A diabetes retinopathy screening kit based on neldi-ms whole blood metabolic fingerprint and use method

Abstract

This invention discloses a screening kit for diabetic retinopathy (DR) based on NELDI-MS whole blood metabolic fingerprinting and its usage method. The kit includes: a blood cell lysis diluent, an iron-based nanoparticle matrix solution, a whole blood metabolite extraction reagent, quality control reference samples / calibration baseline samples and their stock solutions, and data processing instructions. In use, after lysis and extraction of whole blood samples, high-throughput metabolic fingerprinting is performed using NELDI-MS. Feature clusters are constructed using core m / z features and derived ion features. Cross-batch drift correction is performed using the quality control reference samples, and the results are input into a discriminant model to output DR risk probability or classification results. This invention eliminates the need for chromatographic separation and serum / plasma centrifugation pretreatment, has a short single-sample detection time, is suitable for large-scale rapid screening in outpatient clinics and primary healthcare institutions, exhibits high diagnostic sensitivity and specificity, good stability, and has good clinical translational potential.

Description

Technical Field

[0001] This application belongs to the field of medical testing kits, and involves the fields of metabolomics detection, nanoparticle-enhanced laser desorption / ionization mass spectrometry and in vitro assisted screening and early warning technology. Specifically, it relates to a screening kit for diabetic retinopathy based on NELDI-MS whole blood metabolic fingerprinting and its usage method, as well as its application in early screening and risk warning of diabetic retinopathy. Background Technology

[0002] Diabetic retinopathy (DR) is one of the most common microvascular complications of diabetes and a leading cause of blindness in working-age populations. DR is characterized by its insidious nature and lack of obvious clinical symptoms in its early stages. Without timely screening and intervention, it can progress to proliferative changes and lead to irreversible visual impairment. Currently, the screening and standardized treatment rates for DR in my country are far below ideal levels, resulting in a heavy disease burden and posing a serious challenge to clinical work and public health strategies.

[0003] Early-stage diabetic retinopathy (DR) often presents with no obvious symptoms. Current clinical screening and diagnosis primarily rely on imaging examinations such as fundus photography, optical coherence tomography (OCT / OCTA), and fluorescein angiography. These methods are highly dependent on specialized equipment and ophthalmologists, are costly and time-consuming, and have low availability in primary healthcare institutions, making it difficult to meet the needs of large-scale population screening. Meanwhile, hematological tests, mainly based on metabolic indicators such as blood glucose and glycated hemoglobin, have poor specificity and insufficient sensitivity for DR lesions, failing to achieve early, large-scale, rapid screening and risk warning.

[0004] In recent years, metabolomics methods based on mass spectrometry have provided a new approach for the study of DR biomarkers. Existing reports mostly employ liquid chromatography-tandem mass spectrometry (LC-MS / MS) for targeted or non-targeted metabolite detection in serum / plasma samples. However, while LC-MS / MS possesses high detection sensitivity and qualitative / quantitative capabilities, it relies on chromatographic separation, resulting in long single-sample analysis times and limited throughput, making it unsuitable for large-scale clinical screening scenarios. Furthermore, serum / plasma samples require centrifugation, a cumbersome pretreatment step that limits sample stability. In addition, although LC-MS / MS has been applied to the metabolomics analysis of a limited number of whole blood samples, the whole blood matrix contains more proteins, blood cells, and inorganic salts than plasma and serum, easily leading to chromatographic system contamination and signal inhibition. The stability and reproducibility of large-scale analyses remain challenging, hindering the establishment of DR metabolic screening methods based on whole blood matrix.

