Gastric cancer naic efficacy prediction method based on metabolomics and specific metabolites

Abstract

This invention discloses a metabolomics-based method for predicting the efficacy of non-invasive pretreatment intervention (NAIC) in gastric cancer and specific metabolites, belonging to the field of biodetection technology. The method includes: obtaining plasma samples from gastric cancer patients before treatment, detecting the abundance of their metabolites, and obtaining baseline plasma metabolomics data; classifying patients into a response group and a non-response group based on post-treatment imaging assessment results; comparing the baseline metabolomics data of the two groups to screen for differentially expressed metabolites; using these differentially expressed metabolites as features, training the model using LASSO regression, determining the optimal parameters through cross-validation, and constructing an efficacy prediction model composed of specific metabolites and their regression coefficients; for the sample to be predicted, detecting the abundance of its specific metabolites and inputting them into the model to obtain a risk score, and predicting the patient's response to treatment by comparing it with a risk threshold. This invention enables accurate efficacy prediction before treatment, providing an important basis for individualized clinical treatment decisions.

Description

Technical Field

[0001] This invention belongs to the field of biological detection technology, specifically relating to a method for predicting the efficacy of gastric cancer NAIC based on metabolomics and specific metabolites. Background Technology

[0002] Gastric cancer (GC) is the fifth most common malignant tumor, accounting for approximately 33% of cancer-related deaths worldwide. Alarmingly, about 70% to 90% of GC patients are diagnosed at an advanced stage, significantly impacting prognosis. While surgery and chemotherapy have improved survival rates for patients with advanced GC to some extent, the overall survival rate remains below 40%, and more than half of GC patients experience recurrence after treatment. The MAGIC trial showed that patients receiving perioperative chemotherapy had a 5-year survival rate of 36%, significantly higher than those who underwent surgical resection alone (23%). Preoperative chemotherapy is considered a standard treatment and a promising approach to improving survival rates for patients with locally advanced GC. However, despite neoadjuvant chemotherapy (NAC) and radiotherapy being recommended as standard treatments for locally advanced GC, the overall prognosis remains poor, with a median overall survival of less than 12 months.

[0003] In recent years, immune checkpoint inhibitors (ICIs), such as toripalimab, tislelizumab, sintilimab, and nivolumab, have made significant progress in the treatment of advanced gastric cancer, effectively improving patient survival rates. Other studies have shown that preoperative neoadjuvant immunotherapy combined with chemotherapy can significantly delay the progression of gastric cancer. For example, a prospective, single-arm, phase II clinical trial showed that neoadjuvant immunochemotherapy (NAIC) for gastric adenocarcinoma patients achieved a pathologic complete regression rate (pCR) of 30% and a major pathologic regression rate (MPR) of 43%, with good tolerability. The CheckMate 649 study demonstrated that NAIC treatment provided beneficial therapeutic effects for patients. Verschoor et al. found the safety and significant antitumor activity of neoadjuvant atezolizumab combined with chemotherapy in G / GEJ adenocarcinoma, with 70% of patients achieving MPR and 45% achieving pCR. Crucially, the study also found a strong correlation between pathological response and survival rate after NAIC treatment.

[0004] However, a significant proportion of gastric cancer patients do not benefit from non-invasive intravascular coagulation (NAIC). This is because current clinical prognosis and efficacy assessment for gastric cancer primarily rely on imaging, endoscopy, or histopathological examination. Clinicians use these methods to collect clinical characteristics such as tumor location, TNM staging, and Lauren classification for efficacy evaluation, but the accuracy of these assessments is limited. Imaging examinations expose patients to radiation, and while endoscopy is the gold standard for gastric cancer diagnosis, it is invasive and expensive. These drawbacks hinder real-time monitoring of patient efficacy, highlighting the urgent need to identify reliable biomarkers for patient stratification during NAIC treatment.

[0005] Finding easily detectable and highly specific biomarkers to predict the efficacy of neoadjuvant chemotherapy combined with immunotherapy for gastric cancer is a pressing technical problem that needs to be solved. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the prior art and to provide a method for predicting the efficacy of NAIC in gastric cancer based on metabolomics and specific metabolites.

