Application of urinary FLRT2 protein in the preparation of biomarkers for IgA nephropathy

Abstract

This invention relates to the application of urinary FLRT2 protein in the preparation of IgA nephropathy biomarkers, belonging to the field of materials science involving specific methods for the study or analysis of proteins, peptides, or amino acids. The application refers to the use of urinary FLRT2 protein as a biomarker in the preparation of predictive models for IgA nephropathy, in the preparation of online diagnostic systems for IgA nephropathy, and in the preparation of reagent kits for the detection of IgA nephropathy. This invention has the following advantages: 1. Achieves high-precision non-invasive diagnosis; 2. Overcomes the limitations of the "black box" and provides interpretable decision support; 3. Completes clinical translation and provides real-time online tools; 4. Verifies the unique added value of FLRT2 biomarkers; 5. Possesses strong generalization ability and robustness; 6. Provides a potentially efficient and low-cost screening pathway.

Description

Technical Field

[0001] This invention belongs to the field of special methods for studying or analyzing materials involving proteins, peptides or amino acids, specifically relating to the application of urinary FLRT2 protein in the preparation of IgA nephropathy biomarkers. Background Technology

[0002] The main characteristic of IgA nephropathy (IgAN) is the deposition of IgA and its immune complexes in the glomeruli. The 2024 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease, issued by the Kidney Disease Improvement Global Organization (KDIGO), emphasizes that renal biopsy is the authoritative method for diagnosing IgA nephropathy [1, KDIGO 2024 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease. Kidney Int. 2024;105(4s):S117-s314.]. However, although renal biopsy is considered the "gold standard" diagnostic method, this technique has significant limitations. Therefore, there is an urgent need to find a non-invasive and accurate early diagnostic method [2, Moresco RN, Speeckaert MM, Delanghe JR. Diagnosis and monitoring of IgA nephropathy: the role of biomarkers as an alternative to renal biopsy. Autoimmun Rev. 2015;14(10):847-53.].

[0003] Recent studies have focused on constructing a non-invasive diagnostic system for IgA nephropathy (IgAN) through combinations of laboratory indicators. Gao's team [3, Gao J, Cui J, Wang Y, Dong Z, Tian Y, Xu Y. Identification of potential predictors for subtype IgA nephropathy through analyses of blood biochemical indicators. Clin Chim Acta. 2011;412(5-6):441-5.] developed a predictive model based on 58 cases of IgAN and 63 cases of non-IgA nephropathy (NigAN), integrating serum IgA, fibrinogen, and 33 clinical indicators. The diagnostic accuracy of their logistic regression and discriminant analysis models both exceeded 77%. However, due to sample size limitations, the generalization ability of this model still needs to be validated in a larger cohort. Another study involving 536 patients with glomerular disease showed that a combined model consisting of age, IgA / C3 ratio, albumin, and IgG achieved a sensitivity of 79.4% and 81% in the training set and a specificity of 90.9% and 88.2% in the validation set [4, Zhou Liping, Liu Juan, Wang Yue, Liu Tingting, Hou Jie. Construction and validation of a non-invasive predictive model for IgA nephropathy based on logistic regression. Chinese Journal of Experimental Diagnostics. 2021;25(09):1339-42.]. However, this study only included individuals with normal renal function (eGFR≥60 mL / min / 1.73m²) and did not include patients with impaired renal function, which may affect its clinical applicability. Han et al. [5, Han QX, Wang Y, Zhu HY, Zhang D, Gao J, Liu ZS, Cai GY, Chen XM. A non-invasive diagnostic model of immunoglobulin A nephropathy and serological markers for evaluating diseaseseverity. Chin Med J (Engl). 2019;132(6):647-52.] included 272 patients with IgA nephropathy and 518 patients without IgA nephropathy. Through serum biomarker analysis, they found that a multivariate predictive model based on total cholesterol, fibrinogen, D-dimer, IgA concentration, and age had an area under the receiver operating characteristic (AUC) of 0.86 for identifying IgA nephropathy, with a sensitivity of 85.98% and a specificity of 73.85%.Meng Xiaohui [6, Meng Xiaohui, Nanjing Zhu, Gao Jing, Wang Yong. Establishment and validation of a mathematical model for the combined diagnosis of IgA nephropathy. Marker Immunoassay and Clinical. 2017;24(02):133-41.] et al. constructed a combined diagnostic model including seven indicators such as albumin, creatinine, and immune complexes by including 588 patients diagnosed by renal biopsy (including 200 IgAN). The study confirmed that the multi-parameter model significantly improved the area under the curve of a single indicator from 0.747 to 0.890, with an overall accuracy of 81.3%. Although the preliminary data of the above models showed good discriminative ability, the studies were all single-center data, and the case selection was biased, resulting in relatively weak representativeness. The lack of external validation from independent cohorts made the data analysis relatively simple and unable to deeply analyze the complex relationship between indicators and the disease.

