A lung nodule metabolic marker, a screening method thereof, a construction method of a screening model, and an application thereof

Abstract

The application discloses a lung nodule metabolic marker and a screening method and a construction method and application of a screening model. Specifically disclosed are the application of metabolic markers and / or reagents for detecting the metabolic markers in the preparation of products for lung nodule screening; the metabolic markers include hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate and bethanidine. The application further discloses a method for constructing a lung nodule screening model. The metabolic markers and the lung nodule screening model constructed by using the metabolic markers can effectively screen out individuals with lung nodules from healthy people, have very high sensitivity and good specificity, are fast, accurate, non-invasive and non-radiation, are suitable for wide promotion, can be used as a pre-screening tool of LDCT, are used for the preliminary screening of lung nodules of a large-scale population, and have important significance for the early discovery of lung cancer and the improvement of the prognosis of patients.

Description

Technical Field

[0001] This invention belongs to the field of clinical laboratory diagnostic technology, specifically relating to a metabolic biomarker for pulmonary nodules and its screening method, a screening model construction method, and its application. Background Technology

[0002] Lung cancer is the leading cause of cancer-related death worldwide. The five-year survival rate after surgery for early-stage lung cancer (stage I) can reach 77%-84%, while the five-year survival rate for late-stage lung cancer (stage IV) is only 18%-21%. Therefore, early detection and timely intervention are crucial for improving the prognosis of lung cancer patients. Pulmonary nodules are focal, round or oval, solid or subsolid shadows with increased density, not exceeding 3 cm in diameter, visible on lung imaging. They can be solitary or multiple. With the widespread adoption of low-dose spiral CT (LDCT) screening and increased public awareness of health checkups, the detection rate of pulmonary nodules has risen rapidly. Timely detection of pulmonary nodules is crucial for the early detection of lung cancer—the 10-year survival rate after surgical radical resection of early-stage lung cancer can reach 90%, with a significantly better prognosis than that of patients with intermediate or late-stage disease. Therefore, efficient screening for pulmonary nodules in a large population is a key first step in achieving "early detection, early diagnosis, and early treatment" of lung cancer.

[0003] Currently, LDCT is the main imaging method for detecting pulmonary nodules. However, LDCT has the following significant limitations as a large-scale population screening tool: (1) The examination cost is high, making it difficult to achieve large-scale coverage of the entire population; (2) It requires professional CT scanning equipment and senior radiologists for interpretation, which severely limits its accessibility in primary healthcare institutions and community health check centers; (3) There is a risk of cumulative radiation exposure, making it unsuitable for repeated screening of the same population in a short period of time; (4) Current international guidelines only recommend LDCT screening for high-risk groups (such as long-term heavy smokers, those aged ≥50 years, etc.), however, a cohort study of 10,000 people published in JAMA in 2025 showed that if screening is carried out entirely according to the high-risk criteria of the current guidelines, the lung cancer missed detection rate can be as high as 44%-81%, and a large number of pulmonary nodule patients in the medium and low-risk populations cannot be covered; (5) Screening compliance is low, and some people fail to participate in screening due to concerns about radiation, cumbersome examination procedures, or poor accessibility at the primary level.

[0004] Therefore, if a non-invasive blood test method that requires only a small amount of peripheral blood can be developed as a pre-LDCT triage screening method, it is expected to bring significant improvements in the following aspects: (1) greatly expand the population coverage of lung nodule screening, so that screening is no longer limited by CT equipment and professional radiologists; (2) significantly reduce the overall screening cost and reduce a large number of unnecessary CT scans and their cumulative radiation exposure; (3) improve the screening compliance of examinees and promote the hierarchical management of lung nodule screening and early diagnosis and treatment of lung cancer.

[0005] Metabolomics is an important component of systems biology. Through comprehensive detection and analysis of small-molecule metabolites in an organism, it can sensitively reflect the physiological and pathological state of an individual. Cellular metabolic reprogramming is one of the prominent characteristics of tumors, and circulating metabolites can serve as "liquid biopsy biomarkers" reflecting pathological changes in the body. However, the application value of aqueous metabolites (including polar small molecules such as amino acids, nucleotides, organic acids, and acylcarnitines) in pulmonary nodule screening has not been fully explored. Therefore, developing a non-invasive pulmonary nodule screening method based on plasma aqueous metabolites has significant clinical and socioeconomic value. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to perform rapid, sensitive, specific and accurate screening for lung nodules. The technical problem to be solved is not limited to the technical subject matter described herein. Other technical subjects not mentioned herein can be clearly understood by those skilled in the art through the following description.

[0007] To address the aforementioned technical problems, the present invention first provides the application of metabolic biomarkers and / or reagents for detecting said metabolic biomarkers in any of the following: A1) Application in the preparation of products for pulmonary nodule screening; A2) Application in the preparation of products for the diagnosis or auxiliary diagnosis of pulmonary nodules; A3) Application in the preparation of products for predicting the risk of pulmonary nodules; A4) Application in constructing a pulmonary nodule screening model; The metabolic markers include hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate (DHEA-S), and cotinine.

[0008] Furthermore, the metabolic markers may be hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine in the subject's plasma.

[0009] Furthermore, among the metabolic markers, the concentrations of hypoxanthine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine in the plasma of patients with pulmonary nodules are lower than their concentrations in the plasma of healthy individuals; while the concentration of L-ornithine in the plasma of patients with pulmonary nodules is higher than its concentration in the plasma of healthy individuals.

[0010] Furthermore, the application includes using data on the concentrations of hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine in the plasma of known patients with pulmonary nodules and healthy subjects as training samples, employing statistical analysis methods (such as support vector machine analysis, random forest analysis, decision tree analysis, logistic regression analysis, and neural network analysis) to establish a pulmonary nodule screening model, and screening of subjects for pulmonary nodules based on the model.

[0011] The hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine are aqueous metabolites.

[0012] Furthermore, the reagent may include reagents for detecting the concentration of the biomarker in plasma by methods such as mass spectrometry, chromatography, spectroscopy, nuclear magnetic resonance, liquid chromatography-mass spectrometry, or gas chromatography-mass spectrometry.

[0013] The present invention also provides a kit for screening pulmonary nodules, the kit comprising reagents for detecting the metabolic markers described herein.

[0014] Furthermore, the reagents may include standards of the metabolic markers described herein.

[0015] Furthermore, the reagent may also include an isotopic internal standard of the metabolic biomarker described herein.

[0016] Furthermore, the reagent may also include an extraction solvent and a reconstitution solution.

[0017] Furthermore, the isotopic internal standards of the metabolic markers described herein may include: [²H4]-hypoxanthine, [²H7]-L-ornithine hydrochloride, [²H3]-decanoyl-L-carnitine hydrochloride, [²H6]-dehydroepiandrosterone sulfate sodium salt, and [methyl-²H3]-DL-cotinine.

[0018] This invention also provides the application of the kit described herein in the preparation of products for screening pulmonary nodules.