[0007] The specific technical solution adopted in this invention is as follows:

[0008] This invention provides a metabolomics-based method for predicting the efficacy of non-invasive prenatal testing (NAIC) in gastric cancer, with the following specific steps:

[0009] S1: Obtain plasma samples from gastric cancer patients before they receive neoadjuvant immunochemotherapy; detect the abundance of metabolites in the plasma samples to obtain baseline plasma metabolomics data;

[0010] S2: After the gastric cancer patients described in step 1 receive neoadjuvant immunochemotherapy, the gastric cancer patients are divided into a treatment-responsive group and a treatment-non-responsive group based on imaging assessment.

[0011] S3: Compare baseline plasma metabolomic data between the treatment-responsive group and the treatment-unresponsive group, and screen out metabolites that show significant differences between the two groups as differential metabolites;

[0012] S4: Using the abundance data of the differential metabolites as features and the treatment response as a variable, a LASSO regression model is used for training, and the optimal regularization parameter is determined through cross-validation to construct an efficacy prediction model; the efficacy prediction model consists of several specific metabolites and their corresponding regression coefficients.

[0013] S5: Detect the abundance data of the specific metabolites in the plasma sample to be predicted; input the abundance data of the specific metabolites into the efficacy prediction model and output the efficacy risk score; compare the efficacy risk score with the risk threshold to predict its response to neoadjuvant immunochemotherapy.

[0014] Preferably, the metabolite abundance data in step S1 is standardized; the standardization is performed using the Z-score method.

[0015] Preferably, in step S3, principal component analysis and orthogonal partial least squares discriminant analysis are used to screen candidate differential metabolites.

[0016] As a preferred option, the method for determining the optimal regularization parameter in step S4 is as follows: the parameters of the LASSO regression model are optimized using 10-fold cross-validation, and the optimal regularization parameter λ is determined using the λ.1se rule.

[0017] Preferably, the specific metabolites mentioned in step S4 include one or more combinations of 8,9-dihydroxyeicosalicylic acid, 13,14-dihydroprostaglandin F-1a, 12S-hydroxyhexadecanosalicylic acid, thiolysine ketoimine, 4-pyridoxine, 3-hydroxydecanoylcarnitine, coenzyme Q2, 4-hydroxy-2-ketoglutarate, corticosterone, 15,16-dihydroxyoctadecanodienoic acid, 2-aminoadenosine, 3-hydroxy-cis-5-tetradecanoylcarnitine, 3-hydroxynonanoylcarnitine, aspartic acid-tyrosine dipeptide, di-high-gamma-linolenic acid, hexanoylcarnitine, hydroxyprogesterone, hydroxytryptamine, N6,N6-dimethyllysine, S-adenosine homocysteine, and thymidine deoxyribonucleoside.

[0018] Preferably, the formula for the efficacy prediction model in step S4 is as follows:

[0019] ;

[0020] In the formula: P is the efficacy risk score; Let be the regression coefficient of the i-th specific metabolite; Here are the standardized abundance data for the i-th metabolite.

[0021] Preferably, the risk threshold is calculated as follows: the baseline plasma metabolome data of gastric cancer patients obtained in step S1 is used as the training set; the training set is input into the efficacy prediction model in step S5, and the efficacy risk score is output; the median of all efficacy risk scores is used as the risk threshold.

[0022] Preferably, in step S5, if the efficacy risk score of the plasma sample to be predicted is less than the risk threshold, the gastric cancer patient is determined to have a high risk of benefit from neoadjuvant immunochemotherapy; otherwise, the risk of benefit is low.

[0023] Preferably, the plasma sample to be predicted is a plasma sample from a gastric cancer patient before receiving neoadjuvant immunochemotherapy.

