CN115954046A

CN115954046A - Stomach cancer personalized treatment decision-making method and system and storage medium containing stomach cancer personalized treatment decision-making system

Info

Publication number: CN115954046A
Application number: CN202211542941.9A
Authority: CN
Inventors: 李明珠
Original assignee: Shanghai Aipu Tikang Biotechnology Co ltd
Current assignee: Shanghai Aipu Tikang Biotechnology Co ltd
Priority date: 2022-12-02
Filing date: 2022-12-02
Publication date: 2023-04-11

Abstract

The invention discloses a stomach cancer individualized treatment decision-making method, a stomach cancer individualized treatment decision-making system and a storage medium containing the stomach cancer individualized treatment decision-making system. The decision method for individualized treatment of gastric cancer respectively obtains prediction models of different treatment schemes of treated patients, performs prediction by combining the models according to sample mass spectrum detection results provided by the patients to be treated, and recommends the treatment scheme corresponding to the group to the patients to be treated. The invention constructs a prediction model for predicting the treatment response of the gastric cancer based on the proteome panel typing and a machine algorithm, can recommend an individualized treatment scheme suitable for the patient to be treated, and realizes the individualized treatment to a certain extent.

Description

Stomach cancer personalized treatment decision-making method and system and storage medium containing stomach cancer personalized treatment decision-making system

Technical Field

The invention belongs to the field of personalized treatment of human diseases, and relates to a prediction model for personalized treatment decision of gastric cancer, a personalized treatment decision method and a system thereof, a computer readable storage medium containing the prediction model, the personalized treatment decision method and the system, which are constructed by a machine learning algorithm based on proteome panel typing.

Background

Gastric Cancer (GC) is the fourth most common malignancy worldwide and also the second leading cause of cancer-related death worldwide. Low early detection rate and low late cure rate of gastric cancer. The preferred method of treatment for patients with advanced gastric cancer is surgery. For some patients without surgical treatment opportunities, the quality of life can be improved or even the survival can be maintained only by the medicine. Despite significant advances in recent years in the iterative updating of first-line therapeutic drugs, second-line therapeutic drugs, and the intensive development of the targeted drugs trastuzumab (Herceptin ) and Ramucirumab (Ramucirumab; cyramxa), the problem of drug tolerance has not been solved, and the overall prognosis of gastric cancer treatment is still at a poor level. Clinically, the treatment effect of the gastric cancer has great individual difference, but the personalized selection basis of the medicine is too few, and the clinical urgent need is to guide the markers of personalized precise medical treatment, so that the problem of medicine tolerance is relieved.

The development of genomics and transcriptomics technology promotes the research and large-scale discovery of mutation modes of related genes and important driving genes for tumorigenesis, and promotes the generation and development of precise medicine. However, the clinical needs are not satisfied by merely studying precise diagnosis and treatment of tumors at the gene level. Proteomics research explains the reasons of development of specific biological phenomena from the protein level, reveals the development rule, and has great significance for the development of life science research and precise medical diagnosis and treatment.

Most of the current research is reported to identify the characteristic genes and signal networks of tumors by integrating genomic data sets, and genome-based molecular classification systems have been proposed for different human populations. In 2014, the cancer genome map (TCGA) project mapped the genome of 295 gastric cancer patients and divided the gastric cancer into four subtypes: (1) Epstein-Barr virus positive (EBV positive) gastric cancer; (2) microsatellite high instability (MSI-H) gastric cancer; (3) chromosome Instability (CIN) gastric cancer; (4) gene Stability (GS) gastric cancer. A study conducted in 2015 by the Asian Cancer Research Group (ACRG) in succession divided gastric cancer into four subtypes based on gene expression data, each associated with a different clinical outcome. In the past few years, many similar studies have been conducted by large-scale cancer genome sequencing, and it can be seen that gastric cancer is divided into molecularly-genetically distinct heterogeneous subgroups, rather than a single type of cancer. Therefore, in order to realize individualized treatment of gastric cancer, it is necessary to identify subclasses according to molecular genetic and pathological characteristics, and to find and apply corresponding target genes. In addition, in the research of gastric cancer, results that can classify the prognosis of gastric cancer according to its subtype have been reported. Many research patents that have been introduced at present are systems for classifying gastric cancer based on gene expression levels of genome and transcriptome, such as systems for predicting postoperative prognosis or suitability of anticancer drugs for patients with advanced gastric cancer (patent No. CN110168106 a), and clustering classification and prognosis prediction systems based on biological characteristics of gastric cancer (patent No. CN110177886 a). However, in the clinic, first-line treatment regimens for gastric cancer patients, such as XELOX regimens, DOS regimens and herceptin (HER 2) based targeted treatment regimens, have not yet achieved an effective treatment regimen choice.

As a "life performer," proteins determine phenotype and may bridge the gap between research and clinical practice. At present, molecular typing based on diffuse gastric cancer, protein markers for typing and a screening method and application thereof (patent number CN 108445097A) are realized on the protein level, and the strong relevance between the protein-based molecules and the survival and treatment response of patients is suggested. However, no personalized treatment regimen for gastric cancer has been achieved at present.

Disclosure of Invention

In order to overcome the defect that a prediction model for gastric cancer treatment is lacked in the prior art, so that a computer-aided personalized treatment decision scheme is effectively provided, the invention provides a personalized treatment decision method and system for gastric cancer and a storage medium containing the same.

The invention utilizes the proteomics technology combined with a machine learning algorithm to construct a prediction model aiming at different treatment reactions of the gastric cancer and the intestinal cancer according to the protein expression data of treated patients, and the prediction model covers the current first-line treatment scheme of the gastric cancer, such as DOS scheme, XELOX scheme and HER2 targeted treatment scheme of the gastric cancer, thereby being beneficial to the clinician to establish a personalized medicine treatment scheme according to the coincidence degree of the protein expression data of untreated patients and the prediction model, leading the patients to obtain the maximum benefit and laying a foundation for clinical medicine decision.