[0019] This invention also provides a method for screening metabolic biomarkers for pulmonary nodules. The method includes: performing non-targeted metabolomics analysis and data preprocessing on plasma samples from patients with pulmonary nodules and healthy subjects; identifying combinations of five plasma aqueous phase metabolic biomarkers—hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine—through support vector machine feature screening and primary and secondary mass spectrometry library searches. The product of the screening method is the chemical identity of the above metabolic biomarkers and their combinations.

[0020] The present invention also provides a method for constructing a pulmonary nodule screening model, the method comprising using data on the known concentrations of hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate and cotinine in the plasma of pulmonary nodule patients and healthy subjects as training samples, and employing statistical analysis methods to establish a pulmonary nodule screening model.

[0021] The statistical analysis methods described above may include support vector machine analysis, random forest analysis, decision tree analysis, logistic regression analysis, and neural network analysis.

[0022] Furthermore, the method may include the following steps: S1) Obtain peripheral blood plasma samples from patients with pulmonary nodules and healthy subjects; S2) The concentrations of five metabolic markers—hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine—in the plasma sample were detected using the targeted multiple reaction monitoring (MRM) method to obtain targeted absolute quantitative concentration data. S3) The target absolute quantitative concentration data is used as the input feature of the support vector machine classifier. The target MRM detection method remains fixed, and only the support vector machine classifier is trained to obtain the trained support vector machine model, which is the lung nodule screening model.

[0023] The targeted multiple reaction monitoring (MRM) method can be performed using liquid chromatography-mass spectrometry in multiple reaction monitoring (MRM) mode.

[0024] In this document, the products may include reagents, kits, detection chips, test strips, test cards, immunosensors, or devices.

[0025] The present invention also provides a device for screening pulmonary nodules, the device comprising: The data receiving module is used to receive data on the concentrations of hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine in plasma samples of test subjects from at least one terminal. The data processing module is used to input the data into the lung nodule screening model; the lung nodule screening model is a lung nodule screening model constructed using any of the methods described herein; The data output module is used to output the lung nodule screening results to at least one client.

[0026] The data processing module can be used to perform the following operations: inputting the data received by the data receiving module into the lung nodule screening model, obtaining the output value through the lung nodule screening model, thereby providing an indication of whether the subject to be tested is a lung nodule patient or a healthy subject.

[0027] The output value can be the class label of the sample (which category the sample belongs to), such as {0, 1} or {positive, negative}. The output value can be mapped to a numerical value by mapping "positive" to 1 and "negative" to 0.

[0028] Furthermore, the present invention provides the following methods: 1. A method for screening metabolic biomarkers of pulmonary nodules (non-targeted discovery); 2. A method for absolute quantitative analysis of targeted metabolomics based on the metabolic biomarkers; 3. A method for constructing a pulmonary nodule screening model (targeted data training).

[0029] I. Screening methods for metabolic biomarkers in pulmonary nodules (non-targeted detection) This invention also provides a screening method for the metabolic biomarkers (hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine) described herein. This screening method is used to identify and pinpoint the above five metabolic biomarkers from aqueous plasma metabolites. The products are the chemical identities of the biomarkers and their combinations, rather than a lung nodule screening model. The screening method may include the following steps: 1) Non-targeted metabolomics analysis, retrieval, and data preprocessing were performed on plasma samples from patients with pulmonary nodules and healthy individuals to obtain two-dimensional matrix data; wherein, the patients with pulmonary nodules refer to individuals whose lungs are found to have nodular lesions (diameter ≤3cm) through imaging examination, including patients whose lung cancer or benign lesions are confirmed by postoperative pathology. 2) The above two-dimensional matrix data was learned using the machine learning support vector machine (SVM) algorithm. The patients with pulmonary nodules were treated as a whole as the nodule group and binary classification was performed with the healthy control group. 3 / 4 of the sample data was randomly used as the training set and 1 / 4 as the test set. The model was randomly iterated 2000 times. By statistically analyzing the average accuracy of the final support vector machine model, it was determined that the support vector machine model can effectively classify the metabolome data of patients with pulmonary nodules and healthy people. 3) Based on the support vector machine model obtained above, feature selection is performed using a greedy algorithm based on machine learning. The feature weights of the support vector machine model are scored, and the features with the highest weights are continuously accumulated in the forward feature selection method to form the model to be tested. The classification accuracy of the model is evaluated to show the classification performance of different models, and finally the relatively optimal number of features and combination method are displayed. The criterion for selecting the optimal number of features and combination method is: the model accuracy no longer increases when the number of features is increased. 4) The optimal features obtained from the above screening, i.e. the target differential metabolites, are optimized and screened based on mass spectrometry. MS-Dial software is used to screen based on the chromatographic peak shape and secondary mass spectrum (MS / MS) data quality to remove metabolite features with poor or mismatched spectral quality, as well as features corresponding to the same metabolic marker in positive and negative ion modes, and to obtain potential metabolic markers. 5) Based on the primary and secondary mass spectrometry information of the aforementioned potential metabolic biomarkers, the molecular weight and molecular formula of the biomarkers were inferred and compared with the spectral information in the MassBank metabolite database to identify the metabolites. Ultimately, five plasma aqueous phase metabolic biomarkers were identified: hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine. The non-targeted screening process was only used to determine the chemical identity of the aforementioned metabolic biomarkers; its product was the biomarker combination itself, not a lung nodule screening model.

[0030] Furthermore, the non-targeted metabolomics analysis, retrieval, and data preprocessing may include the following steps: (A) Collect plasma samples from patients with pulmonary nodules and healthy individuals respectively as analytical samples; (B) Untargeted metabolomics analysis was performed on each analytical sample using high performance liquid chromatography-mass spectrometry to obtain the original metabolic fingerprint of each plasma sample; (C) The original metabolic fingerprint spectrum was processed using MS-Dial software (version 4.70), including retention time correction, peak identification, peak matching, peak alignment, noise filtering, and data normalization, to obtain a two-dimensional matrix with metabolite information in each row and analytical samples in each column; and metabolite peak identification and peak area integration were performed on the two-dimensional matrix, including isotope peaks, adducts, and fragment ions, for further machine learning.

[0031] In this study, the analytical samples (plasma samples) and standard quality controls can undergo the following pretreatment steps before injection: Mix the analytical sample or standard quality control with the extraction solvent, centrifuge, collect the supernatant, evaporate to dryness to obtain a dry powder, reconstitute with a rehydration solution, centrifuge, collect the supernatant, and store at -80℃ for LC-MS detection. Specifically, this may include the following steps: 1) Use a pipette to draw 50 μL of the analytical sample or standard quality control and place it into a 0.6 mL EP tube; 2) Add 200 μL of extraction solvent methanol (pre-cooled at -20℃ for more than 2 h) and mix. Vortex for 30 seconds, shake for 20 minutes at 4℃ and 1200 rpm on a horizontal shaking mixer, then centrifuge at 12000 rpm at 4℃ for 20 minutes, collect the supernatant, and evaporate to dryness in a vacuum concentrator. 3) After reconstitute the dry powder with acetonitrile-water reconstitution solution (1:1, V / V), centrifuge at 12000 rpm for 20 minutes at 4°C in a high-speed centrifuge to obtain the supernatant; 4) Transfer the supernatant from step 3) into an LC-MS vial and store it at -80°C for LC-MS detection.