[0005] In summary, the existing technologies have the following main drawbacks: (1) Imaging screening is costly, time-consuming, and dependent on specialized resources, making it unsuitable for rapid screening in outpatient clinics and primary care settings; (2) Traditional blood glucose metabolism indicators lack specificity for diabetic renal disease (DR) and have insufficient early warning capabilities; (3) Existing mass spectrometry detection markers are limited in combination, have low throughput, and lack sufficient sensitivity and accuracy, resulting in limited diagnostic capabilities; (4) Serum / plasma sample pretreatment is complex and has poor stability, making it unsuitable for rapid clinical testing; (5) There is a lack of whole blood direct detection kits suitable for people with unknown history of diabetes in ophthalmology outpatient clinics; (6) Existing mass spectrometry methods have long single-sample detection times, which cannot meet the throughput requirements of large-scale population screening. Summary of the Invention

[0006] Terminology Explanation:

[0007] Diabetic retinopathy (DR) is a retinal complication caused by diabetes, which can lead to a rapid decline in vision or even blindness.

[0008] "Whole blood metabolic fingerprint" refers to the set of spectral features composed of multiple m / z features and their peak intensities obtained after a sample is detected by NELDI-MS.

[0009] "Core m / z features" refer to m / z features with disease discrimination value obtained from the training set or discovery queue after statistical analysis, machine learning or manual rule screening.

[0010] "Derived ion characteristics" refer to the characteristic peaks of protonation, dehydration, metal ion addition, complex addition, or other ionization derivative forms that correspond to the core m / z characteristics.

[0011] A “feature cluster” refers to a combined unit consisting of a core m / z feature and one or more derived ion features, which can be characterized by single peak intensity, peak intensity ratio, weighted sum, statistical projection value or a combination thereof.

[0012] To address the aforementioned problems in existing technologies, this application proposes a novel diabetic retinopathy screening kit and its usage method based on NELDI-MS whole blood metabolic fingerprinting, using core m / z features and their derived ion features as discrimination objects, and combining quality control references and model interpretation. This kit can be used for early screening and risk warning of diabetic retinopathy, and can also be further extended to the screening of microvascular complications related to metabolic diseases. This kit is suitable for ophthalmology clinics, primary healthcare institutions, health check-up centers, and third-party medical testing institutions.

[0013] This invention aims to solve the following technical problems:

[0014] (1) How to collect metabolic fingerprints of whole blood samples rapidly and reproducibly without relying on chromatographic separation and absolute quantification of specific metabolites;

[0015] (2) How to maintain consistency in feature extraction, data correction and model output under complex whole blood matrix and multi-batch testing conditions;

[0016] (3) How to integrate the core m / z features with their common derived ion forms into a feature cluster that can be used for disease discrimination, so as to avoid discrimination vulnerability caused by unimodal instability;

[0017] (4) How to construct a DR-assisted screening process that is suitable for both main cohort modeling and subsequent deployment batches and extended sampling methods such as finger-prick blood;

[0018] (5) How to achieve early, rapid, and high-throughput DR screening, adapt to real clinical scenarios in outpatient / primary healthcare institutions, and improve the popularity and accuracy of early DR screening;

[0019] (6) How to establish a stable, reproducible, and industrially producible reagent kit detection system.

[0020] To achieve the above objectives, this application provides the following technical solution:

[0021] In a first aspect, this application provides a screening kit for diabetic retinopathy based on NELDI-MS whole blood metabolic fingerprinting. The kit is used to preprocess whole blood samples to adapt to the NELDI-MS detection platform and obtain a whole blood metabolic fingerprint spectrum containing multiple m / z features and their peak intensities. Screening for diabetic retinopathy is achieved by constructing feature clusters and a discrimination model. The kit includes:

[0022] (1) Diluent for blood cell lysis;

[0023] (2) NELDI-MS nanoparticle matrix liquid, which contains iron-based nanoparticles, is used to enhance laser desorption ionization efficiency and improve detection repeatability;

[0024] (3) Whole blood metabolite extraction reagent;

[0025] (4) Quality control reference samples / calibration benchmark samples and their stock solutions;

[0026] (5) Data processing instructions for performing feature extraction, drift correction, feature alignment and risk assessment.

[0027] Furthermore, in the kit of the present invention, the whole blood metabolite extraction reagent is a mixed solvent prepared by methanol and water at a volume ratio of 4:1.