[0024] Secondly, the present invention provides a specific metabolite for predicting the efficacy of NAIC in gastric cancer, wherein the specific metabolite is one or more of the following combinations: 8,9-dihydroxyeicosatetrienoic acid, 13,14-dihydroprostaglandin F-1a, 12S-hydroxyhexadecanoic acid, thiolysine ketoimine, 4-pyridoxine, 3-hydroxydecanoic acid, coenzyme Q2, 4-hydroxy-2-ketoglutarate, corticosterone, 15,16-dihydroxyoctadecanoic acid, 2-aminoadenosine, 3-hydroxy-cis-5-tetradecanoic acid, 3-hydroxynonanoic acid, aspartic acid-tyrosine dipeptide, di-high-gamma-linolenic acid, hexanoic acid, hydroxyprogesterone, hydroxytryptamine, N6,N6-dimethyllysine, S-adenosine homocysteine, or thymidine deoxyribonucleoside.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] (1) The method provided by this invention integrates metabolomics and machine learning technologies to construct a predictive model based on pre-treatment plasma metabolites. By observing the comprehensive predictive ability of pre-treatment plasma metabolite indicators and clinical indicators of NAIC, it provides important clinical intervention basis for the prognosis of gastric cancer and helps to provide individualized and precise intervention for gastric cancer patients.

[0027] (2) Compared with traditional methods that rely on post-treatment assessment, this invention can predict the efficacy before treatment begins, significantly advancing the prediction window. This helps clinicians develop more rational medication strategies, avoid ineffective treatment, thereby improving patients' quality of life and optimizing the allocation efficiency of medical resources. Attached Figure Description

[0028] Figure 1 The principal component analysis diagram (a) and orthogonal partial least squares discriminant analysis diagram (b) for the treatment-responsive group and the treatment-non-responsive group in the example are shown.

[0029] Figure 2 Volcano plot analysis of the treatment-responsive group and the treatment-non-responsive group in the examples;

[0030] Figure 3 The regression coefficients for 21 specific metabolites in the LASSO regression model;

[0031] Figure 4 The results of the constructed efficacy prediction model in the training set (a) and prediction set (b) are shown, where the left side is the receiver operating curve and the right side is the confusion matrix;

[0032] Figure 5The results of the risk score and its clinical association analysis based on the efficacy prediction model are shown, where (a) is the receiver operating curve; (b) is the distribution of treatment-responsive and treatment-non-responsive groups in the high-benefit-risk group and the low-benefit-risk group; and (c) is the distribution of efficacy results for different solid tumors in the high-benefit-risk group and the low-benefit-risk group. Detailed Implementation

[0033] The present invention will be further described and illustrated below with reference to the accompanying drawings and specific embodiments. The technical features of each embodiment of the present invention can be combined accordingly, provided that there is no mutual conflict.

[0034] Unless otherwise specified, the techniques used in the examples are conventional methods well known to those skilled in the art. All reagents used in this invention are of analytical grade or higher. The chromatographic column used is an ACQUITY UPLC HSST3, manufactured by Waters; the liquid chromatograph is an Orbitrap Exploris 120 mass spectrometer, manufactured by Thermo. The reagents used in this invention include: mass spectrometry grade formic acid (FA), mass spectrometry grade methanol, mass spectrometry grade acetonitrile, ketoprofen, and sulfamethazine.

[0035] Example

[0036] I. Clinical Sample Collection

[0037] (1) This embodiment collected data from 108 gastric cancer patients who received NAIC treatment at the First Affiliated Hospital of Zhejiang University School of Medicine from October 30, 2021 to June 30, 2023.

[0038] The inclusion criteria for the study were: (1) patients diagnosed with gastric adenocarcinoma by endoscopic biopsy histology before treatment; (2) patients diagnosed with locally advanced gastric cancer, regional unresectable or clinical stage IV disease according to the AJCC / UICC 8th edition staging system, mainly assessed by computed tomography (CT); and (3) patients who received chemotherapy combined with immune checkpoint inhibitor therapy (with or without trastuzumab).

[0039] Exclusion criteria were: (1) those with a history of other malignant tumors other than cured basal cell carcinoma of the skin or cervical carcinoma in situ; (2) those who had previously undergone major gastric surgery; (3) those who had previously received systemic treatment or radiotherapy for gastric cancer; (4) those who lacked efficacy assessment information; and (5) those who lacked baseline plasma samples.