[0032] When analyzing each sample using LC-MS plasma non-targeted metabolomics, the preferred chromatographic column used in liquid chromatography is a Waters XBridge BEH Amide column, preferably 100 × 2.1 mm, 2.5 μm in diameter; the preferred injection temperature is 4℃, and the preferred injection volume is 5 μL; the mobile phase contains two solvents, A and B, where A is an aqueous solution of acetonitrile containing 5 mM ammonium acetate, the aqueous solution containing 5% acetonitrile and 95% water by volume; and B is pure acetonitrile; the preferred gradient elution conditions are: 95% B for 0–2 min, gradually decreasing from 95% B to 50% B for 2–15 min, maintaining 50% B for 15–18 min, gradually increasing from 50% B to 95% B for 18–19 min, and then maintaining 95% B for 4 min; the flow rate is 0.35 mL / min; and the column temperature is 30℃. For mass spectrometry detection, a quadrupole-electrostatic field orbital trap high-resolution mass spectrometer Q-Exactive (Thermo Fisher Scientific) is preferred, and an electrospray ion source is used to switch between positive ion mode ESI+ and negative ion mode ESI- for detection. The mass-to-charge ratio range of the spectral data acquisition is 60–800 m / z, and the acquisition mode is data-dependent (DDA).

[0033] The random forest signal correction algorithm based on quality control samples (QC) (QC-RFSC, statTarget package) can be used to correct non-targeted metabolomics data; the missing parts in the data are filled with 1 / 2 minimum values; and the data are standardized and logarithmically transformed to base 2 using the normalizeBetweenArrays function in the limma package.

[0034] In this invention, before modeling with Support Vector Machines (SVMs), all data undergoes L2 normalization to ensure consistent scaling across the dimensions of each sample. When constructing the SVM classification model, the modeling parameter C=5. During the greedy algorithm selection process, the focus is on the top 100 features with the highest SVM weights. A forward feature selection method is employed, evaluating each candidate feature combination through 500 random four-fold cross-validations (a total of 100 × 500 = 50,000 evaluations) to ensure the local optimum of the selected feature combinations.

[0035] For the five metabolic biomarkers identified in the non-targeted screening, a targeted quantitative detection method based on multiple reaction monitoring (MRM) was further established (including ion pairs, isotope internal standards, and standard curves, detailed below). This targeted detection method, once established, was fixed and used as the unified method for absolute quantification of all subsequent cohort samples. Based on this, plasma samples from patients with pulmonary nodules and healthy subjects were detected using the described targeted method. The obtained targeted absolute quantitative concentration data were used as the training set to train only the support vector machine classifier, resulting in a pulmonary nodule screening model.

[0036] II. Targeted Metabolomics Absolute Quantification Method Based on the Metabolic Biomarkers The targeted metabolomics absolute quantification method based on the metabolic biomarkers described in this article may include the following steps: 1) Preparation of standards: (a) Prepare standard stock solutions and isotope internal standard stock solutions for each metabolic marker; (b) Prepare mixed standards with corresponding concentration gradients: the standard working calibration solution is obtained by serial dilution of the combined standard stock solutions, and calibrates at 12 concentration gradients (S1 to S12); (c) Prepare standard quality control products: defined as high-concentration quality control (HQC), medium-concentration quality control (MQC) and low-concentration quality control (LQC), respectively.

[0037] 2) Sample preparation: (a) Preparation of extraction solution containing isotope internal standards: Mix the five isotope internal standard stock solutions obtained in step 1) with methanol pre-cooled at -20℃ for more than 2 h and shake to mix well; (b) Extraction of samples: Pipette 20 μL of analytical sample / standard working calibration sample / standard quality control sample into a 0.6 mL EP tube, add 80 μL of the extraction solution containing internal standards prepared in step (a) to each sample, vortex for 30 seconds, shake horizontally for 15 minutes, centrifuge at 4℃ and 13000 rpm for 20 minutes, collect the supernatant, and evaporate to dryness in a vacuum concentrator; (c) Reconstitution of samples: Reconstitute the dry powder with 200 μL of acetonitrile-water reconstitution solution (3:1, V / V), centrifuge at 4℃ and 12000 rpm for 20 minutes in a high-speed centrifuge, and transfer the supernatant into an LC-MS injection bottle.

[0038] 3) LC-MS / MS analysis: Targeted metabolomics analysis of metabolic biomarkers was performed on each analytical sample, standard working calibration sample, or standard quality control sample using liquid chromatography-mass spectrometry to obtain the targeted metabolomics profile of each sample.

[0039] 4) Data Analysis: MultiQuant software was used to process and quantify the targeted metabolome map, resulting in a two-dimensional matrix where each row contains metabolite information and each column contains biomarkers of the analyzed samples. The standard curve was plotted using the internal standard method, by comparing the peak area of ​​the standard with the area of ​​its corresponding isotope internal standard. After fitting, a linear equation and correlation coefficient R² were obtained, and the quantitative concentration of the corresponding metabolite was calculated.

[0040] In this invention, the targeted metabolomics analysis can be performed using an AB Sciex high-performance liquid chromatography-mass spectrometry system (Jasper HPLC system / QTRAP 4500MD). The preferred chromatographic column used for liquid chromatography is a Waters XBridgeBEH Amide column with dimensions of 100 × 2.1 mm and a diameter of 2.5 μm. Mass spectrometry analysis employs multiple reaction monitoring (MRM) mode. The ionization temperature is set to 500°C, the spray voltage to 5500V, the injection voltage to 10V, and the collision chamber exit voltage to 11V; the collision gas and curtain gas are set to 8 and 35, respectively; and the ion source gases GS1 and GS2 are set to 55 and 60, respectively. The entire targeted detection process takes only 16 minutes.

[0041] III. Construction Method of Lung Nodule Screening Model (Targeted Data Training) The present invention also provides a method for constructing a pulmonary nodule screening model, the method comprising using data on the known concentrations of hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate and cotinine in the plasma of pulmonary nodule patients and healthy subjects as training samples, and employing statistical analysis methods to establish a pulmonary nodule screening model.

[0042] The statistical analysis methods described above may include support vector machine analysis, random forest analysis, decision tree analysis, logistic regression analysis, and neural network analysis.

[0043] Furthermore, the method may include the following steps: S1) Obtain peripheral blood plasma samples from patients with pulmonary nodules and healthy subjects; S2) The concentrations of five metabolic markers—hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine—in the plasma sample were detected using the targeted multiple reaction monitoring (MRM) method to obtain targeted absolute quantitative concentration data. S3) The target absolute quantitative concentration data is used as the input feature of the support vector machine classifier. The target MRM detection method remains fixed, and only the support vector machine classifier is trained to obtain the trained support vector machine model, which is the lung nodule screening model.