In order to solve the above technical problems, one of the technical solutions of the present invention is: a method for constructing a gastric cancer treatment prediction model is provided, which comprises the following steps:

(1) Dividing sample groups: obtaining clinical samples of gastric cancer patients treated by different treatment schemes, and dividing the patients into sensitive groups and non-sensitive groups according to treatment results; performing mass spectrum detection on the clinical sample to obtain protein expression data, screening differential expression protein DEP (deoxyribose nucleic acid) of which the protein expression amount of the sensitive group relative to the non-sensitive group is more than 2 times or the protein expression amount of the non-sensitive group relative to the sensitive group is more than 2 times and p is less than 0.05, and inputting the differential expression protein DEP into a matrix; the treatment regimens include DOS, XELOX, and HER2 regimens, and the samples are divided into DOS sensitive, DOS insensitive, XELOX sensitive, XELOX insensitive, HER2 sensitive, and HER2 insensitive groups;

(2) Selecting characteristic proteins: calculating an erythroid information criterion (AIC) of the DEP by adopting a multivariate binary classification stepwise logistic regression method for the DEP in the step (1) to obtain a group of DEP with the minimum AIC value as characteristic proteins of the treatment schemes DOS, XELOX and HER 2;

(3) Constructing a prediction model: randomly extracting at least 50% of clinical samples in each treatment scheme, taking the corresponding characteristic protein as a training set, taking the characteristic protein corresponding to the rest clinical samples as a test set, and performing cross validation on the training set by adopting multivariate binary classification stepwise logistic regression; thus obtaining a prediction model of DOS, XELOX and HER2 treatment schemes.

In some preferred embodiments:

in the step (2), R studio software is used for selecting the AIC; adopting a pROC data packet, and setting family = binary in a glm function and direction = 'backward' in a step function; and/or the presence of a gas in the atmosphere,

in the step (3), the cross validation is 10 times cross validation, and the cross validation is repeated for 10 times; and/or the presence of a gas in the gas,

providing clinical treatment response information including treatment efficacy assessed using normal X-ray, CT scan, and/or MRI scan; and/or the presence of a gas in the gas,

the patients were grouped based on RECIST criteria, wherein the sensitive group included complete remission and partial remission and the non-sensitive group included stable disease and progression of disease.

In some more preferred embodiments, the obtaining of the protein expression data comprises the steps of:

(A) Sample preparation: preparing the sample into a formalin-fixed paraffin-embedded tissue section, and/or making gastric cancer cells of the sample account for more than 80% of the total number of cells; and/or, the sample uses a lysis buffer of 0.1M Tris-HCL, pH 8.0, and 0.1M DTT and 1mM PMSF are added; and/or, the protein of the sample is treated with 50mM NH containing trypsin ₄ HCO ₃ Incubating at 37 deg.C for 18-20 hr for enzymolysis; and/or, centrifugally collecting to obtain an enzymolyzed peptide fragment; and/or the presence of a gas in the gas,

(B) Mass spectrum detection: detecting the peptide segment by using a Q-active HF-X mixed quadrupole orbitrap mass spectrometer and a high performance liquid chromatography system, and obtaining mass spectrum data corresponding to the peptide segment; and/or, controlling data acquisition by using Xcalilibur software; and/or the presence of a gas in the gas,

(C) Data processing: the acquired data is processed by using a Firmiana database and MaxQuant software; and/or, the first search mass tolerance is 20ppm, the main search peptide tolerance is 0.5da; and/or, the calculated method is label-free iBAQ and FOT is used to represent the normalized abundance of protein in the sample; and/or selecting a protein having at least one specific peptide stretch and an FDR of less than 1%.

In order to solve the above technical problems, the second technical solution of the present invention is: provided is a decision-making method for personalized gastric cancer treatment, which comprises the following steps:

(a) Obtaining prediction models of different treatment schemes obtained by the construction method of the gastric cancer treatment prediction model according to one of the technical schemes of the invention;

(b) Extracting protein expression data of clinical samples of patients to be treated, combining prediction models of different treatment schemes, calculating prediction probability sensitive to each treatment scheme by using a prediction function of R studio software and parameter setting type = 'prob', and recommending the treatment scheme with the maximum prediction probability to the patients.

In order to solve the technical problems, the third technical scheme of the invention is as follows: provided is a device for constructing a gastric cancer treatment prediction model, which comprises:

a sample grouping module: the sample grouping module is used for obtaining clinical samples of patients treated by different treatment schemes and dividing the patients into sensitive groups and non-sensitive groups according to treatment results; performing mass spectrum detection on the clinical sample to obtain protein expression data, screening differential expression protein DEP (deoxyribose nucleic acid) of which the protein expression amount of the sensitive group relative to the non-sensitive group is more than 2 times or the protein expression amount of the non-sensitive group relative to the sensitive group is more than 2 times and p is less than 0.05, and inputting the differential expression protein DEP into a matrix; the treatment regimens include DOS, XELOX, and HER2 regimens, and the samples are divided into DOS sensitive, DOS insensitive, XELOX sensitive, XELOX insensitive, HER2 sensitive, and HER2 insensitive groups; and (c) and (d),

a characteristic protein selection module: the characteristic protein selection module calculates an erythroid information criterion (AIC) of the DEP by adopting a multivariate binary classification stepwise logistic regression method for the DEP in the step (1) to obtain a group of DEP with the minimum AIC value as characteristic proteins of DOS, XELOX and HER2 treatment schemes; and the combination of (a) and (b),

a prediction model construction module: the prediction model construction module randomly extracts at least 50% of clinical samples in each treatment scheme, takes the corresponding characteristic proteins as a training set and the characteristic proteins corresponding to the rest clinical samples as a test set, and performs cross validation on the training set by adopting multivariate binary classification stepwise logistic regression; thus obtaining a prediction model of DOS, XELOX and HER2 treatment schemes.