[0040] (2) Record the baseline clinical characteristics of all patients, including age, sex, tumor differentiation status, Lauren classification, ECOG performance status score, etc.

[0041] (3) Before the start of treatment, the patient’s plasma sample (baseline sample) was collected. After fasting for 12 hours, blood was collected in an empty stomach and plasma was separated according to the following procedure: The collected blood was centrifuged at 1200 rpm for 10 min at 4°C, the supernatant was collected and frozen at -80°C until metabolite extraction was performed.

[0042] II. Extraction and Detection of Plasma Metabolites

[0043] Plasma samples from 108 patients were slowly thawed on ice. 100 μL of plasma was then mixed with 400 μL of pre-chilled methanol, vortexed for 3 minutes, and then centrifuged at 13,000 rpm for 15 minutes at 4°C. The supernatant was collected and evaporated to dryness using a centrifugal vacuum evaporator at 4°C. The dried metabolite samples were stored at -80°C until LC-MS / MS analysis was performed.

[0044] To ensure consistent processing and data reliability across all samples, a quality control (QC) sample was prepared. The method involved extracting 6 μL of plasma from each study patient, pooling it into a single "pooled sample," and then subjecting this pooled sample to the same extraction and processing procedures as described above. In subsequent instrumental analysis, the QC sample underwent the same treatment as other experimental plasma samples.

[0045] A comprehensive metabolomics analysis was performed on 369 serum samples from 108 gastric cancer patients using ultra-high performance liquid chromatography-tandem mass spectrometry (UHPLC-Q-Orbitrap-MS / MS).

[0046] Take 300 μL of the supernatant and add 50 μL of working solution containing internal standards (sulfamethazine and ketoibuprofen). Vortex for 30 seconds, sonicate on ice, and centrifuge at 13,000 rpm for 15 minutes at 4°C. Finally, take 30 μL of the supernatant for LC-MS / MS analysis. Metabolites were then separated and detected using an OrbitrapExploris 120 ultra-high performance liquid chromatography-tandem mass spectrometry instrument (Thermo Fisher, USA).

[0047] To eliminate noise and instrument fluctuation interference during instrument operation, quality control samples are added during the liquid chromatography-mass spectrometry (LC-MS) analysis of plasma samples. Specifically, a QC test is inserted every other sample of equal volume. Ultimately, the expression and dispersion coefficients of metabolites in the QC samples are used as the basis for eliminating some interference caused by system noise.

[0048] III. Identification of Plasma Metabolites

[0049] The plasma samples were analyzed using Compound Discover version 3.2 (Thermo Fisher Scientific Inc.) software, and the metabolites were structurally identified using the mzCloud (www.mzcloud.org / ), mzVault (https: / / mytracefinder.com / tag / mzvault / ), KEGG (www.kegg.jp / ), HMDB (https: / / hmdb.ca / ), and ChemSpider (www.chemspider.com / ) databases.

[0050] To eliminate technical errors and ensure data comparability, the identified metabolite abundance data were processed as follows:

[0051] The support vector regression algorithm embedded in the MetNormalizer R package (version 1.3.02) was used to adjust for possible loading / pipette errors and batch effects. The relative standard deviation (RSD) of each metabolite feature in all QC samples was calculated. Features with an RSD ≥ 30% in QC samples were considered unstable and discarded. The retained metabolite abundance data were first logarithmically transformed to base 2 to approximate a normal distribution. Subsequently, Z-score standardization (i.e., subtracting the mean of all samples and dividing by the standard deviation) was performed on each metabolite feature to eliminate dimensional differences between different metabolites and make the data suitable for subsequent statistical analysis.

[0052] IV. Evaluation and Grouping of NAIC Treatment Efficacy

[0053] After all patients completed the prescribed NAIC regimen, a gastrointestinal specialist clinician evaluated their results based on pre- and post-NAIC CT scans, according to the R1.1 criteria for evaluating the efficacy of treatment for solid tumors. Results were categorized into four groups:

[0054] Complete Response (CR): The tumor completely disappears;

[0055] Partial response (PR): The tumor has shrunk significantly;

[0056] Stable disease (SD): The tumor shows little change;

[0057] Disease progression (PD): The tumor grows larger or new lesions appear.

