretreatment sample of the malignant mesothelioma patient and the healthy control plasma sample are collectively called LC-MS analysis samples, and the quality control plasma pretreatment sample is called QC sample;

(3) performing LC-MS detection and analysis on each plasma pretreatment sample one by one to obtain an original metabolic fingerprint of each sample; in the LC-MS detection and analysis, a QC sample is added into every other several LC-MS analysis samples;

(4) sequentially utilizing MSConvert software and an R language software package XCMS to perform format conversion and ion peak information extraction on the original metabolism fingerprint of each sample to obtain data containing all ion peak information; removing data of ions meeting any one of the following standards (a) to (c) by utilizing Meta X filtering to obtain a preprocessed data matrix, wherein the preprocessed data matrix comprises a training set data matrix and a test set data matrix:

(a) ions not detected in more than 50% of QC samples;

(b) (ii) no ions detected in more than 80% of the LC-MS analysis samples;

(c) ions with relative standard deviation > 30% in QC samples;

(5) modeling the preprocessed data matrix by using OPLS-DA, calculating the variable projection importance of each ion, and screening out differential ions which simultaneously meet the following criteria (i) to (iii):

(i) a variable projection importance score >1 in the OPLS-DA model;

(ii) significance of difference (P-value) <0.05 in t-test;

(iii) the relative concentrations in plasma of a malignant mesothelioma patient differ by a factor >1.5 or <0.667 from that of healthy control plasma;

(6) matching the MS/MS spectrogram corresponding to the m/z value and retention time of each differential ion obtained by screening in the step (5) with the spectrogram in a database, and identifying to obtain n differential metabolites;

(7) in a training set data matrix, using a random forest algorithm to score the feature importance of the n different metabolites, sequencing the n different metabolites from high to low according to the feature importance score, and substituting 2-n metabolic features according to the sequencing of the different metabolites for modeling for multiple times to obtain a plurality of classification models; and comparing the accuracy and the Kappa value of different classification models to evaluate the classification efficiency of different models, and finally screening out an optimal classification model, wherein the optimal classification model is a random forest model containing the following 9 metabolites: uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, and histidine;

(8) and verifying the optimal classification model in a test set data matrix.

In some embodiments of the invention, in step (1), the plasma is collected from a malignant mesothelioma patient and a healthy individual, and is collected from a malignant mesothelioma patient and a healthy individual who are fasted at night.

In some embodiments of the present invention, in step (1), the ratio of the number of samples in the training set to the number of samples in the test set is 2:1 to 4: 1. In one embodiment of the present invention, in step (1), the ratio of the number of samples in the training set to the number of samples in the test set is 43: 14.

In some embodiments of the invention, in step (1), the anticoagulation treatment comprises the following steps: adding dipotassium salt of ethylenediamine tetraacetic acid as anticoagulant into collected blood plasma, centrifuging at 4 ℃, and collecting supernatant to obtain a blood plasma sample.

In some embodiments of the invention, the LC-MS analytical pretreatment in step (2) comprises the following steps: adding pre-cooled acetonitrile at 4 ℃ into the plasma sample, performing vortex for 30 seconds to obtain a mixture, performing centrifugation for 15 minutes at the temperature of 4 ℃ at the speed of 13,000 rpm, separating to obtain a first supernatant, transferring the first supernatant into a new centrifugal tube, and performing freeze drying to obtain a freeze-dried product; and adding acetonitrile/water composite solution with the volume ratio of 1:4 into the freeze-dried product, mixing by vortex for 30 seconds, centrifuging for 15min at the temperature of 4 ℃ at the speed of 13,000 rpm, and separating to obtain a second supernatant, namely the plasma pretreatment sample.

In some embodiments of the invention, in step (3), one QC sample is added every 8-10 LC-MS analysis samples. In one embodiment of the present invention, in step (2), one QC sample is added every 10 LC-MS analysis samples.

In some embodiments of the present invention, in step (4), the format conversion and the ion peak information extraction are as follows: using MSConvert software to convert the original metabolic fingerprints of each plasma sample into mzXML format files; extracting ion peak information in each plasma sample through an R language software package XCMS, and performing peak matching among different plasma samples through the ion peak information to obtain data containing all ion peak information, wherein the ion peak information comprises retention time and charge-to-mass ratio m/z of ions.