[0044] The patients with pulmonary nodules mentioned in step S1) can be those who have been definitively diagnosed with pulmonary nodules.

[0045] Further, in step S2), the targeted MRM map can be processed using MultiQuant software to extract the peak areas of each metabolic biomarker and its corresponding isotopic internal standard. The peak area ratio of each biomarker to its corresponding isotopic internal standard is calculated, resulting in a two-dimensional matrix where each row represents metabolite information and each column represents the analyzed samples. This matrix is ​​used for subsequent machine learning modeling. During model training, the model can be evaluated to determine its performance. Evaluation metrics typically include ROC curves (AUC values) and accuracy.

[0046] The concentration data obtained in step S2 can be normalized using L2.

[0047] Furthermore, the method may include the following steps: 1) Peripheral blood plasma samples were collected from patients with pulmonary nodules and healthy subjects. The concentrations of the metabolic markers described in this paper (5 metabolites) in each sample were determined using liquid chromatography-mass spectrometry in multiple reaction monitoring (MRM) mode to obtain the training set. 2) The feature vector composed of the concentrations of the five metabolites in each sample is L2 normalized according to equation (1): ; 3) Using the normalized feature vector as input and whether the patient has pulmonary nodules as the classification label (+1 for nodules, -1 for healthy), the radial basis function kernel support vector machine algorithm is used for binary classification training. The kernel function is shown in equation (2): ; The classification decision function is shown in equation (3): ; In equations (2) and (3), x i x j Let α be the feature vector of the training samples, γ be the kernel function bandwidth parameter, and α be the feature vector of the training samples. i For support vector weights, y i b represents the class labels of the training samples, and b is the bias term. 4) Use four-fold cross-validation and repeat iteratively for no less than 2000 times. Take the α and b parameters corresponding to the highest average classification accuracy as the model parameters to obtain the lung nodule screening model.

[0048] Furthermore, in step 3), the soft margin penalty coefficient C=5, and γ is taken as the default value of scikit-learn 1 / (n·Var(x)), where n is the number of features and Var(x) is the variance of the training set features.

[0049] Further, calculate f(x) according to formula (3); when f(x)>0, the output can be "suggest the presence of pulmonary nodules and recommend further imaging examinations", and when f(x)≤0, the output can be "no risk of pulmonary nodules found".

[0050] Further, the plasma sample volume in step 1) is 20 μL, and the determination may include: mixing the plasma sample with pre-cooled methanol extract containing an isotope internal standard at a volume ratio of 1:4, vortexing, centrifuging at 13000 rpm for 20 minutes at 4°C, evaporating the supernatant to dryness, reconstituted with acetonitrile water, and then injecting the sample; the chromatographic column used is a Waters XBridge BEH Amide column (100×2.1 mm, 2.5 μm), and the mass spectrometry is performed in multiple reaction monitoring mode with switching detection of positive and negative ions.

[0051] The method for constructing the pulmonary nodule screening model described in this invention is simple and exhibits high sensitivity (up to 90.0%) and good specificity in screening for pulmonary nodule patients from healthy individuals. This application specification provides the training samples, test samples, model, and model parameters included in the machine learning model. Those skilled in the art can verify or apply this method and screening model using the combination of screening biomarkers according to the specification.

[0052] When using the lung nodule screening model of this invention, the test can be completed by collecting only 20 μL of peripheral blood plasma, which is convenient, fast, non-invasive, radiation-free, and has high sensitivity and specificity. The model output can provide screening suggestions such as "further imaging examination recommended" or "no abnormalities found," and can be used as a pre-screening tool for LDCT, which has good clinical application value. The comparison between this invention and existing lung nodule / lung cancer screening programs is shown in the table below.

[0053]

[0054] This invention is not a substitute for LDCT, but rather a complement. This invention is positioned as a pre-screening tool for LDCT: in large-scale population screening scenarios, the blood testing method of this invention is used to quickly and economically screen high-risk individuals who may have pulmonary nodules, and then recommends them for LDCT imaging examination for confirmation. This achieves a tiered screening strategy of "blood pre-screening → imaging confirmation," significantly expanding screening coverage, reducing overall screening costs, and minimizing unnecessary CT scans and radiation exposure.

[0055] Compared with the prior art, the present invention has the following advantages: (1) Minimal blood volume required and rapid detection: Only 20 μL of peripheral blood plasma is needed, and the entire targeted mass spectrometry detection process can be completed in 16 minutes, making it suitable for high-throughput clinical applications and large-scale population screening.

[0056] (2) Non-invasive and radiation-free: This invention only requires the collection of a small amount of venous blood, with no risk of radiation exposure, and completely avoids the cumulative radiation problem caused by LDCT. It is especially suitable for people who need regular screening and monitoring as well as low- to medium-risk groups.

[0057] (3) High screening sensitivity: The lung nodule screening model constructed by the combination of the five metabolic biomarkers described in this invention achieved an AUROC of 0.921 and a sensitivity of 90.0% in the independent validation cohort test set, which can effectively screen out lung nodule patients from healthy people with a low false negative rate. Currently, there is no universally accepted hematological method that can be used for lung nodule screening, and this invention fills the gap in this field.

[0058] (4) It can be used as a pre-screening tool for LDCT, with significant economic benefits: This invention is positioned as a pre-screening method for LDCT screening, rather than a replacement for LDCT. By first using the blood testing method of this invention to screen high-risk individuals who may have pulmonary nodules from a large population, and then recommending them for LDCT imaging confirmation examination, the screening coverage can be significantly expanded, the overall screening cost can be reduced, and unnecessary CT scans and radiation exposure can be reduced, realizing a hierarchical management strategy for pulmonary nodule screening. Compared with directly performing LDCT screening on the entire population, the pre-screening scheme of this invention has significant advantages in terms of economy, accessibility and safety.

[0059] (5) Multicenter independent validation with reliable results: The screening biomarker of the present invention has undergone dual external validation in a three-center cohort of 300 cases and an independent cohort of 214 cases, which confirms its robustness and reproducibility across centers and populations.

[0060] (6) Absolute quantification using isotope internal standard, accurate and reliable method: This invention uses isotope-labeled internal standard method for absolute quantification, effectively correcting for matrix effects and extraction losses, and the standard curve has good linearity (r 2 (All values ​​> 0.98), ensuring the accuracy and comparability of the quantitative results.

[0061] In summary, the metabolic biomarkers described in this invention can effectively screen individuals with pulmonary nodules from healthy populations. They exhibit extremely high sensitivity and good specificity, require only 20 μL of plasma, and can be completed in 16 minutes. They are fast, accurate, non-invasive, and radiation-free, making them suitable for widespread application. They can be used as a pre-screening tool for LDCT for preliminary screening of pulmonary nodules in large populations, which is of great significance for the early detection of lung cancer and improving patient prognosis.

[0062] Terminology Definition In this invention, unless otherwise stated, the scientific and technical terms used herein have the meanings commonly understood by those skilled in the art. Furthermore, to better understand this invention, definitions and explanations of relevant terms are provided below.