[0004] With the rapid advancement of computing technology, artificial intelligence (AI) and its important branch, machine learning (ML), are increasingly widely used in the medical field. In recent years, the application of machine learning algorithms in the field of IgA nephropathy has increasingly attracted the attention of researchers. On the one hand, machine learning has made significant progress in the risk of progression and prognosis assessment of IgA nephropathy. Geddes' team [7, Geddes CC, Fox JG, Allison ME, Boulton-Jones JM, Simpson K. An artificial neural network can select patients at high risk of developing progressive IgA nephropathy more accurately than experienced nephrologists. Nephrol DialTransplant. 1998;13(1):67-71.] screened clinical and laboratory indicators from kidney biopsies of patients with IgA nephropathy and compared the prediction results of the artificial neural network (ANN) model with those of six nephrologists. The results showed that the ANN was more accurate than experienced nephrologists in predicting the risk of poor prognosis in IgA nephropathy. Pesce et al. [8、Pesce F,Diciolla M, Binetti G, Naso D, Ostuni VC, Di Noia T, Vågane AM, Bjørneklett R, Suzuki H, Tomino Y, Di Sciascio E, Schena FP. Clinical decision support system for end-stage kidney disease risk estimation in IgA nephropathy patients. Nephrol Dial Transplant. 2016;31(1):80-6.] developed a clinical decision support system (CDSS) using artificial neural networks to predict the risk of end-stage kidney disease (ESRD) in IgA patients using two ANN models.Subsequently, Liu et al. [9, Liu Y, Zhang Y, Liu D, Tan X, Tang X, Zhang F, Xia M, Chen G, He L, Zhou L, Zhu X, Liu H. Prediction of ESRD in IgA Nephropathy Patients from an Asian Cohort: A Random Forest Model. Kidney Blood Press Res. 2018;43(6):1852-64.] conducted a retrospective analysis of Asian IgAN patients and established a model for predicting ESRD using the random forest (RF) algorithm. The area under the curve (AUC) of this model was as high as 92.57%. In 2019, Barbour's research group developed the international IgAN prognostic assessment model (IIGAN-PT) based on Cox regression analysis. In 2021, the KDIGO guidelines recommended that clinicians use IIGAN-PT when making decisions [10, Rovin BH, Adler SG, Barratt J, et al. Executive summary of the KDIGO 2021 Guideline for the Management of Glomerular Diseases. KidneyInt. 2021;100(4):753-79.]. However, independent cohort data also show that the model does not perform well in external validation [11, Hwang D, Choi K, Cho NJ, Park S, Yu BC, Gil HW, Lee EY, Choi SJ, Park MY, KimJK, Hwang SD, Kwon SH, Jeon JS, Noh H, Han DC, Kim H. Validation of an international prediction model including the Oxford classification in Korean patients with IgA nephropathy. Nephrology (Carlton). 2021;26(7):594-602.].

[0005] On the other hand, machine learning has also gradually developed and improved in the field of diagnosis and prediction of IgA nephropathy. For example, an IgA nephropathy cohort study used the RF algorithm to establish an IgA nephropathy diagnostic model based on gut microbe data. The model achieved an AUC of 86.85% in the training set [12, Dong Y, Chen J, Zhang Y, Wang Z, Shang J, Zhao Z. Development and validation of diagnostic models for immunoglobulin Anephropathy based on gut microbes. Front Cell Infect Microbiol. 2022;12:1059692.]. Zhang et al. [13, Zhang X, Chao P, Jiang H, Yang S, Muhetaer G, Zhang J, Song X, Lu C. Integration of three machine learning algorithms identifies characteristic RNA-binding proteins linked with diagnosis, immunity and pyroptosis of IgA nephropathy. Front Genet. 2022;13:975521.] studied characteristic RNA-binding proteins (RBPs) in IgA nephropathy by integrating three machine learning algorithms: LASSO, RF, and SVM-RFE. Other studies have also conducted integrated proteomic analysis on plasma from patients with IgA nephropathy, and identified four potential diagnostic biomarkers from differential expression profiles using the LASSO feature screening model [14, Zhang D, Li Y, Liang M, Liang Y, Tian J, He Q, Yang B, Jin J, Zhu W. LC-MS / MS based metabolomics and proteomics reveal candidate biomarkers and molecular mechanism of early IgA nephropathy. ClinProteomics. 2022;19(1):51.].Meanwhile, our research group previously conducted a retrospective study using machine learning algorithm (XGBoost) to analyze 460 patients with IgA nephropathy and established a differential diagnosis model with an AUC value as high as 0.99 [15, Yang Han. Research on establishing a diagnostic and grading model for primary IgA nephropathy by combining clinical and laboratory data [D]. 2022.], but it has limitations such as limited sample size and lack of external validation. Gupta et al. [16, Gupta R, Sharma A, Singh S, DindaAK. Rule-based decision support system in the biopsy diagnosis of glomerular diseases. J Clin Pathol. 2011;64(10):862-5.] constructed a rule-driven diagnostic platform based on decision tree classifier to achieve automated diagnosis of glomerular diseases with an accuracy of 83.2%, but this system also has problems such as limited sample size, unverifiable generalization ability, rules that may not cover all clinical situations, and lack of external validation.

[0006] In addition, some studies have applied machine learning to the histopathological auxiliary diagnosis of IgA nephropathy through pathological image segmentation and morphological analysis techniques. Sato's team [17, Sato N, Uchino E, Kojima R, Sakuragi M, Hiragi S, Minamiguchi S, Haga H, Yokoi H, Yanagita M, Okuno Y. Evaluation of Kidney Histological Images Using Unsupervised Deep Learning. Kidney Int Rep.2021;6(9):2445-54.] retrospectively collected renal pathological images of 68 patients with IgA nephropathy. By combining unsupervised clustering with CNN, they were able to successfully extract features related to clinical variables from renal biopsy images. Simon et al. combined support vector machine (SVM) and CNN to collect kidney pathological slice information for model training [18, Simon O, Yacoub R, Jain S, Tomaszewski JE, Sarder P. Multi-radial LBP Features as a Tool for Rapid Glomerular Detection and Assessment in Whole Slide HistopathologyImages. Sci Rep. 2018;8(1):2032.].