In some embodiments of the present invention, in step (6), n is 20.

In some embodiments of the present invention, in step (7), substituting 2 to n metabolic features according to the sequence of the differential metabolites for modeling for multiple times to obtain multiple classification models, which means: substituting the difference metabolite at the first 2 bits of the sequence to model 2 metabolic features, substituting the difference metabolite at the first 3 bits of the sequence to model 3 metabolic features, substituting the difference metabolite at the first 4 bits of the sequence to model 4 metabolic features … and so on, substituting the difference metabolites at the n bits of the sequence to model the n metabolic features, thereby obtaining a plurality of classification models containing different numbers of metabolic features.

In some embodiments of the invention, in step (7), the modeling is to use a machine learning algorithm to randomly forest learn data from a training set, wherein 10 repetitions and 5-fold cross validation are performed during training.

The invention also provides application of the blood marker for identifying malignant mesothelioma.

In some embodiments of the present invention, the blood marker for identifying malignant mesothelioma is used in the preparation of a diagnostic reagent for identifying malignant mesothelioma.

The invention is used for identifying the blood marker of malignant mesothelioma, which is a combination of the following nine metabolites: uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, and histidine, and the 9 plasma metabolic markers themselves are each found in malignant mesothelioma diagnosis for the first time, and the combination thereof is also proposed for the first time as a potential diagnosis marker of malignant mesothelioma.

In the invention, all plasma samples are randomly divided into a training set and a testing set, and the original metabolism fingerprint of each plasma sample is obtained by LC-MS detection and analysis; after format conversion, ion peak information extraction and partial ion filtration are carried out on the original metabolism fingerprint of all plasma samples, a preprocessing data matrix is obtained; modeling the preprocessed data matrix by using OPLS-DA, screening out characteristic difference ions meeting the requirements, and identifying to obtain n difference metabolites; in a training set data matrix, using a random forest algorithm to score the feature importance of the n different metabolites, sequencing the n different metabolites from high to low according to the feature importance score, and substituting 2-n metabolic features according to the sequencing of the different metabolites for modeling for multiple times to obtain a plurality of classification models; comparing the accuracy and the Kappa value of different classification models to evaluate the classification efficiency of different models and finally screening out the optimal classification model; and verifying the optimal classification model in a test set data matrix. The optimal classification model is a combination of nine metabolites, can be used as the blood marker for identifying malignant mesothelioma, has very important significance for diagnosis and treatment of malignant mesothelioma, and can be used as a diagnosis marker for constructing a diagnosis model and carrying out diagnosis application, for example, in preparation of a diagnosis reagent for identifying malignant mesothelioma.

Compared with the prior art, the invention has the following beneficial technical effects:

(1) the invention provides a blood marker for identifying malignant mesothelioma for the first time, which is a classification model containing nine metabolites: uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, and histidine, the classification model accuracy reaches 1.0. Moreover, the AUC values of this classification model were significantly higher in the test set than the AUC values of any single metabolite. The blood marker for identifying malignant mesothelioma has good sensitivity and specificity and good diagnostic capability.

(2) The invention combines non-targeted metabonomics and machine learning algorithm for the first time to research the differential metabolites between the plasma sample of a malignant mesothelioma patient and the healthy individual contrast matched with age and gender, thereby excavating and verifying an optimal classification model consisting of nine metabolites and realizing high-sensitivity and specific detection on the malignant mesothelioma.

(3) The blood marker for identifying malignant mesothelioma can be applied to the preparation of a diagnostic reagent for identifying malignant mesothelioma, is beneficial to the diagnosis of malignant mesothelioma, improves the screening efficiency, reduces the screening cost, provides a direction for individualized treatment, and improves the survival rate and the survival quality of patients.

Drawings

FIG. 1 is a schematic diagram of the process flow of the present invention.

Fig. 2 is a total ion chromatogram for all 5 mass control samples (QC).

FIG. 3 shows the discrimination of all samples by PCA modeling.

FIG. 4 is a classification of healthy group samples (HC) and malignant mesothelioma group samples (MM) based on OPLS-DA.

FIG. 5 shows a verification persistence test diagram of the OPLS-DA model.