[0028] Furthermore, in the kit of the present invention, the NELDI-MS nanoparticle matrix solution has a concentration of 1 mg·mL⁻¹. -1 An ultrapure aqueous dispersion of iron oxide nanoparticles.

[0029] Furthermore, in the kit of the present invention, the iron-based nanoparticles are in the form of nanospheres, with Fe and O elements uniformly distributed and having a crystalline structure.

[0030] Furthermore, the detection platforms compatible with the kit of the present invention include NELDI-MS, MALDI-TOF, or LC-MS / MS;

[0031] The kit is also suitable for the detection of serum or plasma samples.

[0032] Secondly, this application provides a method for using the aforementioned NELDI-MS whole blood metabolic fingerprint-based diabetic retinopathy screening kit, such as... Figure 1 As shown, it includes the following steps:

[0033] S1. Sample collection and preservation: Collect venous whole blood samples and transfer them to a low-temperature storage state within 4 hours, freezing them at -80℃; thaw them at 4℃ before testing;

[0034] S2. Blood cell lysis: Take a thawed whole blood sample, add the blood cell lysis diluent, vortex mix, and let stand at 4°C to allow the blood cells to fully lyse and release intracellular metabolites.

[0035] S3. Metabolite extraction: Add the whole blood metabolite extraction reagent, let stand at -20℃, centrifuge, and collect the supernatant as the NELDI-MS sample;

[0036] S4. NELDI-MS detection: The sample to be tested is mixed with the NELDI-MS nanoparticle matrix solution or sequentially spotted onto a mass spectrometry target plate to form a sample spot array; each sample is repeatedly detected and the average spectrum is taken to obtain the raw metabolic fingerprint spectrum data.

[0037] S5. Feature extraction: Peak detection, peak alignment, feature filtering, and normalization are performed on the original metabolic fingerprint data to construct a main feature table; potential multi-charge species are filtered according to the mass defect rule to obtain core m / z features and their derived ion features;

[0038] S6. Feature Cluster Construction: The core m / z features and their corresponding derived ion features are integrated into a feature cluster. The derived ion features include ion peaks in the form of protonation, dehydration, metal ion addition, or complex addition. The feature cluster is characterized by single-peak normalized intensity, peak intensity ratio, weighted sum, statistical projection value, or a combination thereof, forming the feature vector of the input model.

[0039] S7. Quality Control and Drift Correction: Insert the quality control reference sample into each analysis batch, establish a batch drift curve based on the feature intensity or feature cluster response value of the quality control reference sample, and perform drift correction on the feature intensity; retain features with a coefficient of variation below a preset threshold for subsequent modeling.

[0040] S8. Model discrimination and auxiliary screening output: Input the drift-corrected feature vector into the discrimination model and output the risk probability of diabetic retinopathy, positive / negative classification results or risk stratification level.

[0041] Furthermore, in the method of the present invention, in step S2, 20 μL of whole blood sample is added to 80 μL of ultrapure water, vortexed for 30s, and then allowed to stand at 4℃ for 30 min.

[0042] In step S3, a mixed solvent of methanol and water at a volume ratio of 4:1 is added, and the mixture is allowed to stand at -20℃ for 1 h. After centrifugation at 12000 rpm for 15 min, the supernatant is collected as the NELDI-MS sample.

[0043] In step S4, the final volume of the sample point is 1 μL, and the amount of nanoparticles used is 1 μg; each sample is tested 5 times.

[0044] Furthermore, in the method of the present invention, in step S5, the feature extraction is performed based on MALDIquant or an equivalent software process;

[0045] In step S6, the derived ion characteristics include [M + H]. + [M + H H2O] + [M + Na] + [M + K] + [M + 2Na - H] + [M + Na + K - H] + and [M + 2K - H] + At least one of them;

[0046] In step S7, 10-100 quality control reference samples are inserted into each analysis batch; the drift correction uses the Lowess method; and the preset threshold is 40%.