[0007] However, existing diagnostic methods have the following drawbacks: 1. Current diagnostic methods rely on invasive biopsies. Renal biopsy, as the gold standard, has obvious limitations such as being invasive, complicated to operate, having a risk of complications, and not being suitable for repeated dynamic monitoring[2], which leads to a large number of patients being missed, especially in the early stages when the best treatment plan is lost due to the failure to detect the disease in time.

[0008] 2. Existing non-invasive diagnostic models have insufficient performance and generalization ability. (1) The sample size is small and single-center studies cannot reflect the heterogeneity of this disease. Most models based on conventional indicators (such as the studies by Gao et al. [3], Han et al. [5], and Meng Xiaohui et al. [6] mentioned above) have limited sample sizes (tens to hundreds of cases) and are single-center data, with bias in case selection and relatively weak representativeness.

[0009] (2) Lack of rigorous external verification The aforementioned studies [3, 5, 6] and most machine learning diagnostic studies [12, 13, 15, 16] generally lack independent multi-center external validation cohorts, and the generalization ability of the models is uncertain.

[0010] (3) The model performance is relatively insufficient. Traditional models have an AUC of 0.75-0.89 [3, 5, 6], which still has room for improvement; and some models are only applicable to people with normal renal function [4].

[0011] 3. Machine learning models suffer from a "black box" problem and have not been translated into clinical practice. (1) The diagnostic model has poor interpretability. Machine learning algorithms (such as ANN, RF, and XGBoost) have complex internal working mechanisms and lack transparency, and are regarded as "black box" models

[19] . The decision-making process is difficult to explain, which reduces the trust of clinicians and limits the application and promotion.

[0012] (2) The model cannot be applied in practice. Existing research (such as [12, 13, 15, 16, 17, 18]) mostly remains at the stage of model building and preliminary validation, without further development into convenient clinical tools, and lacks translation from model to clinical application.

[0013] (3) Lack of integration of novel and efficient biomarkers Existing studies mostly utilize routine laboratory indicators or specific omics data, and there are no reports of integrating the novel urinary biomarker FLRT2 validated in this study with clinical indicator systems and using interpretable machine learning to construct diagnostic systems. Summary of the Invention

[0014] The purpose of this invention is to provide a novel use for urinary FLRT2 (Fibronectin Leucine Rich Transmembrane Protein 2) protein, specifically its use in the preparation of kits, predictive models, or online systems for non-invasive diagnosis or risk assessment of IgA nephropathy. Specifically, it aims to overcome the deficiencies of the prior art by collecting multi-center clinical data, combining the urinary biomarker FLRT2 with routine laboratory indicators and clinical data, utilizing various machine learning algorithms to construct and optimize a high-performance, interpretable diagnostic model for IgA nephropathy, performing external validation, and ultimately building an online accessible risk prediction program. This achieves the clinical translation of a non-invasive diagnostic tool for IgA nephropathy, and the actual diagnostic value of FLRT2 is clarified through comparative evaluation.

[0015] The purpose of this invention is to disclose the application of urinary FLRT2 protein in the preparation of IgA nephropathy biomarkers.

[0016] The objective of this invention is achieved through the following technical solution: Application of urinary FLRT2 protein in the preparation of IgA nephropathy biomarkers.

[0017] The application described in the above technical solution, wherein: the application in the preparation of IgA nephropathy biomarkers refers to the application of urinary FLRT2 protein as a biomarker in the preparation of a predictive model for detecting IgA nephropathy.

[0018] In the application described in the above technical solution, the prediction model is the LightGBM model.

[0019] In the application described above, the input features in the LightGBM model are FLRT2 / Ucr, Age, eGFR, ACR, BUN, NAG(24h), SP, IgA, TC, ALB, and HA.

[0020] The application described in the above technical solution, wherein: the application in the preparation of IgA nephropathy biomarkers refers to the application of urinary FLRT2 protein as a biomarker in the preparation of an online diagnostic system for detecting IgA nephropathy.

[0021] The application described in the above technical solution, wherein: the application in the online system refers to deploying the trained LightGBM model as an accessible web application.

[0022] In the application described above, the input features in the LightGBM model are FLRT2 / Ucr, Age, eGFR, ACR, BUN, NAG(24h), SP, IgA, TC, ALB, and HA.

[0023] The application described in the above technical solution, wherein: the application in the preparation of IgA nephropathy biomarkers refers to the application of urinary FLRT2 protein as a biomarker in the preparation of a kit for detecting IgA nephropathy.