Figure 6 shows a volcano based on total ions to show the distribution of the ions.

Figure 7 shows the results of feature importance scoring of 20 differential metabolites using a random forest algorithm.

Figure 8 shows the accuracy of different random forest models containing different numbers of metabolic features.

FIG. 9 shows Kappa values for different random forest models containing different numbers of metabolic features.

FIG. 10 is a ROC curve obtained by ROC analysis of the test set using the optimal random forest model containing 9 metabolic features.

FIG. 11 is the probability that an optimal random forest model containing 9 metabolic features predicts a sample in the test set as MM.

FIG. 12 is a ROC curve obtained by establishing a ROC model based on taurocholic acid in the training set.

Figure 13 shows the difference in the expression levels of taurocholic acid in MM and HC samples of the training set.

FIG. 14 is a ROC curve obtained by building a ROC model based on uracils in the training set.

Figure 15 shows the difference in the expression levels of uracil in MM and HC samples of the training set.

FIG. 16 is a ROC curve obtained by establishing a ROC model based on biliverdin in the training set.

Figure 17 shows the difference in the expression levels of biliverdin in MM and HC samples of the training set.

FIG. 18 is a ROC curve obtained by constructing a ROC model based on histidines in the training set.

Figure 19 shows the difference in expression levels of histidine in MM and HC samples of the training set.

FIG. 20 is a ROC curve obtained by establishing a ROC model based on taurodeoxycholic acid in the training set.

Figure 21 shows the difference in the expression levels of taurodeoxycholic acid in MM and HC samples of the training set.

FIG. 22 is a ROC curve obtained by establishing a ROC model based on pyrroline hydroxycarboxylic acids in the training set.

Figure 23 shows the difference in the expression levels of pyrroline hydroxy acid in MM and HC samples of the training set.

FIG. 24 is a ROC curve obtained by constructing a ROC model based on phenylalanine in the training set.

FIG. 25 shows the difference in the expression levels of phenylalanine in MM and HC samples of the training set.

Fig. 26 shows the prediction accuracy in the test set for the ROC model based on individual metabolites and for the ROC model based on a combination of nine metabolites (machine learning).

Detailed Description

Embodiments of the present invention will be described in detail below with reference to the accompanying drawings and specific examples. It should be understood that these examples are for illustrative purposes only and are not to be construed as limiting the scope of the present invention.

Study subjects: the study contained a total of 25 plasma samples of malignant mesothelioma patients from the tumor hospital in Zhejiang and 32 normal healthy control plasma samples.

The experimental apparatus comprises: high performance liquid chromatography-mass spectrometry (U3000/QOxctive, Thermo Fisher), high-speed low-temperature centrifuge (Thermo Fisher), vibration vortex apparatus, centrifugal concentrator, and refrigerator at 4 deg.C.

The experiment consumptive material includes: ACQUITY UPLC HSS T3 chromatographic column (specification: 2.1mm × 100mm, particle diameter: 1.8 μm), 1.5mL EP tube, 1.5mL sample injection bottle, 300 μ L internal cannula, pipette, 1000 μ L tip, 200 μ L tip, marker pen, latex glove, and mask.

The experimental reagent comprises: methanol (TEDIA, chromatographically pure), acetonitrile (TEDIA, chromatographically pure), acetone (TEDIA, chromatographically pure), purified water (wara).

FIG. 1 gives a schematic representation of the process flow of the present invention. The screening process for identifying blood markers for malignant mesothelioma will be described in detail below.

First, sample collection

Plasma samples from 25 malignant mesothelioma patients and 32 normal healthy control plasma samples from the tumor hospital, zhejiang province were collected: all plasma was collected from overnight fasted subjects and anticoagulated as follows:

to each plasma was added dipotassium salt of ethylenediaminetetraacetic acid (K2-EDTA) as an anticoagulant, and immediately centrifuged at 13,000 rpm at 4 ℃ for 15 minutes, and the supernatant was collected and 80. mu.L of the supernatant was used as a plasma sample. Plasma samples were presented at-80 ℃ until analysis.

All plasma samples were randomly set to 2 cohorts: training set and testing set. Wherein, 43 training sets comprise 19 malignant mesothelioma patients and 24 healthy controls; the test set comprised 14 patients with malignant mesothelioma 6 and healthy controls 8.