[0063] The term "data receiving module" generally refers to a functional module used to obtain relevant data. In this invention, the data receiving module may refer to a functional module used to obtain data on the concentrations of hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine in the plasma sample of the subject being tested.

[0064] The term "data processing module" typically refers to a functional module responsible for data processing (including data preprocessing). In this invention, the data processing module can be used to process and / or analyze the concentration data of five metabolites—hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine—in the plasma samples of test subjects, and to differentiate between individuals with pulmonary nodules and healthy individuals based on the analysis results. The data processing module can process and analyze the data based on machine learning models.

[0065] The term "data output module" generally refers to a functional module used to display results information to the user. In this invention, the data output module may refer to a functional module used to display lung nodule screening results. The data output module may include, but is not limited to, a display, printer, projector, monitor, audio output device, etc.

[0066] The term "device" generally refers to a system comprising modules or units operatively interconnected to enable computation. It can be implemented through software (such as computer programs, mobile applications, web applications, and standalone applications), hardware (such as processors or memory), or a combination thereof, comprising one or more modules or units. Devices encompass a wide range of forms, such as server computers, desktop computers, laptop computers, minicomputers, netbook computers, netpad computers, set-top computers, streaming media devices, mobile smartphones, handheld computers, tablet computers, and personal digital assistants.

[0067] The term "machine learning model" typically refers to a computational model or algorithm. Machine learning models learn and analyze input data to perform tasks such as prediction, classification, recognition, or decision-making. The learning process of a machine learning model can be based on statistical principles and data pattern recognition. It uses training datasets to adjust model parameters and optimize the model to improve its predictive or inference capabilities. It can predict or classify new data by learning the mapping relationship between features and labels in the training data. In machine learning, a "label" typically refers to the true category or target value of a data sample, serving as a "reference answer" during model training. In classification tasks, a label indicates which category a data sample belongs to. In regression tasks, a label can be a continuous target value or a real-valued target value to be predicted. A "feature" typically refers to an attribute or characteristic describing a data sample. In machine learning, features are input variables used to train the model and can be any type of data, such as numerical or categorical. "Training" typically refers to the model learning process, which uses labeled training data to adjust the model's parameters so that the model can predict labels with maximum accuracy. During training, the model continuously optimizes its parameters to approximate the true data distribution. Machine learning models can be trained and optimized using various algorithms and techniques through supervised learning, unsupervised learning, or reinforcement learning. Exemplary machine learning models may include support vector machine models, random forest models, linear discriminant analysis models, recursive feature elimination models, logistic regression models, neural network models, CART models, flextree models, LART models, MART models, k-nearest neighbor classification models, clustering analysis models, principal component analysis models, Bayesian classification models, and hidden Markov models, etc. In one or more embodiments of this invention, a support vector machine model is used to establish a lung nodule screening model.

[0068] The terms "module" and "unit" generally have the same meaning and are used interchangeably.

[0069] The term "comprising" is not intended to be restrictive, but rather inclusive and implies the presence of other elements besides those listed, and can be interpreted as "including but not limited to". The term "comprising" also encompasses the terms "consisting of" and "substantially consisting of". In this document, the terms "comprising" and "including" are used interchangeably. Attached Figure Description

[0070] Figure 1 The PLS-DA score map for non-targeted metabolomics analysis provided by this invention is shown in Figure A, where A is the PLS-DA map of QC samples in positive ion mode and B is the PLS-DA map of QC samples in negative ion mode, demonstrating good clustering of QC samples and the separation trend between the nodule group and the healthy control group.

[0071] Figure 2 The feature selection score map of the SVM model provided by this invention (forward feature selection based on greedy algorithm); where A is the accuracy score map of the positive ion mode lung nodule screening model, B is the accuracy score map of the negative ion mode lung nodule screening model, the horizontal axis is the cumulative number of features, and the vertical axis is the model accuracy.

[0072] Figure 3 Typical selected ion chromatograms (XIC / MRM chromatograms) and MRM parameter optimization tables for constructing a targeted MRM method using five plasma aqueous phase metabolic markers are provided for this invention.

[0073] Figure 4 Standard curves for five plasma aqueous phase metabolic markers provided by this invention.

[0074] Figure 5 The results of the targeted validation cohort 1 (300 cases) provided by this invention are shown in the figure; the classification efficiency statistics of the SVM model for pulmonary nodule screening (nodule group vs. healthy group) are shown in the figure.

[0075] Figure 6 The key results of the targeted validation cohort 2 (214 cases) provided by this invention are shown in Figure A; A is the hierarchical clustering heatmap (nodule group vs. healthy group), and B is the ROC curve of the test set (AUROC=0.921). Detailed Implementation

[0076] The present invention will now be described in further detail with reference to specific embodiments. The given embodiments are merely illustrative of the invention and not intended to limit its scope. The embodiments provided below can serve as a guide for further improvements by those skilled in the art and do not constitute a limitation on the invention in any way.

[0077] Unless otherwise specified, the experimental methods used in the following examples are conventional methods, performed according to the techniques or conditions described in the literature in this field or according to the product instructions. Unless otherwise specified, the materials and reagents used in the following examples are commercially available.

[0078] Statistical methods: Continuous variables are expressed as mean ± standard deviation, and independent samples t-tests are used for comparisons between groups (Mann-Whitney U test is used if normality is not satisfied); categorical variables are expressed as number of cases (percentage), and chi-square test or Fisher's exact test is used for comparisons between groups; P < 0.05 is considered statistically significant.

[0079] Example 1: Screening of metabolic biomarkers for pulmonary nodules (non-targeted discovery) 1. Research Subjects Plasma samples were collected from patients with pulmonary nodules and healthy individuals for analysis. Patients with pulmonary nodules were defined as individuals with nodular lesions (≤3 cm in diameter) found in the lungs through imaging examinations, including those whose lesions were confirmed malignant (lung cancer) or benign by postoperative pathology. Healthy individuals were those with normal physical examination results. Specifically, in the discovery cohort, the preferred number of patients with pulmonary nodules was 287 (198 with pathologically confirmed lung cancer and 89 with benign pulmonary nodules), and the preferred number of healthy individuals was 74, totaling 361 cases. Inclusion criteria included age over 18 years, chest CT scan showing nodular lesions in the lungs not exceeding 3 cm in diameter, surgical resection with a clear pathological diagnosis, and consent to participate in the study and signing an informed consent form. Exclusion criteria included a history of malignant tumors, preoperative radiotherapy, chemotherapy, or targeted therapy, active infection at the time of plasma collection, abnormal liver or kidney function, and lack of key clinical data.

[0080] The cohort included 361 plasma samples (Table 1), of which 287 were patients with pulmonary nodules (confirmed by postoperative pathology, including 198 cases of lung cancer and 89 cases of benign pulmonary nodules) and 74 healthy controls with normal physical examination results.