[0024] The technical solution provided by this invention has the following significant advantages and effects, which are impossible or difficult to achieve simultaneously with existing technologies: 1. Achieved high-precision non-invasive diagnosis The final model based on LightGBM constructed in this invention achieved excellent diagnostic performance on the internal validation set, with an AUC of 0.987, accuracy of 93.2%, sensitivity of 92.5%, specificity of 93.9%, and F1 score of 0.937. This performance is significantly better than most models based on conventional indicators reported in the background art (AUC 0.75-0.89) [3, 5, 6] and some early machine learning models [12, 16]. Existing models, whether traditional Logistic Regression [3, 5, 6] or rudimentary machine learning applications [12, 16], have an AUC of at most about 0.99 (but the sample size is only 460 cases and not validated

[15] ), or generally between 0.86 and 0.89. This invention achieves an approximately ultra-high AUC (0.987) on a larger-scale multicenter dataset and has undergone a more rigorous internal and external validation process, making its high performance more robust and convincing.

[0025] 2. Overcoming the limitations of the "black box" and providing explainable decision support. By integrating the SHAP interpretation framework, the model of this invention not only makes accurate predictions, but also makes the decision-making process transparent. Clinicians can understand how each prediction result is derived from the patient's 11 features (including FLRT2), which features play a key role and their direction of influence (positive or negative). This interpretability greatly enhances the credibility and acceptance of the model in clinical practice, and solves the core defect of machine learning models in the background art that are difficult to interpret and thus difficult to promote

[19] . Machine learning diagnostic models in the background art, such as RF models based on gut microbiota

[12] , ensemble models based on RBP

[13] , or decision tree platforms

[16] , have not applied advanced interpretability methods such as SHAP, and their decision-making process is still a "black box", which clinicians cannot trust and adopt.

[0026] 3. Complete clinical translation and provide real-time online tools. Unlike the general "modeling terminates" approach in background research [12, 13, 15, 16], this invention successfully transforms the model into a real-time, convenient online web calculator. Clinicians or researchers can access it through a browser, input parameters, and immediately obtain risk assessment results, achieving immediate availability of diagnostic aids—something that has never been achieved in existing technologies. Currently, all reported machine learning diagnostic studies of IgA nephropathy [12, 13, 14, 15, 16, 17, 18] have not provided publicly accessible and usable online tools; their results remain only at the level of papers and algorithms, unable to directly serve clinical practice.

[0027] 4. The unique added value of the FLRT2 marker was verified. Rigorous comparative analysis confirmed that adding the FLRT2 / Ucr ratio effectively improves the diagnostic capability of the model. With the same sample size (365 cases), Model 1, which included FLRT2, showed an AUC increase from 0.880 to 0.912 and a significant increase in sensitivity from 73.5% to 81.6% compared to Model 2, which did not. Net Reclassification Index (NRI) analysis showed a significant improvement in reclassification of event samples (IgAN patients) (Event NRI = 0.079, p = 0.042). This indicates that FLRT2, as a novel urinary biomarker, provides diagnostic information independent of traditional clinical indicators, particularly helpful in improving the sensitivity of detecting true patients—a value that existing technology models [3, 4, 5, 6, 12, 16] that do not integrate FLRT2 cannot provide. Existing models do not involve the FLRT2 biomarker and therefore cannot achieve the diagnostic sensitivity improvement and patient reclassification enhancement it brings.

[0028] 5. Possesses strong generalization ability and robustness. This invention was developed based on multi-center data (5 hospitals) and specifically validated using external test sets from different hospitals. Although the AUC of the external test set (0.778) decreased compared to internal validation, revealing the real challenge of cross-center data discrepancies, this process itself demonstrates the necessity of external validation and the forward-looking nature of this approach. Furthermore, the stable results of 50% and 100% cross-validation (mean AUCs of 0.92±0.02 and 0.91±0.04, respectively), and the stable model performance after expanding the sample size to 2029 cases (Model 3 AUC 0.901), all demonstrate the robustness of the model architecture and selected features, outperforming most existing single-center, small-sample techniques. Most existing studies [3, 5, 6, 12, 15] lack multi-center external validation, and their claimed performance may significantly decrease in new patient populations. This invention, however, explicitly evaluates and demonstrates the model's generalization potential and challenges through external testing.

[0029] 6. It provides an efficient and low-cost screening pathway for potential candidates. This invention's model requires only routine laboratory test results and a single urine ELISA test (FLRT2) to achieve efficient risk assessment. For patients with a high predicted risk, a kidney biopsy can be recommended more specifically; for low-risk patients, unnecessary invasive procedures can be avoided. This tiered management strategy helps optimize the allocation of medical resources and reduce overall healthcare costs. Attached Figure Description

[0030] Figure 1 The differences in urinary FLRT2 protein levels among IgA nephropathy, non-IgA nephropathy, and healthy individuals; Figure 2 ROC curves of FLRT2 in different control groups; Figure 3 For the localization and quantitative analysis of FLRT2 tissue; among which Figure 3 Image A shows the immunofluorescence of IgA nephropathy tissue from SD rats. DAPI (blue): cell nuclei; FITC (green): FLRT2; Merge: overlap of the first two images. Figure 3 A shows that FLRT2 protein is localized in renal tubular epithelial cells; Figure 3 B is the quantitative immunofluorescence image of rat IgA nephropathy tissue. ， *express p <0.05, Figure 3 B showed that the immunofluorescence of rat IgA nephropathy tissue was significantly higher than that of the control group; Figure 4 Develop a mind map for the model; Figure 5 Figure A shows the performance evaluation results of the LightGBM model. Among them: Figure A: Five-fold cross-validation: The data is divided into 5 parts, and 4 parts are used for training and 1 part for validation each time. The results are repeated 5 times and the average validation result is taken. Figure B: Ten-fold cross-validation: The data is divided into 10 parts, and 9 parts are used for training and 1 part for validation each time. The results are repeated 10 times and the average validation result is taken. Figure 6 Figure A shows the ROC curves of the LightGBM model after external validation; Figure B shows the ROC curves of the internal validation set and the external validation set. Figure 7 Example of a web calculator. Detailed Implementation

[0031] To facilitate understanding of the technical solution of the present invention, the application of urinary FLRT2 protein in the preparation of IgA nephropathy biomarkers will be further explained below with reference to specific embodiments.