[0047] Furthermore, in the method of the present invention, in step S8, the discrimination model is selected from logistic regression, support vector machine, K-nearest neighbor, ensemble learning or neural network model.

[0048] Furthermore, in the method of the present invention, step S8 further includes: inputting the metabolic feature output score of the discrimination model together with at least one clinical indicator among age, gender, glycated hemoglobin, random blood glucose and body mass index into the joint model, and outputting the risk probability of diabetic retinopathy, positive / negative classification result or risk stratification level.

[0049] In summary, compared with the prior art, this application has the following technical advantages:

[0050] (1) This invention focuses on whole blood metabolic fingerprint and its characteristic clusters, rather than relying on the absolute quantification of a single metabolite, which is more in line with the rapid screening properties of NELDI-MS in complex matrices.

[0051] (2) This invention incorporates the core m / z features and their derived ion features into the discrimination vector, which helps to improve the robustness of feature expression and tolerance to ionization fluctuations;

[0052] (3) The present invention introduces a whole blood sample quality control reference sample and a cross-batch calibration mechanism, which can maintain the consistency of model deployment under large sample, multiple batch and cross-center conditions;

[0053] (4) This invention is applicable to venous whole blood samples, requires no professional pretreatment, reduces potential deviations caused by centrifugation and hemolysis sensitivity, and is easy to connect with the routine blood collection process of community primary medical institutions and ophthalmology clinics. It is suitable for high-throughput assisted screening and risk warning scenarios. Moreover, this kit can be adapted to LDI-MS equipment that is widely equipped in medical institutions, and has good translation potential in high-throughput outpatient / primary medical institution screening scenarios.

[0054] (5) The present invention can also be combined with conventional clinical indicators to construct a combined model, further enhancing its clinical application value.

[0055] Other features and advantages of this application will be set forth in detail in the following description, or will become apparent through the implementation of the relevant technical solutions of this application. The objectives and other advantages of this application can be achieved through the technical features and means explicitly pointed out in the description, claims, and drawings, and will be obtained through the implementation of these technical contents. Attached Figure Description

[0056] To more clearly illustrate the technical solution of this application, the accompanying drawings involved in the description of this invention will be briefly introduced below. It should be noted that the drawings only show some embodiments of the invention. For those skilled in the art, other related drawings can be derived from these drawings without creative effort.

[0057] Figure 1This is a flowchart illustrating the implementation method of the NELDI-MS whole blood metabolic fingerprint screening kit for diabetic retinopathy based on the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are only some embodiments of this application, and this application can also be implemented or applied through other different specific implementation methods. The details in this specification can also be modified or changed based on different viewpoints and applications.

[0059] At the same time, it should be understood that the scope of protection of this application is not limited to the specific implementation schemes described below; it should also be understood that the terminology used in the embodiments of this application is for describing specific implementation schemes, and not for limiting the scope of protection of this application.

[0060] When numerical ranges are given in the embodiments, it should be understood that, unless otherwise stated in this application, both endpoints of each numerical range and any value between the two endpoints may be selected. Unless otherwise defined, all technical terms used in this application have the same meaning as commonly understood by those skilled in the art. In addition to the specific methods, equipment, and materials used in the embodiments, based on the knowledge of those skilled in the art and the description in this application, any prior art methods, equipment, and materials similar to or equivalent to those described, used, and materials in the embodiments of this application may be used to implement this application. In this application, unless otherwise specified, all parts and percentages are in units of weight, and unless otherwise specified, all instruments and reagents are commercially available or commonly used in the industry. Unless otherwise specified, the experimental methods and operations in the following embodiments are conventional experimental methods and operations in the art.

[0061] Example: A screening kit for diabetic retinopathy based on NELDI-MS whole blood metabolic fingerprinting

[0062] (I) Components of the reagent kit

[0063] (1) Diluent for blood cell lysis.