[0081]

[0082] 2. Plasma non-targeted metabolomics analysis using liquid chromatography-mass spectrometry (LC-MS) technology All participants had a peripheral blood sample of 4 mL collected after fasting for at least 8 hours using vacuum blood collection tubes containing EDTA. Plasma was separated by two centrifugation steps: first centrifugation at 1600 g for 10 minutes at 4°C, and then centrifugation at 16000 g for 10 minutes. The supernatant, i.e., plasma, was obtained, aliquoted into pre-cooled cryopreservation tubes, and rapidly transferred to an ultra-low temperature freezer at -80°C for storage.

[0083] Plasma samples were collected for the study. After sample pretreatment (50 μL of plasma was added to 200 μL of methanol pre-cooled to -20°C for protein precipitation; the supernatant was evaporated to dryness after vortexing, shaking, and centrifugation, and then reconstituted with acetonitrile and water at a 1:1 ratio), and non-targeted metabolomics analysis was performed using a high-performance liquid chromatography-mass spectrometry (HPLC-MS / MS) system (UltiMate 3000 / Q-Exactive, Thermo Fisher Scientific). Hydrophilic interaction chromatography (HILIC) mode was used with a Waters XBridge BEH Amide column (100 × 2.1 mm, 2.5 μm), and data-dependent acquisition (DDA) was performed under switching between positive and negative ion modes to obtain raw metabolic fingerprints of the samples containing chromatographic and mass spectrometric information.

[0084] Simultaneously, 10 μL of plasma was taken from each analytical sample, mixed thoroughly, and aliquoted to form a quality control (QC) sample. The QC samples were evenly arranged in a sequence of one sample inserted for every 15 analytical samples, used to monitor the stability of the entire analytical process in real time.

[0085] When analyzing the samples using LC-MS plasma non-targeted metabolomics, the chromatographic column used was a Waters XBridge BEH Amide column, with dimensions of 100 × 2.1 mm and a diameter of 2.5 μm. The injection temperature was 4 °C, and the injection volume was 5 μL. The mobile phase contained two solvents, A and B. A was an aqueous solution of acetonitrile containing 5 mM ammonium acetate, which contained 5% acetonitrile and 95% water by volume. B was pure acetonitrile. The chromatographic gradient elution conditions were as follows: 95% B for 0–2 min, gradually decreasing from 95% B to 50% B for 2–15 min, maintaining 50% B for 15–18 min, gradually increasing from 50% B to 95% B for 18–19 min, and then maintaining 95% B for 4 min. The flow rate was 0.35 mL / min, and the column temperature was 30 °C. Mass spectrometry detection was performed using a quadrupole-electrostatic field orbital trap high-resolution mass spectrometer Q-Exactive (Thermo Fisher Scientific), and an electrospray ionization source was used to switch between positive ion mode ESI+ and negative ion mode ESI-. The mass-to-charge ratio range of the spectral data acquisition was 60–800 m / z, and the acquisition mode was data-dependent (DDA).

[0086] 3. Screening and identification of plasma metabolic markers 3.1 Spectral Data Preprocessing Preprocessing was performed using MS-Dial software (version 4.70), including retention time correction, peak identification, peak matching, peak alignment, and noise filtering. The results were then compared and identified with public MS / MS spectral databases such as MassBank. Parameters were set as follows: mass deviation of 0.01 Da for primary mass spectrometry, 0.05 Da for secondary mass spectrometry, and an identification score cutoff of 70%. The processed data yielded a two-dimensional matrix of 791 metabolite peaks in positive ion mode and 481 peaks in negative ion mode.

[0087] The non-targeted metabolomics data were corrected using the QC-RFSC algorithm from the statTarget package. Missing values ​​were filled with 1 / 2 minimum values, and the data were normalized and transformed using the normalizeBetweenArrays function of the limma package.

[0088] 3.2 LC-MS Experimental Quality Control The coefficients of variation for all metabolites in the QC sample were controlled below 30%, indicating good quality control throughout the sample pretreatment and analysis process. (PLS-DA plot) Figure 1 The results showed that the QC samples were tightly clustered, which verified the stability of the detection system and the reliability of the data.

[0089] 3.3 Machine Learning SVM Modeling The Support Vector Machine (SVM) algorithm was used to learn the two-dimensional matrix data obtained from the preprocessed atlas to construct a lung nodule screening model, namely a binary classification model of the nodule group (BL, which combines patients with lung cancer and benign lung nodules into one group) and the healthy control group (H). Three-quarters of the plasma sample data was randomly used as the training set and one-quarter as the test set for learning, and the model was randomly iterated 2000 times.

[0090] Before modeling, the feature vector for each sample is x=(x1, x2, …, x…). n The L2 normalization process is performed, and the normalization formula is shown in equation (1): ; SVM uses a radial basis function (RBF kernel) to map samples to a high-dimensional feature space. The kernel function expression is shown in equation (2): ; In equation (2), x i x j are the normalized feature vectors of the i-th and j-th samples, respectively; γ is the kernel bandwidth parameter, taking the default value of scikit-learn γ=1 / (n·Var(x)), where n is the number of features and Var(x) is the variance of the training set features.

[0091] The SVM classification decision function is shown in equation (3): ; In equation (3), α i The Lagrange multipliers (i.e., support vector weights, obtained by solving a quadratic programming problem) are y. i ∈{+1, -1} represents the class label of the i-th training sample (+1 represents the nodule group, -1 represents the healthy control group), b is the bias term; if f(x)>0, x is determined to belong to the nodule group (output 1), and if f(x)≤0, x is determined to belong to the healthy control group (output 0). The model soft margin penalty coefficient C=5.

[0092] 3.4 Screening and Identification of Plasma Metabolic Markers Based on the SVM model obtained above, metabolic features are screened using a forward feature selection algorithm based on a greedy algorithm. First, a linear kernel SVM is used to model all features, and the absolute value of the weight |w| of each feature j is calculated according to equation (4). j |As a feature importance score: ; In equation (4), w j Let α be the weight of the j-th feature. i y i The meaning is the same as in equation (3), x ij Let be the j-th dimension feature value of the i-th training sample.

[0093] Press |w j Sort the features from largest to smallest and select the top 100 features as the candidate feature set. Then, add the candidate features one by one to the combination of features to be tested in a forward selection manner. After each feature is added, use radial basis kernel SVM and calculate the classification accuracy of the model through 500 random four-fold cross-validation. Evaluate the accuracy according to formula (5): Acc = (TP + TN) / (TP + TN + FP + FN) (5); In equation (5), TP, TN, FP, and FN represent the number of true positive, true negative, false positive, and false negative samples, respectively. The entire screening process involves a total of 100 × 500 = 50,000 evaluations. The stopping condition is "the average accuracy of the model no longer increases when the number of features is increased," thus determining the optimal number of features and the corresponding feature combination (…). Figure 2 In the lung nodule screening model, the SVM was locally optimal with 39 feature combinations in the positive ion mode and with 44 feature combinations in the negative ion mode.