Preparation of quality control samples (QC samples):

from each of the plasma samples of malignant mesothelioma patients and the healthy plasma samples, 10 μ L of each of the plasma samples was taken and mixed, and then the mixed solution was divided into a plurality of portions on an average, one for each quality control plasma sample.

Second, plasma sample pretreatment

And (3) performing LC-MS analysis pretreatment on all the plasma samples of the malignant mesothelioma patients, the healthy plasma samples and the quality control samples one by one to respectively obtain plasma pretreatment samples of the malignant mesothelioma patients, healthy control plasma samples and quality control plasma pretreatment samples.

The LC-MS analysis pretreatment comprises the following steps:

a single plasma sample of 80 μ L was taken, 320 μ L of pre-chilled acetonitrile (acetonitrile pre-chilled at 4 ℃) was added thereto, and vortexed for 30 seconds to obtain a mixture; the mixture was centrifuged at 13,000 rpm for 15 minutes at 4 ℃ to separate 350. mu.L of supernatant; transferring the supernatant into a new centrifugal tube, and freeze-drying to obtain a freeze-dried product; adding 80 μ L of composite solution (acetonitrile/water, 1:4, v/v) into the lyophilized product, mixing by vortex for 30 s, centrifuging at 13,000 rpm for 15min at 4 deg.C, and separating to obtain 60 μ L of supernatant, which is the plasma pretreatment sample.

Three, liquid chromatography-mass spectrometry (LC-MS) detection and analysis

For all the plasma pretreatment samples of malignant mesothelioma patients, the healthy control plasma samples and the quality control plasma pretreatment samples, 5 mu L of each sample is taken and transferred into an injection bottle, and the samples are sequentially sequenced, injected one by one, and subjected to LC-MS detection and analysis. One QC sample was added every 10 LC-MS analysis samples.

The LC-MS analysis sample is a general term of a plasma pretreatment sample of a malignant mesothelioma patient and a healthy control plasma sample, and the QC sample is a quality control plasma pretreatment sample.

The liquid chromatography and mass spectrometry methods used were as follows:

the mobile phase is as follows: 0.1% formic acid water (phase a) and acetonitrile (phase B); wherein, the formic acid water comprises 0.1 percent of formic acid and 99.9 percent of purified water by volume percentage;

flow rate: 0.3 mL/min; column temperature: 35 ℃;

chromatographic gradient elution conditions: 0-1 min, 2% phase B; 1-10 minutes, 2-100% phase B; 10-13 min, 100% phase B; 13-13.1 minutes, and 100-2 percent of phase B; 13.1-16 min, phase B2%.

The mass spectrometry method comprises the following steps: capillary voltages for the positive electrospray ionization (ESI +) and negative electrospray ionization (ESI-) modes were 3500v and 2500v, respectively. The ion transfer tube temperature was set to 350 ℃, the mass scan range (m/z) was set to 70-1000, and the mass resolution in both ESI + and ESI-modes was 70000. In ESI + and ESI-modes, the sheath gas is set to 35 and 40arb, respectively. For MS/MS chromatography acquisition, the data-dependent acquisition mode was performed in the first 10 modes with mass resolution 17500, step collision energies 10ev, 20ev, and 40 ev.

And (3) analyzing each plasma pretreatment sample of the patient with malignant mesothelioma and the healthy plasma pretreatment sample according to the chromatographic mass spectrometry conditions to obtain the original metabolic fingerprint of each LC-MS analysis sample.

And analyzing each QC sample according to the chromatographic mass spectrometry conditions to obtain the original metabolic fingerprint of all 5 QC samples. The total ion chromatogram of 5 QC samples is shown in FIG. 2. In FIG. 2, the distribution and shape of the ion peaks of 5 QC samples are consistent, which indicates that the LC-MS instrument and the analysis method are stable and reliable.