In some of the more preferred embodiments of the present invention,

in the characteristic protein selection module, R studio software is used for calculating the AIC value; adopting a pROC data packet, and setting family = binary in a glm function and direction = 'backward' in a step function; and/or the presence of a gas in the gas,

in the prediction model building module, the cross validation is 10 times of cross validation, and the cross validation is repeated for 10 times.

In some further more preferred embodiments, the construction apparatus further comprises:

a sample preparation module: the sample preparation module is used for preparing a sample into a formalin-fixed paraffin-embedded tissue section, wherein gastric cancer cells of the sample account for more than 80% of the total number of cells; the lysis buffer used for lysis of the samples was 0.1M Tris-HCl, pH 8.0, 0.1M DTT and 1mM PMSF were added; 50mM NH with trypsin ₄ HCO ₃ Carrying out enzymolysis on the protein of the sample, incubating for 18-20 hours at 37 ℃, centrifuging and collecting to obtain an enzymolysis peptide fragment; and/or the presence of a gas in the gas,

a mass spectrometric detection module: the peptide segment of the mass spectrum detection module is detected by a Q-active HF-X mixed quadrupole orbitrap mass spectrometer and a high performance liquid chromatography system, and mass spectrum data corresponding to the peptide segment is obtained; using Xcalibur software to control data acquisition; and/or the presence of a gas in the gas,

a data processing module: the data acquired by the data processing module is processed by using a Firmiana database and MaxQuant software; the first search mass tolerance was 20ppm and the main search peptide tolerance was 0.5da; the calculation method is label-free iBAQ, and FOT is used for expressing the standardized abundance of protein in the sample; selecting a protein having at least one specific peptide stretch and an FDR of less than 1%.

In order to solve the above technical problems, the fourth technical solution of the present invention is: a personalized gastric cancer treatment decision system is provided, comprising:

the device for constructing the gastric cancer treatment prediction model according to the third technical scheme of the invention; and

a scheme decision module: the protocol decision module extracts protein expression data (obtained by mass spectrometric detection, for example) of a clinical sample of a patient to be treated, calculates prediction probabilities sensitive to each treatment protocol by combining prediction models of different treatment protocols and using a prediction function of R studio software, parameter setting type = 'prob', and recommends the treatment protocol with the highest prediction probability to the patient.

In order to solve the above technical problems, the fifth technical solution of the present invention is: an electronic device is provided that includes a memory and a processor; the memory includes a computer program stored therein that is executable on the processor; wherein the content of the first and second substances,

the processor implements the method for constructing a gastric cancer treatment prediction model according to one aspect of the present invention and the method for determining a personalized gastric cancer treatment according to the second aspect of the present invention when the computer program is executed.

In order to solve the technical problems, the sixth technical scheme of the invention is as follows: there is provided a computer readable storage medium having a computer program stored thereon, wherein the computer program, when being executed by a processor, is capable of implementing the steps of the method for constructing a gastric cancer treatment prediction model according to one aspect of the present invention and the steps of the method for deciding a personalized gastric cancer treatment according to the second aspect of the present invention.

According to the construction method of the gastric cancer treatment prediction model, the combination of each characteristic biomarker in the patient sample corresponding to each treatment scheme is obtained. When the biomarker combination is detected in a sample from a patient to be treated, one skilled in the art can determine which treatment regimen the patient to be treated is appropriate for.

On the basis of the common knowledge in the field, the above preferred conditions can be combined randomly to obtain the preferred embodiments of the invention.

The reagents, materials, equipment and medicaments used in the invention are all commercially available.

The positive progress effects of the invention are as follows:

the invention provides a personalized treatment decision method for diseases and a decision system for executing the decision method, wherein the decision system comprises: the device for constructing the gastric cancer treatment prediction model, the model comparison module and the scheme decision module construct the prediction model of the treatment response by combining a machine algorithm, can recommend an individualized treatment scheme suitable for the patient to be treated, and realize individualized treatment to a certain extent.

Drawings

FIG. 1 is a flow chart of a method for personalized gastric cancer treatment decision making according to the present invention;

FIG. 2 is a block diagram of a personalized gastric cancer treatment decision system;

fig. 3 is a schematic structural diagram of an electronic device according to embodiment 5 of the present invention;

FIG. 4 is a diagram of DOS subcohort DSG and DNSG differential protein volcanoes;

FIG. 5 is a volcano plot of the XSG and XNSG differential proteins of the XELOX subcohort;

figure 6 is a volcano plot of HSG and HNSG difference proteins of HER2 sub-cohort;

FIG. 7 shows a process of constructing a prediction model for gastric cancer treatment;

FIG. 8 is a flowchart for modeling and recommending personalized gastric cancer treatment protocols based on proteomic panel analysis;

FIG. 9 is a DOS subqueue prediction model construction;

FIG. 10 is a prediction model construction of XELOX subcohorts;

figure 11 is a prediction model construction for HER2 subcohort;

FIG. 12 is a diagram of a prediction evaluation for a new inbound DOS queue;

FIG. 13 is a predictive evaluation of the new entry group XELOX cohort;

FIG. 14 is a predictive assessment of the Beijing proteomics research center XELOX cohort.

Detailed Description

The invention is further illustrated by the following examples, which are not intended to limit the scope of the invention. The experimental methods without specifying specific conditions in the following examples were selected according to the conventional methods and conditions, or according to the commercial instructions.

Example 1 method for establishing a predictive model of gastric cancer treatment regimen based on proteome data

The method 201 for constructing a gastric cancer treatment prediction model of the present invention comprises the following steps (as shown in fig. 1):

step 101, dividing sample groups: obtaining clinical samples of patients treated by different treatment schemes, and dividing the patients into sensitive groups and non-sensitive groups according to treatment results; performing mass spectrum detection on the clinical sample to obtain protein expression data, screening differential expression protein DEP (deoxyribose nucleic acid) of which the protein expression amount of the sensitive group relative to the non-sensitive group is more than 2 times or the protein expression amount of the non-sensitive group relative to the sensitive group is more than 2 times and p is less than 0.05, and inputting the differential expression protein DEP into a matrix;

102, selecting characteristic proteins: calculating an erythroid information criterion (AIC) of the DEP by adopting a multivariate binary classification stepwise logistic regression method for the DEP in the step 101 to obtain a group of DEP with the minimum AIC value as a characteristic protein of the treatment scheme;

step 103, constructing a prediction model: randomly extracting at least 50% of clinical samples in each treatment scheme, taking the corresponding characteristic protein obtained in the step 102 as a training set, taking the characteristic protein corresponding to the rest clinical samples as a test set, and performing cross validation on the training set by adopting multivariate two-classification stepwise logistic regression; obtaining the prediction model of the treatment scheme.