[0094] Subsequently, the extracted ion chromatograms (XIC) and high-resolution MS / MS mass spectra of the positive and negative ion features selected by the algorithm were meticulously compared manually. Features with consistent retention times, good peak shapes, and high matching with the MassBank standard library were selected, while exogenous substances and unknown features were removed. After comprehensive analysis and integration of candidate features in the positive and negative ion modes, and multi-faceted screening, the following five plasma aqueous phase metabolic markers were finally identified: Hypoxanthine, L-ornithine, Decanoyl-L-carnitine, Dehydroepiandrosterone sulfate (DHEA-S), and Cotinine.

[0095] Hypoxanthine has a CAS registry number of 68-94-0 and a molecular formula of C5H4N4O.

[0096] L-Ornithine has the CAS registry number 70-26-8 and the molecular formula C5H. 12 N2O2.

[0097] Decanoyl-L-carnitine has a CAS registry number of 3992-45-8 and a molecular formula of C2. 17 H 33 NO4.

[0098] Dehydroepiandrosterone sulfate (DHEA-S) has a CAS registry number of 651-48-9 and a molecular formula of C2. 19 H 28 O5S.

[0099] Cotinine has a CAS registry number of 486-56-6 and a molecular formula of C. 10 H 12 N2O.

[0100] Five plasma water phase metabolic markers showed significant concentration differences in patients with pulmonary nodules (nodule group, including pathologically confirmed lung cancer and benign pulmonary nodules, collectively referred to as the BL group) compared to healthy individuals (H group). The specific trends are shown in Table 2 below:

[0101] In Table 2, pos represents the positive ion mode and neg represents the negative ion mode. ↑ indicates an increase in ion levels in patients with pulmonary nodules relative to healthy individuals, and ↓ indicates a decrease in ion levels in patients with pulmonary nodules relative to healthy individuals. The p-values ​​above are derived from a differential analysis in the discovery cohort.

[0102] Cotinine is the main metabolite of nicotine. In the subsequent independent validation cohort 2, there was no significant difference in the smoking history composition between the nodule group and the healthy control group (P=0.557, see Example 3). Therefore, the discriminative power of the combination of these 5 biomarkers was not due to the imbalance between groups in smoking exposure. Cotinine, as one feature in the combination, provides complementary discriminative information in conjunction with the other four endogenous metabolites.

[0103] Using the above five metabolic biomarkers, an SVM pulmonary nodule screening model was constructed in the discovery cohort (i.e., constructed according to the methods described in equations (1)–(3)) (nodule group vs. healthy group). The average accuracy (Acc, calculated according to equation (5)) after cross-validation was approximately 88%, indicating that the combination of the five metabolic biomarkers can effectively distinguish between pulmonary nodule patients and healthy individuals. The "pulmonary nodule screening model" mentioned in this invention refers to the SVM model constructed according to the methods described in equations (1)–(3); corresponding model instances were fitted for different datasets (non-target relative intensity or target absolute concentration), and their algorithms, kernel functions, and decision function forms were the same.

[0104] Example 2: Establishment and Absolute Quantification of Targeted Metabolomics Methods 1. Establishment of Targeted MRM Method Based on the mass spectrometry characteristics and ion pair information of the five metabolic biomarkers screened in Example 1, the parameters in multiple reaction monitoring (MRM) mode were optimized on an AB Sciex Jasper HPLC / QTRAP 4500MD system (see Tables 3 and 4), establishing a targeted metabolomics detection method for pulmonary nodule screening. The entire detection process takes only 16 minutes. Figure 3 ).

[0105] Mass spectrometry parameter settings: ionization temperature 500℃, spray voltage 5500V, injection voltage 10V, collision chamber emission voltage 11V, collision gas 8, curtain gas 35, GS1 55, GS2 60.

[0106]

[0107]

[0108] 2. Preparation of standard products Preparation of stock solutions for each standard: The standards of hypoxanthine and L-ornithine hydrochloride were dissolved in 1:1 methanol-water, and the standards of decyl-L-carnitine, dehydroepiandrosterone sulfate sodium salt, and cotinine were dissolved in pure methanol. The concentration of all stock solutions was prepared to be 1 mg / mL.

[0109] Preparation of isotope internal standard stock solutions: [²H4]-hypoxanthine was dissolved in 1:1 methanol-water to prepare a 1 mg / mL stock solution; [²H7]-L-ornithine hydrochloride was dissolved in 1:1 methanol-water to prepare a 2 mg / mL stock solution; [²H3]-decanoyl-L-carnitine hydrochloride was dissolved in pure methanol to prepare a 1 mg / mL stock solution; [²H6]-dehydroepiandrosterone sulfate sodium salt was dissolved in pure methanol to prepare a 1 mg / mL stock solution; [methyl-²H3]-DL-cotinine was dissolved in pure methanol to prepare a 1 mg / mL stock solution. All stock solutions were stored at -80°C.

[0110] Use the above-mentioned mother liquor to prepare 12 calibration points with concentration gradients (S1 to S12).

[0111] The concentration ranges of the standard curves for each metabolic marker and the concentrations of the QC samples are summarized in Table 5 below.

[0112]

[0113] Note: For dehydroepiandrosterone sulfate (DHEA) at calibration point S12 (12000 ng / mL), the detection signal exceeds the linear range (ND), so the upper limit of quantitation is taken as S11 (9000 ng / mL); for cotinine, calibration points S1–S6 (4.6875–37.5 ng / mL) are below the method lower limit of quantitation (ND), so the lower limit of quantitation is taken as S7 (56.25 ng / mL). LQC, MQC, and HQC are obtained by taking the corresponding concentrations of each biomarker's standard curve at S3, S7, and S11, respectively, for batch-to-batch precision and accuracy verification.

[0114] The standard curve points for the five metabolic markers are shown in Tables 6-10.

[0115]

[0116]

[0117]

[0118]

[0119]

[0120] 3. Standard Curve and Methodological Validation By constructing a standard curve ( Figure 4 The correlation coefficients of the standard curves for each metabolic marker are as follows: hypoxanthine r 2 =0.9994, L-ornithine r 2 =0.9994, Decanoyl-L-carnitine r 2 =0.9997, dehydroepiandrosterone sulfate r 2 =0.9978, Cotinine r 2 =0.9847, both of which demonstrate the high accuracy and good linearity of the method.

[0121] During targeted detection, a set of QC samples (HQC, MQC, LQC) was inserted for every 24 plasma samples. The coefficient of variation of all metabolic markers in the QC samples was controlled below 15%, indicating that the method has good precision and repeatability.

[0122] Example 3: Multi-center Independent Validation 1. Validation Cohort 1 – 300 Cases Multicenter Cohort The targeted MRM method described above was used to test a multicenter validation cohort of 300 patients (Table 11). This cohort included 200 patients with pulmonary nodules (100 of whom were pathologically confirmed to have lung cancer and 100 of whom had benign nodules) and 100 healthy controls.