Fourth, map data preprocessing

4.1 using MSConvert software to convert the original metabolic fingerprint of each sample into an mzXML format file; each sample comprises an LC-MS analysis sample and a QC sample;

4.2 extracting ion peak information (retention time and charge-to-mass ratio m/z) of each sample through R packet XCMS, and then comparing and matching (peak matching) through the ion peak information among different samples to obtain data containing all ion peak information, wherein the ion peak information comprises the retention time and the charge-to-mass ratio m/z of ions;

4.3, further processing the data by using an R packet MetaX (v1.4.16), filtering and removing data of ions meeting any one of the following standards (a) to (c), and obtaining a preprocessed data matrix, wherein the preprocessed data matrix comprises a training set data matrix and a test set data matrix:

(a) ions not detected in more than 50% of QC samples;

(b) ions not detected in more than 80% of the LC-MS analysis samples (i.e. non-QC samples);

(c) ions with relative standard deviation > 30% in QC samples;

relative Standard Deviation (RSD), also called Coefficient of Variation (CV), is the Standard Deviation/average value 100%.

Fifthly, screening and identifying differential metabolites

5.1 Principal Component Analysis (PCA) of the preprocessed data of all samples using R-packs, i.e. PCA modeling under unsupervised conditions (i.e. without grouping information) with all ions retained after preprocessing, the PCA model score plot is shown in FIG. 3.

Fig. 3 gives the discrimination of the principal component analysis method for the classification of QC samples (QC), healthy group samples (HC) and malignant mesothelioma group samples (MM) without giving grouping information:

in FIG. 3, the consistency of all QC samples is good, which indicates that the LC-MS instrument is stable and the analysis result is reliable, and the experiment quality control is good; meanwhile, the healthy group sample (HC) and the malignant mesothelioma group sample (MM) do not have a good classification trend in fig. 3, and the samples overlap more, which indicates that a supervised learning method is required to realize further classification.

5.2 Pre-processed data matrices for healthy group samples (HC) and malignant mesothelioma group samples (MM) were modeled using supervised analysis methods, orthogonal partial least squares discriminant analysis (OPLS-DA).

FIG. 4 shows the classification of OPLS-DA for healthy group samples (HC) and malignant mesothelioma group samples (MM). In FIG. 4, the healthy group sample (HC) and the malignant mesothelioma group sample (MM) are well distinguished.

FIG. 5 shows a verification persistence test diagram of the OPLS-DA model, wherein the abscissa represents the similarity of the model. R2Y represents the interpretation rate of the model on the Y matrix, and is used for evaluating the goodness of fit, and Q2 is used for evaluating the predictability of the model and indicating the prediction capability of the model. In fig. 5, the slope of the straight line is large and the intercept of Q2 is X, indicating that the OPLS-DA model is not overfit, and also showing the effectiveness of the OPLS-DA model, since the displaced R2 and Q2 are significantly lower than the original values and the two lines R2Y and Q2 have a significant tendency to intersect.

Meanwhile, Variable Projection Importance (VIP) of each ion is calculated in an OPLS-DA model, and then the influence strength and the interpretation capability of the expression mode of each metabolite on the classification discrimination of each group of samples are measured by combining the difference multiple and the difference significance of the relative concentration of each ion in plasma of a malignant mesothelioma patient and the plasma of a healthy control patient, so that the screening of the marker metabolite is assisted.

The volcano plot of fig. 6 shows the change in characteristic ions in the malignant mesothelioma group sample (MM) versus the healthy group sample (HC). In fig. 6, the abscissa represents the direction of change and the fold difference of the metabolic ions, and the ordinate represents the significance of the difference in the statistical sense of the metabolic ions (P value in t test). Dots located above the dotted line intersecting the vertical axis represent ions that are significantly different (P <0.05), dots located to the left of the dotted line intersecting the horizontal axis represent ions that are significantly down-regulated (down) (fold difference <0.667) in the malignant mesothelioma group sample (MM), dots located to the right of the dotted line intersecting the horizontal axis represent ions that are significantly up-regulated (up) (fold difference >1.5) in the malignant mesothelioma group sample (MM), and light circles represent ions that are not significantly different (no sig.).

5.3 screening for differential ions which also satisfy the following criteria (i) to (iii):

(ii) a variable projection importance (VIP) score of >1 for the ion in the OPLS-DA model;

(ii) ions having a P value <0.05 in the t-test;

(iii) the relative concentration of ions in the plasma of a patient with malignant mesothelioma differs by a factor >1.5 or <0.667 from that of healthy control plasma;

the fold difference of the relative concentration of ions in the plasma of a malignant mesothelioma patient and the plasma of a healthy control, also called the fold difference of the relative concentration of ions, refers to the fold value of the relative concentration of ions in the plasma of a malignant mesothelioma patient divided by the relative concentration of ions in the plasma of a healthy individual.