In this embodiment, the gastric cancer is taken as an example, and the step 101 of obtaining the protein expression data comprises the following steps:

1. gastric cancer clinical specimen collection

Clinical samples of the gastric cancer required by the invention are provided by the pathology department of Zhongshan Hospital of Fudan university: 206 patients with gastric cancer, including DOS subcohort (44 treatments with S-1 (referring to drugs with tegafur, gimeracil and oteracil potassium as active ingredients) and oxaliplatin in combination with docetaxel), xeloxx subcohort (70 treatments with capecitabine and oxaliplatin), HER2 subcohort (71 treatments based on HER2 (trastuzumab/herceptin) targeted therapy) and other subcohorts (18 treatments not yet known and 3 apatinib or docetaxel combination treatments). All treatment regimens were administered at the standard dose for first line treatment of gastric cancer.

For judging the curative effect of the target focus, conventional X-ray, CT scanning or MRI scanning are all the commonly used methods at present, and can be other clinically acceptable detection and diagnosis means. Based on the currently accepted criteria for evaluating the efficacy of a tumor (response in solid tumors), the evaluation of the efficacy of a target lesion can be classified as Complete Response (CR), partial Response (PR), stable Disease (SD), or Progressive Disease (PD). Objective Remission Rate (ORR), an index of efficacy assessment of tumor response, is the proportion of patients whose tumors have shrunk to a certain amount and maintained for a certain period of time, including cases of Complete Remission (CR) and Partial Remission (PR). Clinically, the Food and Drug Administration (FDA) defines the Objective Remission Rate (ORR) as the sum of PR and CR, and can directly measure the antitumor activity of a Drug.

Gastric cancer patients were divided into 82 patients (sum of CR and PR) sensitive and 103 non-sensitive patients (sum of SD and PD) based on RECIST criteria. In the DOS subcohort, patients were divided into 22 sensitive (CR, 2 pr,20, named DSG) and 22 non-sensitive (SD, 20 pd,2, named DNSG); in the XELOX subcohort, patients were divided into 27 susceptible (CR, 0 pr,27, named XSG) and 42 non-susceptible (SD, 37 pd,5, named XNSG) (1 of which treatment responses were unknown; in the HER2 subcohort, patients were classified into 32 sensitive (CR, 2 pr,30, designated HSG) and 37 non-sensitive (SD, 28 pd,9, designated HNSG) (2 of which no treatment response was known. The treatment information of the other 21 patients with gastric cancer is incomplete, including 1 sensitive person and 2 non-sensitive persons, and the other 18 treatment schemes are not clear. There was no bias in the choice of cases. All cases were staged according to the seventh version staging system of the united states joint committee for cancer (AJCC). All patients provided written informed consent with approval by the hospital ethics committee (B2019-200R).

2. Preparation of gastric cancer protein samples:

the clinical specimens were Formalin Fixed Paraffin Embedded (FFPE) tissues.

Sample pretreatment: taking 3-10 μm thick slices from FFPE block, performing macro dissection, xylene dewaxing, ethanol washing, and air drying to obtain white slice, and taking Hematoxylin-eosin (H & E) slice as reference under tumor scope. The tumor cell content in the section is judged to be more than 80 percent by evaluation of two gastrointestinal tract pathologists and is added into the group.

Extracting sample proteins and peptide fragments: equal amounts of FFPE tissue were collected in EP tubes, added to lysis buffer (0.1M Tris-HCl pH 8.0, 0.1M DTT,1mM PMSF), and then ground with a grinding bar for 3 minutes; adding Sodium Dodecyl Sulfate (SDS) to make the final concentration 4%, shaking at 1800rpm for 2-2.5 hours at 99 ℃; centrifuging at 12,000g for 5 min, collecting the supernatant in an EP tube, adding 4 volumes of acetone, and standing at-20 deg.C for 4 hr or overnight; 12,000g centrifuging at 4 ℃ for 1 minute, discarding the supernatant, retaining the precipitate, washing the precipitate with cold acetone for three times, and air-drying the protein precipitate in an ultra-clean bench; protein pellets were reconstituted with 8M Urea and 50mM NH4HCO3 and added to the FASP tubes, and Urea was removed by repeated centrifugation with 50mM NH4HCO3; 50 μ L of 50mM NH4HCO3 in which 5.5 μ g trypsin was dissolved was added to the FASP tube and incubated at 37 ℃ for 18-20 hours for enzymatic digestion; 12,800g, centrifuging for 15 minutes, collecting peptide fragments, washing with 200 μ L MS water twice for improving the yield of the peptide fragments, and collecting the peptide fragments; vacuum-pumping at 60 ℃ to obtain the peptide segment required for mass spectrum detection.