[0058] For ease of analysis, in this embodiment, CR and PR patients are defined as treatment-responsive (R), while SD and PD patients are defined as treatment-unresponsive (NR).

[0059] As of June 30, 2023, 65 of the 108 patients were assessed as having a response to treatment (R group) (64 partial remissions (PR) and 1 complete remission (CR)), and 43 patients were assessed as having no response to treatment (NR group) (stable disease (SD) or progressive disease (PD).

[0060] V. Screening of Differential Metabolites

[0061] Principal component analysis (PCA) was performed on the baseline (T0) plasma metabolomics profiles of 108 samples to assess overall differences between groups. Orthogonal partial least squares discriminant analysis (OPLS-DA) was used to further observe the separation trend between the R group and the NR group, and the results are as follows: Figure 1 As shown in the figure. PCA results showed a significant difference between group R (n=65) and group NR (n=43), especially in the OPLS-DA plot, where the two groups exhibited a significant separation. Preliminary screening of inter-group differences yielded 65 metabolites. After removing duplicates (retaining those with smaller p-values), 61 differentially expressed metabolites remained, as shown in the figure. Figure 2 As shown. Figure 2 Red circles represent significantly upregulated metabolites, blue circles represent significantly upregulated metabolites, gray circles represent metabolites with no significant difference between groups, and the horizontal axis represents the difference in metabolite abundance between groups.

[0062] VI. Construction and Training of Efficacy Prediction Model

[0063] The 108 samples were randomly divided into a training set (n=86, 79.63%) and a validation set (n=22, 20.37%) in an 8:2 ratio. No statistically significant differences were found in baseline clinical characteristics between the two groups. The training set contained 52 treatment-responsive samples (R group) and 34 non-responsive samples (NR group). Given the significant sample imbalance (R / NR≈1.53:1), to avoid excessive bias towards the majority class, this embodiment employed the Synthetic Minority Oversampling Technique (SMOTE) for data balancing (DMwR package, R 0.4.1).

[0064] Based on 61 differentially expressed metabolites, a LASSO regression model was used for feature selection (using the R package glmnet, R 4.1-8). The LASSO regression model parameters were optimized using 10-fold cross-validation, and λ=0.03574295 (1se rule) was ultimately determined as the optimal model parameter. The final efficacy prediction model is as follows:

[0065] ;

[0066] In the formula: P is the efficacy risk score; Let be the regression coefficient of the i-th specific metabolite; Here are the standardized abundance data for the i-th metabolite.

[0067] The model screened out 21 specific metabolites with significant discriminative efficacy. The regression coefficients for each specific metabolite in the model are as follows: 8,9-dihydroxyeicosatotrienoic acid (8,9-DiHETrE) 0.257866854, 13,14-dihydroprostaglandin F-1a (13,14-Dihydro PGF-1a) 0.576923531, 12S-hydroxyhexadecanotrienoic acid (12S-HHT) -0.438257212, Thialysine ketimine 0.008887266, 4-Pyridoxic acid -0.496592545, 3-hydroxydecanoyl carnitine -0.452458428, and coenzyme Q2 (coenzyme Q2) -0.452458428. Q2) is -0.234665227, 4-hydroxy-2-ketoglutaric acid (4-Hydroxy-2-oxoglutaric acid) The values ​​for the following amino acids were: 0.428603563 (acid), -0.001607521 (corticosterone), 0.510721355 (15,16-dihydroxyoctadecadienoic acid), 0.045229722 (2-Aminoadenosine), 0.110793804 (3-hydroxycis-5-tetradecenoylcarnitine), 0.09711028 (3-hydroxynonanoylcarnitine), 0.060312089 (Aspartic acid-tyrosine dipeptide), and 0.060312089 (dihomo-gamma-linolenic acid). The values ​​for (acid) are 0.108673309, hexanoylcarnitine is 0.146304358, hydroxyprogesterone is 0.244304361, hydroxytryptophol is 0.21353077, N6,N6-dimethyllysine is 0.259042898, S-adenosylhomocysteine ​​is 0.272855168, and thymidine is 0.122716209. (Example) Figure 3 As shown.