And 5.4, matching an MS/MS spectrogram corresponding to the m/z value and retention time of each differential ion obtained by screening in the step 5.3 with a reference spectrogram of a network Database Metlin and a Human Metamolome Database (HMDB), identifying the reference spectrogram, and finally identifying to obtain 20 differential metabolites, wherein the details are shown in a table 1.

TABLE 1 differential metabolite List

^a"Pos" and "Neg" denote the positive ion scan mode and the negative ion scan mode of the mass spectrum, respectively.

^bThe charge-to-mass ratio.

^cProjection importance of variables calculated based on OPLS-DA.

^dFold difference (mesothelioma plasma/healthy control plasma).

Sixthly, feature screening and establishment of multiple metabolite modeling

In a training set data matrix, a random forest algorithm is used for carrying out feature importance (variable importance) grading on 20 differential metabolites, the 20 differential metabolites are ranked from high to low according to the feature importance grading, 2-n metabolic features are substituted according to the ranking of the differential metabolites for modeling for multiple times, machine learning algorithm Random Forest (RF) is used for learning data from a training set, 10 times of repetition and 5 times of cross validation are carried out during training, the accuracy and Kappa value of the established different classification models are compared, the classification efficiency of the different models is evaluated, and finally the optimal classification model with the optimal feature number and the combination mode on the cross validation set is screened out.

The feature importance scoring is carried out on the 20 differential metabolites, modeling is carried out on all the 20 differential metabolites by combining a random forest algorithm, a model with the highest prediction accuracy is obtained by adjusting parameters, and the contribution rate of each variable to the model in the operation process is ranked from high to low. The ranking results of the feature importance scores of the 20 differential metabolites are shown in fig. 7.

Substituting 2-n metabolic features according to the sequence of the differential metabolites for modeling for many times, which means that: substituting the difference metabolite at the first 2 bits of the sequence for modeling 2 metabolic features, substituting the difference metabolite at the first 3 bits of the sequence for modeling 3 metabolic features, substituting the difference metabolite at the first 4 bits of the sequence for modeling 4 metabolic features … and so on, substituting the difference metabolites at the n bits of the sequence for modeling the n metabolic features, thereby obtaining a plurality of random forest classification models containing different numbers of metabolic features.

Random forest models containing different numbers of metabolic features are illustrated below: in the random forest model containing 2 metabolic features, 2 metabolic features are respectively the 1 st and 2 nd metabolic features of feature importance ordering, and uracil and pyrroline hydroxycarboxylic acid correspond to each other in fig. 7; in the random forest model containing 3 metabolic features, 3 metabolic features are respectively the metabolic features with feature importance ranking 1 st to 3 rd, and uracil, pyrroline hydroxycarboxylic acid and phenylalanine correspond to fig. 7; by analogy, 20 metabolic features are included, the 1 st to 20 th metabolic features are ranked for feature importance respectively, and all 20 metabolites correspond in fig. 7.

The accuracy of the random forest classification models containing different numbers of metabolic features is shown in fig. 8, and the criteria for screening the optimal feature numbers and the combination mode are as follows: the model accuracy does not increase any more with increasing feature numbers. In fig. 8, the abscissa is the number of metabolic features in the model and the ordinate is the accuracy of the random forest model; for example, the abscissa is 2, and the ordinate corresponds to the accuracy of a random forest model containing 2 metabolic features; similarly, the abscissa is 3, and the ordinate corresponds to the accuracy of a random forest model containing 3 metabolic features; by analogy, the abscissa is 20, and the ordinate corresponds to the accuracy of a random forest model containing 20 metabolic features. As can be seen from fig. 8, the model accuracy reached 1.0 for the

numbers

9, 10, 11, 12, 15.