3. Mass spectrometric detection of gastric cancer protein samples:

detection was performed using a Q-exact HF-X hybrid quadrupole orbitrap mass spectrometer (Thermo Fisher Scientific, rockford, IL, USA) and high performance liquid chromatography (EASY nLC 1200, thermo Fisher) and mass spectral data corresponding to the peptide sample was obtained. The drained peptide sample was redissolved in solvent A (0.1% formic acid in water), loaded onto a trap column (100 μm. Times.2 cm; particle size, 3 μm; pore size,

) Then, separation was performed on an analytical column (150. Mu. M.times.12 cm, particle size, 1.9 μm; aperture size->

) Gradient 5-35% mobile phase B (80% acetonitrile and 0.1% formic acid) elution, flow rate of 600nL/min, total elution for 75 minutes. MS analysis of QE-HFX, one full scan (300-1400 m/z, resolution = 12000),60 data-dependent MS/MS scanning orbiter (resolution = 7500) with maximum number of ions allowed to enter in the ion trap (AGC target) at 3e +06 ions, followed by high energy collision induced dissociation (isolation window 1.6m/z, collision energy 27%, AGC target at 5e +04 ions, maximum injection time 30MS, dynamic exclusion set to 18 seconds). The liquid chromatography tandem mass spectrometry system uses Xcalibur software (Thermo Scientific) control for data acquisition.

4. Data processing:

the raw files were retrieved from the Refseq protein database at the human National Center for Biotechnology Information (NCBI) using Firmiana database (https:// phenomics. Fudan. Edu. Cn/Firmiana/gardener /) and MaxQuant software. The Firmiana is a workflow based on a Galaxy system and consists of a plurality of functional modules such as a user login interface, original data, identification and quantification, data analysis, knowledge mining and the like. Trypsin was chosen as proteolytic enzyme, allowing maximally two sites of cleavage missing, the fixed modification being carbamidomethyl (C), the dynamic modification being protein acetyl (protein N-term), oxidation (M). The first search mass tolerance was 20ppm and the main search peptide tolerance was 0.5da. Peptide profile matching (PSMs) and protein False Discovery Rate (FDR) were both less than 1%. The invention adopts an intensity-based absolute quantification method (iBAQ) without a standard quantity and a calculation method without a mark. FOT (fraction of total) is often used in this study to indicate the normalized abundance of a particular protein in a sample. FOT is defined as the value of the iBAQ of one protein divided by the sum of the iBAQ of the experimental sample as a whole. Proteins with at least one unique peptide fragment (unique peptide) and an FDR of less than 1% were selected for further analysis.

Specific results showed that the proteomes of the DOS-sensitive group (DSG) and DOS-insensitive group (DNSG) covered 9119 and 8785 Gene Products (GPs), and 234 GPs were screened for DEP of DSG versus DNSG (expression was more than 2-fold and wilcox p <0.05 by non-parametric test) (fig. 4, table 1).

Table 1

/>

/>

/>

/>

/>

/>

/>

/>

/>

The same analysis was used for both the XELOX and HER2 sub-cohorts. The XELOX sub-cohort identified a total of 10397 GPs, the proteomes of the XELOX Sensitive (XSG) and XELOX non-sensitive (XNSG) groups covered 9039 and 9612 GPs, and the DEP of XSG versus XNSG was screened for a total of 164 GPs (fig. 5, table 2).

Table 2

/>

/>

/>

/>

/>

HER2 sub-cohort identified 10605 GPs, the HER2 sensitive group (HSG) and the HER2 non-sensitive group (HNSG) proteomes covering 9398 and 9554 GPs, and the screening of the DEPs for HSG versus HNSG consisted of 245 GPs (fig. 6, table 3).

Table 3

/>

/>

/>

/>

/>

/>

/>

According to the protein expression data and clinical treatment response information of the treated cancer patient pre-treatment sample respectively adopting DOS, XELOX and HER2 schemes, the sample is divided into a DOS Sensitive Group (DSG), a DOS non-sensitive group (DNSG), a XELOX Sensitive Group (XSG), a XELOX non-sensitive group (XNSG), a HER2 sensitive group (HSG) and a HER2 non-sensitive group (HNSG), the characteristic proteins of the groups are screened to respectively obtain DSG, DNSG, XSG, XNSG, HSG and HNSG proteins, and the input matrix is input to construct a DOS subcohort, a XELOX subcohort and a HER2 subcohort protein profile input matrix.

5. Model construction

Specifically, using R studio software, a prom data analysis packet is loaded, and 80% of randomly extracted samples are set as a training set and 20% of samples are set as a test set for input queues (DOS subqueue, xeloxx subqueue, and HER2 subqueue); and applying a multivariate binary classification logistic regression analysis method to DEP setting, setting family = binary in a glm function, and direction = 'backward' in a step function, and respectively establishing DOS, XELOX and HER2 schemes based on a multivariate binary classification logistic regression prediction model according to AIC values (figure 7).

In DOS subcohorts, a group of DEPs with the smallest AIC as model characteristic proteins show that the prediction accuracy rate is 1 (95% confidence interval [ CI ]: 0.9-1) in 80% of training sets, and the sensitivity and the specificity are both 100%; the predicted accuracy was 0.8889 (95% confidence interval [ CI ]: 0.5175-0.9972) at 20% of the test set, with 100% sensitivity and 83% specificity, respectively (FIG. 9).

In the XELOX subcohort, the minimum group of DEPs of the AIC as model characteristic proteins showed a prediction accuracy of 1 in the 80% training set (95% confidence interval [ CI ]: 0.9351-1), and both sensitivity and specificity were 100%; the prediction accuracy was 0.8889 (95% confidence interval [ CI ]: 0.7684-1) with 100% sensitivity and specificity in the 20% test set (FIG. 10).

In HER2 subcohorts, a group of DEP with the smallest AIC as a model characteristic protein shows that the prediction accuracy rate is 1 in 80% of training sets (95% confidence interval [ CI ]: 0.9351-1), and the sensitivity and the specificity are both 100%; the prediction accuracy was 0.7143 (95% confidence interval [ CI ]: 0.419-0.9161) in the 20% test set, with 71% sensitivity and specificity (FIG. 11).

Example 2 testing the stability of the predictive model

As described in example 1, the present invention establishes a model for predicting the response of gastric cancer treatment based on proteome expression data in combination with a machine learning method. The abundance signal of the discriminatory protein selected after subsequent statistical analysis is directed into a machine learning algorithm to obtain a statistical or mathematical model, i.e. classifier, that classifies the protein spectral data with a certain accuracy, sensitivity and specificity.