[0032] Example 1: Validation of the ELISA detection and diagnostic efficacy of FLRT2 in urine samples 1.1 Experimental Materials Urine samples: 292 patients with IgA nephropathy diagnosed by renal biopsy (IgAN group), 254 patients with non-IgA nephropathy (NIgAN group), and 67 healthy volunteers (NC group).

[0033] Reagent kit: Human FLRT2 ELISA kit (Wuhan Sanlujie Biotechnology, catalog number SLJ-H0445).

[0034] Instruments and equipment: Microplate reader (Synergy HTX, Biotek), centrifuge (Sorvall Legend Micro 17R, Thermo Scientific).

[0035] 1.2 Experimental Procedure 1.2.1 Sample Processing 1. Take a fresh urine sample or one stored at -80°C and thaw it at 4°C.

[0036] 2. Centrifuge at 1000 rpm for 10 minutes and collect the supernatant for later use.

[0037] 3. Use a creatinine assay kit (enzymatic method) to simultaneously measure the urinary creatinine concentration.

[0038] 1.2.2 ELISA Operation 1. Reagent preparation: Dilute 20× washing solution to 1× working solution; dilute the standard with standard diluent in a gradient according to the instructions (0, 0.31, 0.625, 1.25, 2.5, 5, 10, 20 ng / mL).

[0039] 2. Add sample: Add 100 μL of standard or sample to each well and incubate at 37°C for 1 hour.

[0040] 3. Washing: Discard the liquid in the well, add 250 μL of washing solution to each well, repeat washing 3 times, and pat dry.

[0041] 4. Add detection antibody: Add 100 μL of biotin-labeled detection antibody to each well and incubate at 37°C for 1 hour.

[0042] 5. Washing: Same as step 3, wash 3 times.

[0043] 6. Add enzyme conjugate: Add 100 μL of horseradish peroxidase-labeled streptavidin to each well and incubate at 37°C for 30 minutes.

[0044] 7. Washing: Same as step 3, wash 5 times.

[0045] 8. Color development: Add 100 μL of TMB color development solution to each well and incubate at 37°C in the dark for 15 minutes.

[0046] 9. Termination and detection: Add 50 μL of stop solution to each well and read the absorbance value at a wavelength of 450 nm.

[0047] 1.2.3 Data Processing 1. Plot a four-parameter logic (4-PL) standard curve (R² > 0.99) with the concentration of the standard as the x-axis and the absorbance value as the y-axis.

[0048] 2. Calculate the FLRT2 concentration in urine based on the standard curve.

[0049] 3. Correct FLRT2 concentration using urine creatinine: FLRT2 / Ucr (10-³ mg / g) = FLRT2 concentration (ng / mL) ÷ urine creatinine (g / L) × 1000.

[0050] 1.3 Experimental Results Concentration difference results as follows Figure 1 As shown, the median FLRT2 / Ucr in the IgAN group was 0.842 (0.367, 1.80), which was significantly higher than that in the NIGAN group (0.552) and the NC group (0.285) (p < 0.001).

[0051] Diagnostic efficacy results such as Figure 2 As shown: - IgAN vs NC: AUC = 0.773, cut-off = 0.428 × 10-³ mg / g, sensitivity 76.12%, specificity 72.26%.

[0052] - IgAN vs NIgAN: AUC = 0.632, cut-off = 0.643 × 10-³ mg / g, sensitivity 61.1%, specificity 60.6%.

[0053] Example 2: Tissue localization of FLRT2 expression in a rat model of IgA nephropathy 2.1 Animal Model Construction Animals: 18 male SD rats (SPF grade, 6-8 weeks old, weighing 200-220 g).

[0054] Grouping: The animals were randomly divided into a normal control group (NC group, 9 animals) and an IgAN model group (9 animals).

[0055] Modeling method: 1. The model group was given 100 g / L BSA solution (4 mL / kg) by gavage once every other day for 6 consecutive weeks.

[0056] 2. Administer a subcutaneous injection of a mixture of CCl4 (0.1 mL) and castor oil (0.5 mL) once a week for 9 consecutive weeks.

[0057] 3. LPS (0.25 g / L, 0.2 mL / animal) was injected via the tail vein at weeks 6 and 8.

[0058] Sample collection: In week 10, 3 rats from each group were randomly sacrificed for pathological verification; in week 11, all remaining rats were sacrificed and kidney tissue was collected.

[0059] 2.2 Immunofluorescence double labeling assay 2.2.1 Organizational Processing 1. Kidney tissue was collected and fixed in 4% paraformaldehyde for 24 hours.

[0060] 2. Embedded in paraffin and sectioned (4 μm thick).