The Kappa values of the optimal models containing different amounts of metabolic features are shown in fig. 9, and the criteria for screening the optimal feature numbers and the combination method are as follows: the model Kappa value does not rise any more when the feature number is increased. In fig. 9, the abscissa is the number of metabolic features in the model and the ordinate is the Kappa value of the random forest model; for example, the abscissa is 2 and the ordinate corresponds to the Kappa number of a model containing 2 metabolic features. Similarly, the abscissa is 3, and the ordinate corresponds to the Kappa value of the random forest model containing 3 metabolic features; by analogy, the abscissa is 20, and the ordinate corresponds to the Kappa value of a random forest model containing 20 metabolic features. Also, as can be seen from FIG. 9, the Kappa number of the model reached 1.0 at the

numbers

9, 10, 11, 12, 15.

Therefore, we selected a random forest model containing 9 metabolic features as the optimal classification model, where 9 metabolites are: uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, histidine.

Verification of seven, optimal classification model

For this optimal classification model containing 9 metabolic features, ROC (Receiver Operating Characteristic curve) analysis was performed in the test set, and the ROC curve is shown in fig. 10. According to calculation, the AUC value is 1.0000, namely, the prediction accuracy of the optimal model containing 9 metabolic characteristics is as high as 1.0000.

The probability that this optimal classification model containing 9 metabolic features predicted Malignant Mesothelioma (MM) for 14 samples of the test set is shown in fig. 11. In fig. 11, the probability of the model mispredicting the actual healthy group sample (HC) as Malignant Mesothelioma (MM) is very low, whereas the probability of accurately predicting the malignant mesothelioma group sample (MM) is very high (P <0.0001), which means that the optimal classification model can well distinguish malignant mesothelioma group sample (MM) from healthy group sample (HC).

It can be seen that, according to the above-described identification method, we succeeded in identifying a combination of 9 plasma metabolic markers (uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, histidine) as a diagnostic marker suitable for malignant mesothelioma. Not only is the group of the 9 plasma metabolism markers put forward for the first time as a potential diagnosis marker of malignant mesothelioma, but also the 9 plasma metabolism markers are found in the diagnosis of malignant mesothelioma for the first time, so that the invention has very important significance for the diagnosis and treatment of malignant mesothelioma, and can be used as a diagnosis marker for constructing a diagnosis model and carrying out diagnosis application.

Comparative example 1:

AUC values were obtained by packing rcocs with R and building an ROC model based on the individual differential metabolites in the training set, and used to assess the ability of each metabolite to distinguish cancer from healthy samples.

Among the ROC curves for the 20 different metabolites, we selected the ROC curves for the first 7 most discriminatory metabolites, as shown in FIG. 12, FIG. 14, FIG. 16, FIG. 18, FIG. 20, FIG. 22, FIG. 24, and the corresponding AUC_ROCComprises the following steps: taurocholic acid (AUC ═ 0.8421); uracil (AUC ═ 0.8399); biliverdin (AUC 0.8289); histidine (AUC 0.8180); taurodeoxycholic acid (AUC ═ 0.8048); pyrroline hydroxy acid (AUC ═ 0.8026); phenylalanine (AUC ═ 0.8004).

Meanwhile, we analyzed the expression levels of these 7 metabolites (taurocholic acid, uracil, biliverdin, histidine, taurodeoxycholic acid, pyrroline hydroxycarboxylic acid, phenylalanine) in the 43 samples of the training set, as shown in fig. 13, fig. 15, fig. 17, fig. 19, fig. 21, fig. 23, fig. 25, respectively. According to the above-mentioned graphs, taurocholic acid (fig. 13), taurodeoxycholic acid (fig. 21), pyrroline hydroxyhydrocarbon acid (fig. 23) were significantly up-regulated in the malignant mesothelioma group sample (MM), uracil (fig. 15), biliverdin (fig. 17), histidine (fig. 19), phenylalanine (fig. 25) were significantly down-regulated in the malignant mesothelioma group sample (MM) compared to the healthy group sample (HC).

In addition, ROC analysis based on test set data was also performed using R package pROC to evaluate different predictive performance of different individual metabolite models and an optimal random forest model based on 9 metabolites. Fig. 26 shows that the ROC model based on a single metabolite and the optimal random forest model based on 9 metabolites were applied to test set data to obtain different prediction accuracy rates. Wherein the accuracy of the first 3 single metabolites (histidine, pyrroline hydroxy acid, phenylalanine) is 0.8571, 0.8571, 0.7857 respectively, which indicates that the prediction level of each single metabolite is not high, and the single metabolite is not suitable to be used as a biomarker; whereas the accuracy of the optimal random forest model (corresponding to machine learning) of 9 metabolites is 1.0000, which is significantly higher than the accuracy of any one single metabolite.