In order to further evaluate the accuracy of the treatment response prediction model and simulate the decision-making method of individualized gastric cancer treatment, the invention verifies the application of the model in a new independent queue and other data based on the prediction model.

1. Application of gastric cancer treatment prediction model established based on example 1 in new independent queue

To further evaluate the accuracy of the predictive model of treatment response, the present invention further included 13 gastric cancer patients from the pathology department of Zhongshan Hospital, 6 of which received DOS treatment and 7 received XELOX treatment to validate the accuracy of the DOS predictive model and the XELOX predictive model.

When prediction was performed using the DOS model obtained in example 1, 6 DOS-treated gastric cancer patients were added to 44 gastric cancer patients in example 1, and 50 gastric cancer patients were used as a new validation set. In the new validation set, the accuracy of the predictive model for DOS treatment regimens was 0.96, (95% confidence interval [ CI ]: 0.8629-0.9951), sensitivity and specificity were 96.3%, 95.65%, respectively (fig. 12), and the AUC for the test set was 1; the stability and accuracy of the DOS prediction model can be seen.

When prediction was performed using the XELOX model obtained in example 1, 7 XELOX-treated gastric cancer patients were added to 69 gastric cancer patients in example 1, and 76 gastric cancer patients were used as a new validation set. In the new validation set, the predictive model of the XELOX treatment regimen was accurate at 0.9474, (95% confidence interval [ CI ]: 0.8707-0.9855), sensitivity and specificity were 91.49%, 100%, respectively (FIG. 13), and the AUC for the test set was 1; the stability and accuracy of the XELOX prediction model can be seen.

2. Application of gastric cancer treatment prediction model established based on example 1 in other central data

The clinically annotated diffuse gastric carcinoma proteome database established by beijing proteome research center has accession number PXD00884020 for prid file, of which 45 are annotated as receiving XELOX treatment protocol and have long-term disease-free survival follow-up (DFS) (table 4), of which 25 are XELOX sensitive groups and 20 are XELOX non-sensitive groups. When the DGC cohort was predicted using the XELOX model obtained in example 1, the predicted AUC was 0.9474 (FIG. 14). The results demonstrate the applicability and accuracy of the predictive model for xeloxx treatment of the present invention in other gastric cancer proteomic data.

Thus, the DOS and XELOX predictive models obtained as described in example 1 have high predictability, stability, and universality.

Example 3 decision making method for personalized gastric cancer treatment

For the gastric cancer patient to be treated, a personalized treatment plan is recommended according to the prediction model (fig. 8), and the method specifically comprises the following steps (fig. 1).

Step 201: predictive models for different treatment regimens were constructed according to example 1.

Step 202:

(1) Extraction of proteome data of a sample of a patient to be treated

Specifically, 5 patients with gastric cancer to be treated, numbered #1, #2, #3, #4, #5, respectively, were obtained as described in example 1 and subjected to mass spectrometry to obtain proteome data thereof;

(2) Calculating patient predicted probabilities for different treatment regimens

Protein expression data of 5 samples of gastric cancer patients to be treated were input to prediction models of different treatment regimens of DOS, xeloxx, HER2 as in example 1, and prediction probabilities sensitive to each treatment regimen were calculated by R studio software using prediction function, parameter set type = "prob".

(3) Recommendation of treatment regimens

Based on the prediction probabilities in table 4 below, the largest one of the prediction probabilities that is sensitive to the treatment regimen is selected and the treatment regimen is recommended to the patient.

Table 4: prediction results and treatment scheme decision of 5 gastric cancer patients

Embodiment 4 construction device of gastric cancer treatment prediction model and personalized gastric cancer treatment decision system

1. Device for constructing gastric cancer treatment prediction model

The present embodiment provides a device 51 for constructing a gastric cancer treatment prediction model, as shown in fig. 2, including: a sample grouping module 41, a characteristic protein selection module 42 and a prediction model construction module 43.

The sample grouping module 41 is used for obtaining clinical samples of patients treated by different treatment schemes, and dividing the patients into sensitive groups and non-sensitive groups according to treatment results; performing mass spectrum detection on the clinical sample to obtain protein expression data, screening the differentially expressed protein DEP of which the protein expression amount of the sensitive group relative to the non-sensitive group is more than 2 times or the protein expression amount of the non-sensitive group relative to the sensitive group is more than 2 times and p is less than 0.05, and inputting the differentially expressed protein DEP into a matrix.

The characteristic protein selecting module 42 calculates an akage information content criterion (AIC) of the DEP by adopting a multivariate binary classification stepwise logistic regression method to the DEP in the sample grouping module to obtain a group of DEP with the minimum AIC value as the characteristic protein of the treatment scheme.

The prediction model construction module 43 randomly extracts at least 50% of clinical samples in each treatment scheme, uses the corresponding characteristic proteins as a training set, uses the characteristic proteins corresponding to the remaining clinical samples as a verification set, and adopts multivariate binary classification stepwise logistic regression to perform cross-verification on the training set; obtaining the prediction model of the treatment scheme.

2. Personalized gastric cancer treatment decision making system

The personalized gastric cancer treatment decision system 61 includes: a device 51 for constructing a gastric cancer treatment prediction model and a scheme decision module 52 (fig. 2).

Information of the construction apparatus 51 of the gastric cancer treatment prediction model is shown in the first section; and the combination of (a) and (b),

the plan decision module 52 extracts protein expression data of clinical samples of patients to be treated, calculates prediction probabilities of the patients sensitive to each treatment plan by combining prediction models of different treatment plans through R studio software by using a prediction function and parameter setting type = 'prob', and recommends the treatment plan with the highest prediction probability to the patients.