[0061] 3. Dewaxing and hydration: Xylene (2 × 10 min) → Anhydrous ethanol (2 × 5 min) → 95%, 80%, and 70% ethanol for 5 min each → PBS rinse.

[0062] 2.2.2 Antigen retrieval 1. Immerse the slices in pH 6.0 citrate antigen retrieval solution and microwave for 10 minutes.

[0063] 2. Allow to cool naturally to room temperature, then rinse 3 times with PBS.

[0064] 2.2.3 Immunofluorescence staining 1. Blocking: Add 5% goat serum and block at room temperature for 1 hour.

[0065] 2. Primary antibody incubation: - Rabbit anti-FLRT2 antibody (1:200) Incubate overnight at 4°C.

[0066] 3. Washing: Rinse 3 times with PBS, 5 minutes each time.

[0067] 4. Secondary antibody incubation: - FITC-labeled goat anti-rabbit IgG (1:500) - Incubate at room temperature in the dark for 1 hour.

[0068] 5. Washing: Same as step 3.

[0069] 6. Mounting: Add mounting medium containing DAPI, cover with a coverslip, and store in the dark.

[0070] 2.2.4 Image Acquisition and Analysis 1. Images were acquired using an upright fluorescence microscope (Nikon Eclipse Ci-S).

[0071] 2. Use ImageJ software to analyze fluorescence colocalization intensity.

[0072] 2.3 Experimental Results The tissue localization results of FLRT2 expression in the IgA nephropathy rat model are as follows: Figure 3 As shown, FLRT2 expression in renal tubular epithelial cells of the IgAN model group was significantly higher than that of the NC group (p < 0.05).

[0073] Example 3: Specific experiments on FLRT2 protein detection and model construction like Figure 4 The model development approach shown in this example details the entire process from urine sample processing to LightGBM model construction.

[0074] 3.1. Sample Preparation A total of 365 morning urine samples (195 IgAN cases and 170 non-IgAN cases) were collected from the First Affiliated Hospital of Kunming Medical University. All patients were diagnosed by renal biopsy pathology. Samples were centrifuged at 4°C (3000 rpm, 15 minutes) within 2 hours, and the supernatant was aliquoted and stored at -80°C.

[0075] 3.2. FLRT2 Concentration Detection (1) The detection was performed using a commercial human FLRT2 ELISA kit (e.g., Elabscience, catalog number E-EL-H6034). All samples and standards were tested in duplicate.

[0076] (2) Preparation of standard curve The standard was diluted sequentially with sample diluent to concentrations of 2000, 1000, 500, 250, 125, 62.5, 31.25, and 0 pg / mL.

[0077] (3) Detection steps Strictly follow the kit instructions. A brief description is as follows: Add 100 μL of standard or a 1:10 diluted urine sample to each well of a microplate pre-coated with antibody, and incubate at 37°C for 90 minutes; wash the plate 5 times; add 100 μL of biotinylated detection antibody to each well, and incubate at 37°C for 60 minutes; wash the plate 5 times; add 100 μL of horseradish peroxidase-labeled streptavidin to each well, and incubate at 37°C for 30 minutes; wash the plate 7 times; add 90 μL of TMB substrate to each well, and develop the color at 37°C in the dark for 15 minutes; add 50 μL of stop solution (2M H2SO4) to each well. Immediately measure the absorbance (OD value) of each well using a microplate reader at 450 nm. Plot a four-parameter logic (4-PL) standard curve with standard concentration on the x-axis and OD value on the y-axis. Calculate the FLRT2 concentration (pg / mL) from the standard curve based on the sample OD value.

[0078] The concentration of urinary creatinine (Ucr) in the same sample was measured using a fully automated biochemical analyzer (sarcosine oxidase method). The FLRT2 / Ucr ratio was calculated (unit: pg / mmol).

[0079] 3.3. Clinical and Laboratory Data Collection (1) General clinical data: including the patient’s age (years), gender (male / female), whether they are from ethnic minorities, kidney biopsy results, height (cm), weight (Kg), blood pressure (mmHg), duration of hypertension (years), whether they have malignant tumors, infectious diseases, autoimmune diseases, and smoking history (years).

[0080] (2) Laboratory indicators: including urinary bilirubin (BLD), urinary protein (PRO), white blood cell count (WBC), hemoglobin (Hb), total protein (TP), albumin (AlB), blood urea nitrogen (BUN), creatinine (Cre), uric acid (UA), blood glucose (Glu), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), serum potassium (K), serum IgA (IgA), serum sodium (Na), glomerular filtration rate (eGFR), complement C3 (C3), complement C4 (C4), urinary creatinine (UCr), urinary N-acetyl-beta-D-glucosidase (NAG), urinary total protein (UTP), urinary albumin (UALB), urinary microalbumin to creatinine ratio (ACR), and serum IgG (IgG).

[0081] 3.4. Data Preprocessing (1) Handling missing values Remove variables with an overall missing value rate >20% (such as random urine NAG, K, Na, etc.). For the remaining variables, fill missing values ​​with the median for continuous variables and the mode for categorical variables.

[0082] (2) Outlier handling For continuous variables, calculate their first quartile (Q1) and third quartile (Q3). Values ​​less than Q1-1.5IQR or greater than Q3+1.5IQR are considered outliers and replaced with the corresponding lower or upper bound values.