[0068] The baseline metabolomics data of the training set (86 patients) were input into the efficacy prediction model constructed above to calculate the efficacy risk score (P) for each patient. The median of these 86 scores was calculated and set as the risk threshold for clinical stratification.

[0069] VII. Model Performance Validation

[0070] (1) The model exhibits excellent predictive performance on the training set, such as Figure 4 As shown in (a), the area under the receiver operating characteristic (AUC) was 0.9935 (95% confidence interval: 0.9834–1.0000), with an overall accuracy of 97.67%, and both sensitivity and specificity reached 97.67%. The predictive power of the prediction model including 21 metabolite combinations (21-PMR) was better than that of a single metabolite (AUC: 0.9935 vs 0.716), indicating that the multi-metabolite combination model has better predictive value.

[0071] Subsequently, the predictive model was applied to the test set, with an AUC of 0.897 (95% CI: 0.751-1.0000, accuracy: 90.91%, sensitivity: 92.31%, specificity: 88.89%). Figure 4 As shown in (b).

[0072] according to Figure 3 Significant differences were observed between the R and NR groups in 21 metabolites. Sixteen metabolites were significantly upregulated in the R group: thymidine deoxyribonucleoside, thiolysine ketoimine, N6,N6-dimethyllysine, tryptophan, hydroxyprogesterone, hexanocarnitine, di-hypo-γ-linolenic acid, aspartic acid-tyrosine dipeptide, 15,16-dihydroxyoctadecadienoic acid, 113,14-dihydroprostaglandin F-1a, 8,9-dihydroxyeicosatetrienoic acid, 4-hydroxy-2-oxoglutarate, 3-hydroxynonanoylcarnitine, 2-aminoadenosine, and 3-hydroxy-cis-5-tetradecanoylcarnitine. Five metabolites were significantly downregulated in the R group, including corticosterone, coenzyme Q2, 12S-hydroxyheptadecadienoic acid, 4-pyridoxine, and 3-hydroxydecanoylcarnitine. Key metabolites that contribute significantly to the production of prostaglandins include 13,14-dihydroprostaglandin F-1a, 15,16-dihydroxyoctadecadienoic acid, 4-hydroxy-2-oxoglutaric acid, 4-pyridoxic acid, 3-hydroxydecanoylcarnitine, and 12S-hydroxyheptadecadienoic acid.

[0073] (2) Based on the regression coefficients of each specific metabolite in the model, calculate the efficacy risk score for each patient; using the median efficacy risk score as the cutoff, divide patients into a high benefit-risk group (high) and a low benefit-risk group (low); assess the distribution characteristics of different benefit-risk groups in the treatment response group, and the results are as follows: Figure 5As shown.

[0074] The predictive model based on efficacy risk score showed good predictive power in the test set, with an AUC of 0.891 (95% CI: 0.826-0.957).

[0075] Benefit-risk stratification results showed that the majority of treatment-unresponsive (NR) patients were concentrated in the high-benefit-risk group, while treatment-responsive (R) patients were mainly distributed in the low-benefit-risk group, such as... Figure 5 As shown in (b), this is consistent with the predictive power of the efficacy risk score. Further analysis of the RECIST efficacy evaluation results of the two groups revealed that the proportion of patients with disease progression (PD) and stable disease (SD) was significantly higher in the high-benefit-risk group than in the low-benefit-risk group, while the proportion of patients with partial remission (PR) was more pronounced in the low-benefit-risk group. Figure 5 As shown in (c). The above results indicate that the predictive model can effectively identify patient groups that may benefit from treatment optimization.