In addition, according to the method for constructing the gastric cancer treatment prediction model, the combination of each characteristic biomarker in the patient sample corresponding to each treatment scheme is obtained (see tables 1 to 3). When the biomarker combination is detected in a sample from a patient to be treated, one skilled in the art can determine which treatment regimen the patient to be treated is appropriate for. The biomarker combinations include DOS treatment regimen related marker combinations, xeloxx treatment regimen related marker combinations, and HER2 treatment regimen related marker combinations; the DOS treatment regimen-related marker combination comprises one or more of the following markers: AHR, ATP5S, C11orf31, C20orf26, CDC42SE2, CHGA, and CHP2; the marker associated with the XELOX treatment regimen is combined with one or more of the following markers: RFC2, RAB32, FLG2, FNBP1, GCLC, DYNLRB1, NIT1, RBBP7, LPXN, LMAN2, NUB1, WAS, RMDN1, DFFB and MYCBP; the HER2 treatment regimen related marker is in combination with one or more of the following markers: BAIAP2, CAPN5, COMMD4, DDX60L, DSC, IRF6, NPL, SCIN, SEPSECS, SLC39A4, SLC5A5, SRPX2, and TECPR1. For more precise recommendation of the treatment regimen, it is also possible to determine which regimen the patient is suitable for treatment directly from which biomarker combination in tables 1, 2 and 3 the sample of the patient to be treated corresponds.

Embodiment 5 electronic device

The present embodiment provides an electronic device, which may be represented in the form of a computing device (for example, may be a server device), and includes a memory, a processor, and a computer program stored in the memory and running on the processor, where the processor, when executing the computer program, may implement the method for constructing a prediction model in embodiment 1 and the method for deciding personalized gastric cancer therapy in embodiment 3 of the present invention.

Fig. 3 shows a schematic diagram of a hardware structure of the embodiment, and the electronic device 9 specifically includes:

at least one processor 91, at least one memory 92, and a bus 93 for connecting the various system components (including the processor 91 and the memory 92), wherein:

the bus 93 includes a data bus, an address bus, and a control bus.

Memory 92 includes volatile memory, such as Random Access Memory (RAM) 921 and/or cache memory 922, and can further include Read Only Memory (ROM) 923.

Memory 92 also includes a program/utility 925 having a set (at least one) of program modules 924, such program modules 924 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof may comprise an implementation of a network environment.

The processor 91 executes a computer program stored in the memory 92, thereby executing various functional applications and data processing, such as the prediction model in embodiment 1 and the decision method for personalized gastric cancer treatment in embodiment 3 of the present invention.

The electronic device 9 may further communicate with one or more external devices 94 (e.g., a keyboard, a pointing device, etc.). Such communication may be through an input/output (I/O) interface 95. Also, the electronic device 9 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet) via the network adapter 96. The network adapter 96 communicates with the other modules of the electronic device 9 via the bus 93. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 9, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

It should be noted that although in the above detailed description several units/modules or sub-units/modules of the electronic device are mentioned, such a division is merely exemplary and not mandatory. Indeed, the features and functions of two or more units/modules described above may be embodied in one unit/module according to embodiments of the application. Conversely, the features and functions of one unit/module described above may be further divided into embodiments by a plurality of units/modules.

Example 6 computer-readable storage Medium

Embodiments of the present invention provide a computer-readable storage medium on which a computer program is stored, which when executed by a processor, implements the steps of the method for constructing a prediction model in embodiment 1, and the steps of the method for deciding a personalized gastric cancer therapy in embodiment 3 of the present invention.

More specific examples, among others, that the readable storage medium may employ may include, but are not limited to: a portable disk, a hard disk, random access memory, read only memory, erasable programmable read only memory, optical storage device, magnetic storage device, or any suitable combination of the foregoing.

In a possible embodiment, the present invention can also be implemented in the form of a program product including program code for causing a terminal device to perform the steps of implementing the method for constructing a predictive model in embodiment 1, the method for deciding personalized gastric cancer therapy in embodiment 3 of the present invention, when the program product is run on the terminal device.

Where program code for carrying out the invention is written in any combination of one or more programming languages, the program code may be executed entirely on the user device, partly on the user device, as a stand-alone software package, partly on the user device and partly on a remote device or entirely on the remote device.

While specific embodiments of the invention have been described above, it will be appreciated by those skilled in the art that this is by way of example only, and that the scope of the invention is defined by the appended claims. Various changes and modifications to these embodiments may be made by those skilled in the art without departing from the spirit and scope of the invention, and these changes and modifications are within the scope of the invention.

Claims

1. A method for constructing a gastric cancer treatment prediction model is characterized by comprising the following steps:

(1) Dividing sample groups: obtaining clinical samples of gastric cancer patients treated by different treatment schemes, and dividing the patients into a sensitive group and a non-sensitive group according to treatment results; performing mass spectrum detection on the clinical sample to obtain protein expression data, screening differentially expressed protein DEP (Dep) of which the protein expression amount of the sensitive group relative to the non-sensitive group is more than 2 times or the protein expression amount of the non-sensitive group relative to the sensitive group is more than 2 times and p is less than 0.05, and inputting the differentially expressed protein DEP into a matrix; the treatment regimens include DOS, XELOX, and HER2 regimens, and the samples are divided into a DOS sensitive group, a DOS insensitive group, an XELOX sensitive group, an XELOX insensitive group, a HER2 sensitive group, and a HER2 insensitive group;

(3) Constructing a prediction model: randomly extracting 80% of clinical samples in each treatment scheme, taking the corresponding characteristic protein as a training set, taking the characteristic protein corresponding to the rest clinical samples as a test set, and performing cross validation on the training set by adopting multivariate binary classification stepwise logistic regression; thus obtaining a prediction model of DOS, XELOX and HER2 treatment schemes.