[0083] (3) Coding and Standardization Categorical variables such as "hematuria (HA)" are coded as 0 / 1. All continuous variables are standardized using Z-score.

[0084] (4) Split the dataset The 365 data points were randomly divided into a training set (255 cases) and an internal validation set (110 cases) in a 7:3 ratio, with a random seed of 42.

[0085] 3.5. Machine Learning Model Training and Evaluation (Taking LightGBM as an example) Environment and Reservoir Python 3.8, LightGBM 4.2.0, scikit-learn 1.4.2.

[0086] Code Implementation Python import lightgbm as lgb from sklearn.model_selection import train_test_split from sklearn.metrics import roc_auc_score, accuracy_score,recall_score, confusion_matrix # Assume X_train and y_train are the features and labels of the training set, and X_val and y_val are the features and labels of the validation set. # Define LightGBM parameters params = { 'objective': 'binary', 'metric': 'auc', 'boosting_type': 'gbdt', 'num_leaves': 31, 'max_depth': -1, 'learning_rate': 0.1, 'feature_fraction': 0.9, 'bagging_fraction': 0.8, 'bagging_freq': 5, 'min_child_samples': 20, 'reg_alpha': 0.1, 'reg_lambda': 0.1, 'random_state': 42, 'verbose': -1 } # Create a dataset train_data = lgb.Dataset(X_train, label=y_train) val_data = lgb.Dataset(X_val, label=y_val, reference=train_data) # Training the model, early stopping to prevent overfitting model = lgb.train(params, train_data, valid_sets=[val_data], num_boost_round=1000, early_stopping_rounds=50, verbose_eval=False) # Predict and evaluate on the validation set y_val_pred_prob = model.predict(X_val, num_iteration=model.best_iteration) auc = roc_auc_score(y_val, y_val_pred_prob) y_val_pred = (y_val_pred_prob >= 0.5).astype(int) accuracy = accuracy_score(y_val, y_val_pred) recall = recall_score(y_val, y_val_pred) # Sensitivity tn, fp, fn, tp = confusion_matrix(y_val, y_val_pred).ravel() specificity = tn / (tn + fp) # Specificity print(f"Validation set AUC: {auc:.3f}, Accuracy: {accuracy:.3f}, Sensitivity: {recall:.3f}, Specificity: {specificity:.3f}") ``` result After running the above code, the model is obtained after 50-fold and 100-fold cross-validation. Figure 5 ) and internal validation set (110 cases, Figure 6 Performance after evaluation: AUC = 0.987, Accuracy = 0.932, Sensitivity = 0.925, Specificity = 0.939.

[0087] Example 4: Specific steps for deploying and using an online web calculator This example demonstrates how to deploy a trained LightGBM model as an accessible web application.

[0088] 4.1. Model Saving Save the trained `model` from Example 3 as a file `igan_lightgbm_model.pkl` using the `pickle` library.

[0089] 4.2. Web Application Development (1) Environment Python 3.8, Gradio 5.0.

[0090] (2) Code (`app.py`): Python import gradio as gr import pickle import numpy as np # 1. Load the model and feature scale (for denormalization) with open('igan_lightgbm_model.pkl', 'rb') as f: model = pickle.load(f) # Assuming the mean and standard deviation of the training set features have been saved, we will use them to perform the same standardization on new inputs. # with open('scaler_params.pkl', 'rb') as f: # scaler_means, scaler_stds = pickle.load(f) # 2. Define the prediction function def predict_igan(age, egfr, acr, flrt2_ucr, bun, nag_24h,systolic_bp, iga, tc, alb, ha): """ Input 11 features and output the predicted probability and risk level.

[0091] #3. Creating the Grado Interface with gr.Blocks(title="IgA Nephropathy Non-invasive Risk Assessment System", theme=gr.themes.Soft()) as demo: gr.Markdown("## IgA Nephropathy Non-invasive Risk Assessment System (Based on LightGBM Model)") gr.Markdown("Please enter the following clinical and laboratory test results:") with gr.Row(): with gr.Column(scale=1): age = gr.Number(label="age (years)", value=35,minimum=10, maximum=100) egfr = gr.Number(label="Estimated glomerular filtration rate eGFR(mL / min / 1.73m²)", value=75.0) acr = gr.Number(label="urine albumin / creatinine ratio ACR (mg / g)", value=800.0) flrt2_ucr = gr.Number(label="Urine FLRT2 / creatinine ratio (pg / mmol)", value=30.0) bun = gr.Number(label="Blood Urea Nitrogen BUN (mmol / L)", value=6.0) with gr.Column(scale=1): nag_24h = gr.Number(label="24-hour urine NAG (U / L)", value=15.0) systolic_bp = gr.Number(label="systolic blood pressure (mmHg)", value=120.0) iga = gr.Number(label="Serve Immunoglobulin A IgA (g / L)", value=2.9) tc = gr.Number(label="Total cholesterol TC (mmol / L)", value=4.8) alb = gr.Number(label="Serum Albumin ALB (g / L)",value=40.0) with gr.Column(scale=1): ha = gr.Radio(["Yes", "No"], label="Hematuria (HA)", value="Yes") submit_btn = gr.Button("Start Evaluation", variant="primary", size="lg") # Output area with gr.Row(): prob_output = gr.Textbox(label="prediction result") risk_output = gr.Textbox(label="Risk Assessment") suggestion_output = gr.Textbox(label="suggestion") # Bind button click event submit_btn.click(fn=predict_igan, inputs=[age, egfr, acr, flrt2_ucr, bun, nag_24h, systolic_bp, iga,tc, alb, ha], outputs=[prob_output, risk_output, suggestion_output]) gr.Markdown("---") gr.Markdown(" # 4. Launch the application if __name__ == "__main__": demo.launch(server_name="0.0.0.0", server_port=7860, share=True) # share=True generates a temporary public IP address ``` 4.3. Deployment and Access: Run `python app.py` on the server or locally.