[0076] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A metabolomics-based method for predicting the efficacy of non-invasive prenatal testing (NAIC) in gastric cancer, characterized in that, The specific steps are as follows: S1: Obtain plasma samples from gastric cancer patients before they receive neoadjuvant immunochemotherapy; detect the abundance of metabolites in the plasma samples to obtain baseline plasma metabolomics data; S2: After the gastric cancer patients described in step 1 receive neoadjuvant immunochemotherapy, the gastric cancer patients are divided into a treatment-responsive group and a treatment-non-responsive group based on imaging assessment. S3: Compare baseline plasma metabolomic data between the treatment-responsive group and the treatment-unresponsive group, and screen out metabolites that show significant differences between the two groups as differential metabolites; S4: Using the abundance data of the differential metabolites as features and the treatment response as a variable, a LASSO regression model is used for training, and the optimal regularization parameter is determined through cross-validation to construct an efficacy prediction model; the efficacy prediction model consists of several specific metabolites and their corresponding regression coefficients. S5: Detect the abundance data of the specific metabolites in the plasma sample to be predicted; input the abundance data of the specific metabolites into the efficacy prediction model and output the efficacy risk score; The efficacy risk score is compared with a risk threshold to predict the response to neoadjuvant immunochemotherapy.

2. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, The metabolite abundance data in step S1 is standardized; the standardization is performed using the Z-score method.

3. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, In step S3, principal component analysis and orthogonal partial least squares discriminant analysis are used to screen candidate differential metabolites.

4. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, The method for determining the optimal regularization parameter in step S4 is as follows: 10-fold cross-validation is used to optimize the parameters of the LASSO regression model, and the optimal regularization parameter λ is determined by the λ.1se rule.

5. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, The specific metabolites mentioned in step S4 include one or more combinations of 8,9-dihydroxyeicosatotrienoic acid, 13,14-dihydroprostaglandin F-1a, 12S-hydroxyhexadecanosatotrienoic acid, thiolysine ketoimine, 4-pyridoxine, 3-hydroxydecanoylcarnitine, coenzyme Q2, 4-hydroxy-2-ketoglutarate, corticosterone, 15,16-dihydroxyoctadecanodienoic acid, 2-aminoadenosine, 3-hydroxy-cis-5-tetradecanoylcarnitine, 3-hydroxynonanoylcarnitine, aspartic acid-tyrosine dipeptide, di-high-gamma-linolenic acid, hexanoylcarnitine, hydroxyprogesterone, hydroxytryptamine, N6,N6-dimethyllysine, S-adenosine homocysteine, and thymidine deoxyribonucleoside.

6. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, The formula for the efficacy prediction model described in step S4 is as follows: ； In the formula: P is the efficacy risk score; Let be the regression coefficient of the i-th specific metabolite; Here is the standardized abundance data for the i-th metabolite.

7. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, The risk threshold is calculated as follows: the baseline plasma metabolome data of gastric cancer patients obtained in step S1 is used as the training set; the training set is input into the efficacy prediction model in step S5, and the efficacy risk score is output; the median of all efficacy risk scores is used as the risk threshold.

8. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, In step S5, if the efficacy risk score of the plasma sample to be predicted is less than the risk threshold, the gastric cancer patient is determined to have a high risk of benefit from neoadjuvant immunochemotherapy; otherwise, the risk of benefit is low.

9. The method for predicting the efficacy of NAIC in gastric cancer based on metabolomics according to claim 1, characterized in that, The plasma sample to be predicted is a plasma sample from a gastric cancer patient before receiving neoadjuvant immunochemotherapy.

10. A specific metabolite for predicting the efficacy of NAIC in gastric cancer, characterized in that, The specific metabolite is one or more of the following combinations: 8,9-dihydroxyeicosatotrienoic acid, 13,14-dihydroprostaglandin F-1a, 12S-hydroxyhexadecanotrienoic acid, thiolysine ketoimine, 4-pyridoxine, 3-hydroxydecanoylcarnitine, coenzyme Q2, 4-hydroxy-2-ketoglutarate, corticosterone, 15,16-dihydroxyoctadecanodienoic acid, 2-aminoadenosine, 3-hydroxy-cis-5-tetradecanoylcarnitine, 3-hydroxynonanoylcarnitine, aspartic acid-tyrosine dipeptide, di-high-gamma-linolenic acid, hexanoylcarnitine, hydroxyprogesterone, hydroxytryptamine, N6,N6-dimethyllysine, S-adenosine homocysteine, or thymidine deoxyribonucleoside.

Images