[0064] (2) NELDI-MS nanoparticle matrix solution: Iron-based nanoparticles were prepared as the matrix using a solvothermal method, which was specifically designed for NELDI-MS to improve ionic strength and detection repeatability. Transmission electron microscopy and elemental distribution characterization confirmed that the iron-based nanoparticles were in the form of nanospheres with uniform distribution of Fe and O elements; X-ray diffraction confirmed their crystal structure.

[0065] (3) Metabolite extraction reagent: organic solvent methanol / water = 4 / 1 (v / v), used for whole blood metabolite extraction.

[0066] (4) Quality control reference sample / calibration benchmark sample and its stock solution.

[0067] (5) Data processing instructions for performing feature extraction, drift correction, feature alignment and risk assessment.

[0068] In addition, the kit includes a testing instruction manual, which includes instructions on sample processing, mass spectrometry detection, and model interpretation procedures.

[0069] (II) Instructions for use of the reagent kit

[0070] The method of using the kit of the present invention includes steps such as sample collection, whole blood lysis and extraction, NELDI-MS acquisition, feature extraction, feature cluster construction, data correction, feature alignment and model discrimination.

[0071] Sample type: Human venous whole blood sample, no serum / plasma separation required. It should be noted that this kit is also suitable for serum / plasma sample testing; only the extraction reagent ratio needs to be adjusted.

[0072] 1. Collect 1 mL of sample and transfer it to a low-temperature storage state within 4 h, store at -80℃; thaw at 4℃ before testing.

[0073] Preferably, the test samples can be prepared into lyophilized reagents to improve room temperature stability and reduce reliance on the cold chain.

[0074] 2. Take 20 μL of thawed whole blood sample and place it in a centrifuge tube. Add 80 μL of ultrapure water, vortex for 30 s, and let stand at 4℃ for 30 min to fully lyse blood cells and release intracellular metabolites.

[0075] 3. Add extraction reagent (methanol / water = 4 / 1, v / v), and let stand at -20℃ for 1 h; then centrifuge at 12000 rpm for 15 min, and collect the supernatant as the NELDI-MS sample.

[0076] 4. NELDI-MS detection

[0077] The sample to be tested is mixed with the nanoparticle matrix or sequentially spotted onto a mass spectrometry target plate to form a sample spot with a final volume of approximately 1 μL and a nanoparticle dosage of approximately 1 μg. Multiple sample spots can be arranged into a spot array, thereby supporting the automated acquisition of hundreds of samples in a single batch. The nanoparticle matrix is ​​1 mg•mL. -1 An ultrapure aqueous dispersion of iron oxide nanoparticles.

[0078] Each sample was tested five times and the average spectrum was taken to reduce the impact of single spotting and local ionization differences on the final feature extraction.

[0079] It should be noted that, in addition to the NELDI-MS detection platform, this kit is also compatible with the MALDI-TOF and LC-MS / MS detection platforms.

[0080] 5. Feature Extraction

[0081] The raw spectral data can be used for feature extraction based on MALDIquant or an equivalent software process, which includes at least peak detection, peak alignment, feature filtering, and normalization.

[0082] In a preferred embodiment, a principal feature table is constructed by averaging the repeated spectra of each sample; and potential multi-charge species are filtered according to the mass defect rule to reduce the interference of non-target features in complex spectra on subsequent modeling.

[0083] Preferably, the derived ion features in the feature cluster include, but are not limited to, [M + H]. + [M + H H2O] + [M+ Na] + [M + K] + [M + 2Na - H] + [M + Na + K - H] + and [M + 2K - H] + For the same core feature, you can choose to match all derived peaks, or you can select one or more derived peaks with higher stability to participate in the modeling.

[0084] In terms of feature representation, the feature vector of the final input model can be formed by using single-peak normalized intensity, peak intensity ratio between main peak and accompanying peak, weighted sum of multiple peaks in the feature cluster, feature cluster projection score, or a combination thereof.