[0092] The console will output a local URL (such as `http: / / 127.0.0.1:7860`) and a temporary public Gradio link (such as `https: / / xxxxxx.gradio.live`).

[0093] Users can access this link through a browser and enter real patient data: age 34, eGFR 76.1, ACR 654.2, FLRT2 / Ucr 29.1, BUN 6.1, NAG (24h) 14.4, systolic blood pressure 122, IgA 2.89, TC 4.75, ALB 40.7, hematuria "present". After clicking "Start Assessment", the interface displays: "Predicted probability of IgA nephropathy: 99.20%", "Risk level: High risk", "Clinical recommendation: It is recommended to consult a nephrologist as soon as possible and consider a kidney biopsy to confirm the diagnosis." Figure 7 ) Example 5: Quantitative Assessment Experiment of the Diagnostic Value of FLRT2 This embodiment demonstrates in detail how to quantify the contribution of FLRT2 using a comparative model.

[0094] 5.1. Dataset Construction Model 1 (incorporating FLRT2 data) Same as in Example 3, using 365 samples and their 11 features (including FLRT2 / Ucr).

[0095] Model 3 (excluding FLRT2 data) Data from 2029 patients diagnosed by renal biopsy between January 2019 and December 2023 (1019 cases of IgAN and 1010 cases of non-IgAN) were extracted from the electronic medical record system of the First Affiliated Hospital of Kunming Medical University. Ten features other than FLRT2 / Ucr were collected (Age, eGFR, ACR, BUN, NAG(24h), SP, IgA, TC, ALB, HA). The same data preprocessing as in Example 3 was performed.

[0096] 5.2. Model Training (1) For the two datasets, divide them into training and test sets in an 8:2 ratio.

[0097] (2) Train Model 1 and Model 3 using the same LightGBM hyperparameters (`params`) as in Example 3.

[0098] (3) Early stop method was used during training, and the final performance was evaluated on an independent test set.

[0099] 5.3. Performance Evaluation and Statistical Analysis Specific results Model 1 (included in FLRT2, n=365 test set): AUC = 0.912, sensitivity = 0.816, specificity = 0.882.

[0100] Model 3 (excluding FLRT2, n=2029 test set): AUC = 0.901, sensitivity = 0.737, specificity = 1.000.

[0101] NRI analysis: Event NRI = 0.079 (p=0.042), Non-event NRI = -0.118 (p=0.098), Total NRI = -0.039 (p=0.320). The event NRI was significantly positive, indicating that FLRT2 helped the model more accurately reclassify more real IgA nephropathy patients.

[0102] The IDI was 0.018 (p=0.782), which was positive but not statistically significant, possibly due to differences in sample size or variation.

[0103] 5.4. Experimental Conclusions This example, through concrete data and code demonstration, quantitatively proves that adding FLRT2 significantly improves the diagnostic model's ability to identify patients with IgA nephropathy (sensitivity improvement of approximately 8%) while maintaining high specificity. The significant improvement in event NRI indicates that FLRT2 has independent diagnostic incremental value, which cannot be achieved by existing technology models that do not include this biomarker.

[0104] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention in any form or substance. Any modifications, alterations, and variations made by those skilled in the art without departing from the scope of the present invention using the disclosed technical content are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, and variations made to the above embodiments based on the essential technology of the present invention are still within the scope of the present invention.

Claims

1. Application of urinary FLRT2 protein in the preparation of IgA nephropathy biomarkers.

2. The application according to claim 1, characterized in that: The application mentioned in the preparation of IgA nephropathy biomarkers refers to the use of urinary FLRT2 protein as a biomarker in the preparation of a predictive model for detecting IgA nephropathy.

3. The application according to claim 2, characterized in that: The prediction model is the LightGBM model.

4. The application according to claim 3, characterized in that: The input features of the LightGBM model are FLRT2 / Ucr, Age, eGFR, ACR, BUN, NAG(24h), SP, IgA, TC, ALB, and HA.

5. The application according to claim 1, characterized in that: The application mentioned in the preparation of IgA nephropathy biomarkers refers to the use of urinary FLRT2 protein as a biomarker in the preparation of an online diagnostic system for IgA nephropathy.

6. The application according to claim 5, characterized in that: The application in the online system refers to deploying the trained LightGBM model as an accessible web application.

7. The application according to claim 6, characterized in that: The input features of the LightGBM model are FLRT2 / Ucr, Age, eGFR, ACR, BUN, NAG(24h), SP, IgA, TC, ALB, and HA.

8. The application according to claim 1, characterized in that: The application mentioned in the preparation of IgA nephropathy biomarkers refers to the use of urinary FLRT2 protein as a biomarker in the preparation of a kit for detecting IgA nephropathy.

Images