In summary, these metabolites with significant differences between the malignant mesothelioma group sample (MM) and the healthy group sample (HC) were not high in predicted levels of single metabolites and were not suitable for use as biomarkers.

Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and not for limiting the protection scope of the present invention, and although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A blood marker for identifying malignant mesothelioma, which is a combination of nine metabolites: uracil, pyrroline hydroxycarboxylic acid, phenylalanine, biliverdin, taurocholic acid, bilirubin, kynurenine, uridine, and histidine.

2. The screening method for identifying a blood marker of malignant mesothelioma according to claim 1, comprising the steps of:

(2) performing LC-MS analysis pretreatment on each plasma sample of the patient with malignant mesothelioma, each healthy control plasma sample and each quality control plasma sample to respectively obtain a plasma pretreatment sample of the patient with malignant mesothelioma, a healthy control plasma sample and a quality control plasma pretreatment sample; the plasma pretreatment sample of the malignant mesothelioma patient and the healthy control plasma sample are collectively called LC-MS analysis sample, and the quality control plasma pretreatment sample is called QC sample;

(a) ions not detected in more than 50% of QC samples;

(b) (ii) no ions detected in more than 80% of the LC-MS analysis samples;

(c) ions with relative standard deviation > 30% in QC samples;

(i) a variable projection importance score >1 in the OPLS-DA model;

(ii) P value <0.05 in t-test;

(8) and verifying the optimal classification model in a test set data matrix.

3. The screening method according to claim 2, wherein in the step (1), the ratio of the number of samples in the training set to the number of samples in the test set is 2:1 to 4: 1.

4. The screening method according to claim 2, wherein the LC-MS pretreatment before analysis in step (2) comprises the steps of: adding 4 ℃ precooled acetonitrile into the plasma sample, vortexing for 30 seconds to obtain a mixture, centrifuging for 15 minutes at 4 ℃ at the speed of 13,000 rpm, separating to obtain a first supernatant, transferring the first supernatant into a new centrifuge tube, and freeze-drying to obtain a freeze-dried product; and adding acetonitrile/water composite solution with the volume ratio of 1:4 into the freeze-dried product, mixing by vortex for 30 seconds, centrifuging for 15min at the temperature of 4 ℃ at the speed of 13,000 rpm, and separating to obtain a second supernatant, namely the plasma pretreatment sample.

5. The screening method according to claim 2, wherein in step (3), one QC sample is added every 8 to 10 LC-MS analysis samples.

6. The screening method according to claim 2, wherein in the step (4), the format conversion and the ion peak information extraction are performed as follows: using MSConvert software to convert the original metabolic fingerprint of each plasma sample into an mzXML format file; extracting ion peak information in each plasma sample through an R language software package XCMS, and performing peak matching among different plasma samples through the ion peak information to obtain data containing all ion peak information, wherein the ion peak information comprises retention time and charge-to-mass ratio of ions.

7. The screening method according to claim 2, wherein in the step (7), the step of performing modeling for a plurality of times by substituting 2 to n metabolic features according to the sequence of the differential metabolites to obtain a plurality of classification models is characterized in that: substituting the difference metabolite at the first 2 bits of the sequence to model 2 metabolic features, substituting the difference metabolite at the first 3 bits of the sequence to model 3 metabolic features, substituting the difference metabolite at the first 4 bits of the sequence to model 4 metabolic features … and so on, substituting the difference metabolites at the n bits of the sequence to model the n metabolic features, thereby obtaining a plurality of classification models containing different numbers of metabolic features.

8. The screening method of claim 2, wherein in the step (7), the modeling is performed by learning data from a training set by using a machine learning algorithm random forest, and 10 repetitions and 5-fold cross validation are performed during training.

9. Use of the blood marker for identifying malignant mesothelioma of claim 1.

10. The use of the blood marker for identifying malignant mesothelioma of claim 1 in the preparation of a diagnostic reagent for identifying malignant mesothelioma.