2. The method of constructing a prediction model for gastric cancer treatment according to claim 1,

in the step (2), the AIC is calculated by using R studio software; adopting a pROC data packet, and setting family = binary in a glm function and direction = 'backward' in a step function; and/or the presence of a gas in the gas,

in the step (3), the cross validation is 10 times of cross validation, and the cross validation is repeated for 10 times; and/or, providing clinical treatment response information includes treatment efficacy assessed using normal X-ray, CT scan, and/or MRI scan; and/or the presence of a gas in the gas,

3. The method for constructing a gastric cancer treatment prediction model according to claim 1or 2,

the obtaining of the protein expression data comprises the following steps:

(A) Sample preparation: preparing the sample into a formalin-fixed paraffin-embedded tissue section, wherein the gastric cancer cells of the sample account for more than 80% of the total number of the cells; the lysis buffer used for lysing the sample was 0.1M Tris-HCL, pH8.0, 0.1M DTT and 1mM PMSF were added; 50mM NH with trypsin ₄ HCO ₃ Carrying out enzymolysis on the protein of the sample, incubating for 18-20 hours at 37 ℃, and centrifuging and collecting to obtain an enzymolysis peptide segment; and/or the presence of a gas in the gas,

(B) Mass spectrum detection: detecting the peptide segment by using a Q-exact HF-X mixed quadrupole orbitrap mass spectrometer and a high performance liquid chromatography system, and obtaining mass spectrum data corresponding to the peptide segment; controlling data acquisition by using Xcalilibur software; and/or the presence of a gas in the atmosphere,

(C) Data processing: the acquired data is processed by using a Firmiana database and MaxQuant software; the first search mass tolerance was 20ppm and the main search peptide tolerance was 0.5da; the calculation method is label-free iBAQ, and FOT is used for expressing the standardized abundance of protein in the sample; selecting a protein having at least one specific peptide stretch and an FDR of less than 1%.

4. A decision-making method for personalized gastric cancer treatment is characterized by comprising the following steps:

(a) Obtaining prediction models of different treatment regimens obtained by the method for constructing a gastric cancer treatment prediction model according to any one of claims 1 to 3;

5. A device for constructing a gastric cancer treatment prediction model is characterized by comprising:

a sample grouping module: the sample grouping module is used for obtaining clinical samples of patients treated by different treatment schemes and dividing the patients into sensitive groups and non-sensitive groups according to treatment results; performing mass spectrum detection on the clinical sample to obtain protein expression data, screening differential expression protein DEP (deoxyribose nucleic acid) of which the protein expression amount of the sensitive group relative to the non-sensitive group is more than 2 times or the protein expression amount of the non-sensitive group relative to the sensitive group is more than 2 times and p is less than 0.05, and inputting the differential expression protein DEP into a matrix; the treatment regimens include DOS, XELOX, and HER2 regimens, and the samples are divided into a DOS sensitive group, a DOS insensitive group, an XELOX sensitive group, an XELOX insensitive group, a HER2 sensitive group, and a HER2 insensitive group; and (c) and (d),

a characteristic protein selection module: the characteristic protein selection module calculates an erythroid information criterion (AIC) of the DEP by adopting a multivariate binary classification stepwise logistic regression method for the DEP of the sample grouping module to obtain a group of DEP with the minimum AIC value as characteristic proteins of DOS, XELOX and HER2 treatment schemes; and (c) and (d),

a prediction model construction module: the prediction model construction module randomly extracts at least 80% of clinical samples in each treatment scheme, takes the corresponding characteristic proteins as a training set and the characteristic proteins corresponding to the rest clinical samples as a test set, and performs cross validation on the training set by adopting multivariate two-classification stepwise logistic regression; thus obtaining a prediction model of DOS, XELOX and HER2 treatment schemes.

6. The apparatus for constructing a prediction model for gastric cancer treatment according to claim 5,

in the characteristic protein selection module, R studio software is used for calculating the AIC value; adopting a pROC data packet, and setting family = binary in a glm function and direction = 'backward' in a step function; and/or the presence of a gas in the atmosphere,

in the prediction model building module, the cross validation is 10 times cross validation and is repeated for 10 times; and/or, providing clinical treatment response information includes treatment efficacy assessed using normal X-ray, CT scan, and/or MRI scan; and/or the presence of a gas in the gas,

the patients were grouped based on RECIST criteria, where the sensitive group included complete remission and partial remission and the non-sensitive group included disease stabilization and disease progression.

7. The apparatus for constructing a prediction model for gastric cancer treatment according to claim 5 or 6, further comprising:

a sample preparation module: the sample preparation moldPreparing a sample into a formalin-fixed paraffin-embedded tissue section, wherein gastric cancer cells of the sample account for more than 80% of the total number of cells; the lysis buffer used for lysis of the sample was 0.1M Tris-HCl, pH 8.0, to which 0.1M DTT and 1mM PMSF were added; 50mM NH with trypsin ₄ HCO ₃ Carrying out enzymolysis on the protein of the sample, incubating for 18-20 hours at 37 ℃, and centrifuging and collecting to obtain an enzymolysis peptide segment; and/or the presence of a gas in the gas,

a mass spectrometric detection module: the peptide section of the mass spectrum detection module is detected by a Q-exact HF-X mixed quadrupole orbitrap mass spectrometer and a high performance liquid chromatography system, and mass spectrum data corresponding to the peptide section is obtained; using Xcalibur software to control data acquisition; and/or the presence of a gas in the atmosphere,

8. A personalized gastric cancer treatment decision system, comprising:

the device for constructing a gastric cancer treatment prediction model according to any one of claims 5 to 7; and

a scheme decision module: the plan decision module extracts protein expression data of clinical samples of patients to be treated, combines prediction models of different treatment plans, calculates prediction probability sensitive to each treatment plan by using a prediction function of R studio software and parameter setting type = 'prob', and recommends the treatment plan with the maximum prediction probability to the patients.

9. An electronic device comprising a memory and a processor; the memory includes a computer program stored therein that is executable on the processor; wherein the processor implements the method for constructing a gastric cancer treatment prediction model according to any one of claims 1 to 3 or the method for deciding an individualized gastric cancer treatment according to claim 4 when executing the computer program.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, is adapted to carry out the steps of the method for constructing a gastric cancer treatment prediction model according to any one of claims 1 to 3 or the steps of the method for deciding on an individualized gastric cancer treatment according to claim 4.