[0085] 6. Quality control reference samples and drift correction

[0086] To improve the robustness of large-sample, multi-batch analyses, quality control reference samples are inserted into each analytical batch. These quality control reference samples can be prepared by mixing multiple whole blood samples, or by using a simulated whole blood matrix with predetermined reference ions. Their function is to monitor changes in sample pretreatment, spotting, collection, and instrument status, and to support subsequent drift correction.

[0087] In a preferred embodiment, 10-100 quality control reference samples, more preferably about 40, are inserted at fixed positions in each analytical batch. A batch drift curve is established based on the median, mean, or characteristic cluster response value of the quality control reference samples in the batch, preferably using the Lowess method for drift correction.

[0088] After drift correction, features with coefficients of variation below a preset threshold are further retained for subsequent modeling; the preferred threshold is 40%, but those skilled in the art can adjust this threshold according to sample size, instrument status and model tolerance.

[0089] 7. Model building and auxiliary screening output

[0090] The discriminant model of this invention can employ logistic regression, support vector machine, K-nearest neighbors, ensemble learning, or neural network models. Preferably, a logistic regression model with L1 regularization is used to maintain discriminative performance while performing feature compression and model interpretation.

[0091] In a further preferred embodiment, the output score of the metabolic feature model is input together with clinical indicators such as age, gender, glycated hemoglobin, random blood glucose and body mass index into the MetClinical joint model to output the DR risk probability, positive / negative classification result or risk stratification level.

[0092] Clinical trials have shown that the kit of this invention can achieve the following detection results for early screening of diabetic retinopathy:

[0093] (1) Extremely high diagnostic sensitivity and specificity

[0094] Discovery queue: AUC=0.999, sensitivity 0.984, specificity 0.969;

[0095] Internal validation queue: AUC=0.996, sensitivity 0.968, specificity 0.988;

[0096] Independent deployment queue: AUC=0.984, sensitivity 0.910, specificity 0.957;

[0097] MetClinical model: AUC = 0.992 - 0.998.

[0098] (2) Precise identification of clinically challenging populations

[0099] Subgroup with abnormal glucose metabolism (HbA1c 5.7%-6.5%): AUC=0.986-0.999;

[0100] Finger-prick blood minimally invasive test: AUC=0.911;

[0101] Differentiating between diabetic patients with and without DR: AUC=0.891.

[0102] (3) High throughput and fast speed

[0103] A single sample can be tested in less than 30 seconds, and 384 samples can be completed in 1.5 hours.

[0104] (4) High stability

[0105] It exhibits high stability in metabolic fingerprinting of thousands of samples, with a median CV of 24.5% in large-sample QC, effectively matching clinical testing requirements.

[0106] The above description is merely a preferred embodiment of this application and is not intended to limit this application in any way. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to obtain equivalent embodiments without departing from the scope of the technical solution of this application. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of this application without departing from the content of the technical solution of this application should be included within the scope of protection of the claims of this application.

Claims

1. A screening kit for diabetic retinopathy based on NELDI-MS whole blood metabolic fingerprinting, characterized in that, The kit is used for preprocessing whole blood samples to adapt to the NELDI-MS detection platform and obtain a whole blood metabolic fingerprint spectrum containing multiple m / z features and their peak intensities. It then uses the constructed feature clusters and discriminant model to screen for diabetic retinopathy. The kit includes: (1) Diluent for blood cell lysis; (2) NELDI-MS nanoparticle matrix liquid, which contains iron-based nanoparticles; (3) Whole blood metabolite extraction reagent; (4) Quality control reference samples / calibration benchmark samples and their stock solutions; (5) Data processing instructions for performing feature extraction, drift correction, feature alignment and risk assessment.

2. The reagent kit according to claim 1, characterized in that, The whole blood metabolite extraction reagent is a mixed solvent of methanol and water at a volume ratio of 4:

1.

3. The reagent kit according to claim 1, characterized in that, The NELDI-MS nanoparticle matrix solution has a concentration of 1 mg·mL⁻¹ -1 An ultrapure aqueous dispersion of iron oxide nanoparticles.