[0123]

[0124] The samples were obtained from the Department of Thoracic Surgery of Peking University People's Hospital, the Department of Thoracic Surgery of Jiangsu Cancer Hospital, and the Department of Thoracic Surgery of Shenzhen Second People's Hospital. The nodule group and the healthy control group were matched by propensity score based on age and gender. After matching, there were no statistically significant differences in age (P=0.76) and gender (P=0.06) between the two groups. Using the absolute concentration data of five metabolic markers from validation cohort 1, the SVM lung nodule screening model was fitted according to the methods described in equations (1)–(3). The model performance was evaluated according to equations (5), (6), and (7): Sn = TP / (TP + FN) (6); Sp = TN / (TN + FP) (7); In equations (6) and (7), Sn is the sensitivity, Sp is the specificity, and TP, TN, FP, and FN have the same meaning as in equation (5).

[0125] Five metabolic markers showed significant concentration differences between the nodule group and the healthy control group. Among them, hypoxanthine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine were decreased in the nodule group, while L-ornithine was increased, consistent with the findings of the cohort. Figure 5 Both t-SNE dimensionality reduction and PLS-DA showed good clustering and separation effects between the nodule group and the healthy control group. Figure 5 ).

[0126] For the absolute concentration data of the five metabolic markers in validation cohort 1, the SVM lung nodule screening model was refitted according to the method described in equations (1)–(3). After 2000 random four-fold cross-validations, the average accuracy (Acc) was 87.88%, the average sensitivity (Sn) was 90.62%, and the average specificity (Sp) was 72.82% (Acc, Sn, and Sp were calculated according to equations 5, 6, and 7, respectively).

[0127] 2. Validation Queue 2 – 214 Independent Queues (Key Validation Results) Further validation was performed in an independent prospective validation cohort of 214 patients (Table 12). The training set was the validation cohort 1; the independent test set included 164 patients with pulmonary nodules and 50 healthy controls (99 of whom had lung cancer and 65 had benign nodules in the nodule group).

[0128]

[0129] Validation method and validated model: Validation cohort 2 adopted the same SVM construction method as Equations (1)–(3). The target absolute concentration data of validation cohort 1 (multicenter, 300 cases: nodules + healthy controls) were used as the training set to fit and lock the model parameters (including support vector weights α, bias b, and classification threshold), and then independently tested on an independently enrolled cohort of 214 cases (164 nodules + 50 healthy controls). The validated model is the same SVM lung nodule screening model constructed according to Equations (1)–(3) and whose parameters were locked on validation cohort 1. Heatmap ( Figure 6 (A) Hierarchical clustering was performed based on the absolute concentrations of the five biomarkers after Z-score normalization; ROC analysis ( Figure 6 (B) Using the decision function output f(x) of the locked model on the independent test set samples as the score, the AUROC and optimal operating point are calculated. Heatmap visualization shows the expected changes in the five metabolites in the nodule group and the healthy group. Figure 6 (A) Hierarchical clustering confirmed good separation between the nodule group and the healthy group. ROC analysis results ( Figure 6 (B): Independent test set AUROC = 0.921 (95% CI: 0.887-0.954), sensitivity 90.0%, specificity 81.8%.

[0130] Regarding the difference in specificity between the two validation cohorts: the average specificity of cross-validation in validation cohort 1 was 72.82%, while the specificity of the independent test set in validation cohort 2 was 81.8%. The difference is mainly related to the purpose of the two cohorts and the composition of the healthy control population—validation cohort 1 was used for cross-validation and determination of model parameters (including classification thresholds), while validation cohort 2 was used to test the same model with locked parameters on an independent population; the sensitivity of both cohorts remained stable at approximately 90% (90.62% and 90.0%), reflecting the reproducibility and robustness of the model in different populations.

[0131] The results demonstrate that the combination of five plasma water phase metabolic markers has robust and repeatable screening efficacy for pulmonary nodules in independent populations, and can screen individuals with pulmonary nodules from healthy individuals with high sensitivity.

[0132] The present invention has been described in detail above. Those skilled in the art will recognize that the invention can be practiced in a wide range of ways with equivalent parameters, concentrations, and conditions without departing from its spirit and scope, and without requiring unnecessary experiments. While specific embodiments have been provided, it should be understood that further modifications can be made to the invention. In summary, according to the principles of the invention, this application is intended to include any changes, uses, or improvements to the invention, including changes made using conventional techniques known in the art that depart from the scope disclosed herein.

Claims

1. The use of metabolic markers and / or reagents for detecting said metabolic markers in any of the following: A1) Application in the preparation of products for pulmonary nodule screening; A2) Application in the preparation of products for the diagnosis or auxiliary diagnosis of pulmonary nodules; A3) Application in the preparation of products for predicting the risk of pulmonary nodules; A4) Application in constructing a pulmonary nodule screening model; The metabolic markers include inosine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine.

2. The application according to claim 1, characterized in that, The reagents include those for detecting the concentration of the biomarkers in plasma by methods such as mass spectrometry, chromatography, spectroscopy, nuclear magnetic resonance, liquid chromatography-mass spectrometry, or gas chromatography-mass spectrometry.

3. A kit for screening pulmonary nodules, characterized in that, The kit includes reagents for detecting the metabolic markers described in claim 1.

4. The reagent kit according to claim 3, characterized in that, The reagents include standards of the metabolic markers described in claim 1.

5. A method for screening metabolic biomarkers for pulmonary nodules, characterized in that, The method includes: performing non-targeted metabolomics analysis and data preprocessing on plasma samples from patients with pulmonary nodules and healthy subjects; identifying combinations of five plasma aqueous phase metabolic markers—hypoxanthine, L-ornithine, decanoyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine—through support vector machine feature screening and primary and secondary mass spectrometry library searches; and identifying the chemical identities of the above metabolic markers and their combinations therefrom.

6. A method for constructing a pulmonary nodule screening model, characterized in that, The method involves using data on the concentrations of hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine in the plasma of known patients with pulmonary nodules and healthy subjects as training samples, and employing statistical analysis methods to establish a pulmonary nodule screening model.

7. The method according to claim 6, characterized in that, The statistical analysis methods include support vector machine analysis, random forest analysis, decision tree analysis, logistic regression analysis, and neural network analysis.

8. The method according to claim 6 or 7, characterized in that, The method includes the following steps: S1) Obtain peripheral blood plasma samples from patients with pulmonary nodules and healthy subjects; S2) The concentrations of five metabolic markers—hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine—in the plasma sample were detected using the targeted multiple reaction monitoring (MRM) method to obtain targeted absolute quantitative concentration data. S3) The target absolute quantitative concentration data is used as the input feature of the support vector machine classifier. The target MRM detection method remains fixed, and only the support vector machine classifier is trained to obtain the trained support vector machine model, which is the lung nodule screening model.

9. The application according to claim 1 or 2, characterized in that, The products include reagents, kits, detection chips, test strips, test cards, immunosensors, or devices.

10. A device for screening pulmonary nodules, characterized in that, The device includes: The data receiving module is used to receive data on the concentrations of hypoxanthine, L-ornithine, decyl-L-carnitine, dehydroepiandrosterone sulfate, and cotinine in plasma samples of test subjects from at least one terminal. The data processing module is used to input the data into the lung nodule screening model; the lung nodule screening model is a lung nodule screening model constructed using the method described in any one of claims 6-8; The data output module is used to output the lung nodule screening results to at least one client.

Images