4. The reagent kit according to claim 1, characterized in that, The iron-based nanoparticles are in the form of nanospheres, with uniform distribution of Fe and O elements, and have a crystalline structure.

5. The reagent kit according to claim 1, characterized in that, The kit is compatible with detection platforms including NELDI-MS, MALDI-TOF, or LC-MS / MS; The kit is also suitable for the detection of serum or plasma samples.

6. A method of using the reagent kit as described in any one of claims 1-5, characterized in that, Includes the following steps: S1. Sample collection and preservation: Collect venous whole blood samples and transfer them to a low-temperature storage state within 4 hours, freezing them at -80℃; thaw them at 4℃ before testing; S2. Blood cell lysis: Take a thawed whole blood sample, add the blood cell lysis diluent, vortex mix, and let stand at 4°C to allow the blood cells to fully lyse and release intracellular metabolites. S3. Metabolite extraction: Add the whole blood metabolite extraction reagent, let stand at -20℃, centrifuge, and collect the supernatant as the NELDI-MS sample; S4. NELDI-MS detection: The sample to be tested is mixed with the NELDI-MS nanoparticle matrix solution or sequentially spotted onto a mass spectrometry target plate to form a sample spot array; each sample is repeatedly detected and the average spectrum is taken to obtain the raw metabolic fingerprint spectrum data. S5. Feature extraction: Peak detection, peak alignment, feature filtering, and normalization are performed on the original metabolic fingerprint data to construct a main feature table; potential multi-charge species are filtered according to the mass defect rule to obtain core m / z features and their derived ion features; S6. Feature Cluster Construction: The core m / z features and their corresponding derived ion features are integrated into a feature cluster. The derived ion features include ion peaks in the form of protonation, dehydration, metal ion addition, or complex addition. The feature cluster is characterized by single-peak normalized intensity, peak intensity ratio, weighted sum, statistical projection value, or a combination thereof, forming the feature vector of the input model. S7. Quality Control and Drift Correction: Insert the quality control reference sample into each analysis batch, establish a batch drift curve based on the feature intensity or feature cluster response value of the quality control reference sample, and perform drift correction on the feature intensity; retain features with a coefficient of variation below a preset threshold for subsequent modeling. S8. Model discrimination and auxiliary screening output: Input the drift-corrected feature vector into the discrimination model and output the risk probability of diabetic retinopathy, positive / negative classification results or risk stratification level.

7. The method according to claim 6, characterized in that, In step S2, take 20 μL of whole blood sample and add 80 μL of ultrapure water, vortex for 30 s and then let stand at 4℃ for 30 min; In step S3, a mixed solvent of methanol and water at a volume ratio of 4:1 is added, and the mixture is allowed to stand at -20℃ for 1 h. After centrifugation at 12000 rpm for 15 min, the supernatant is collected as the NELDI-MS sample. In step S4, the final volume of the sample point is 1 μL, and the amount of nanoparticles used is 1 μg; each sample is tested 5 times.

8. The method according to claim 6, characterized in that, In step S5, the feature extraction is performed based on MALDIquant or an equivalent software process; In step S6, the derived ion characteristics include [M + H]. + [M + H H2O] + [M + Na] + [M + K] + [M + 2Na - H] + [M + Na + K - H] + and [M + 2K - H] + At least one of them; In step S7, 10-100 quality control reference samples are inserted into each analysis batch; the drift correction uses the Lowess method; and the preset threshold is 40%.

9. The method according to claim 6, characterized in that, In step S8, the discrimination model is selected from logistic regression, support vector machine, K-nearest neighbors, ensemble learning or neural network model.

10. The method according to claim 6, characterized in that, Step S8 also includes: inputting the metabolic feature output score of the discriminant model together with at least one clinical indicator among age, gender, glycated hemoglobin, random blood glucose and body mass index into the joint model, and outputting the risk probability of diabetic retinopathy, positive / negative classification result or risk stratification level.

Images