A kit for osteoporosis detection

Abstract

This invention belongs to the field of biotechnology, and specifically relates to a kit for osteoporosis detection. This invention is the first to discover that a combination of six proteins—FST, NRIP1, OSBPL11, Cystatin SN, CEBPD, and NACC1—can serve as a plasma biomarker for osteoporosis. This invention combines protein chip detection with a machine learning model, outputting the probability of disease through a random forest model, with a clearly defined detection threshold (0.628) and high operability. The kit enables high-throughput detection, allowing simultaneous detection of multiple biomarkers in a single experiment, requiring small sample volumes, and is suitable for large-scale screening. Clinical sample validation shows that this kit has a sensitivity of over 75.58% and a specificity of over 76.67% for osteoporosis detection.

Description

Technical Field

[0001] This invention belongs to the field of biotechnology, and in particular relates to a kit for osteoporosis detection. Background Technology

[0002] Osteoporosis is a systemic skeletal disease characterized by decreased bone mass and destruction of bone microstructure, leading to increased bone fragility and susceptibility to fractures. Osteoporosis is a skeletal disease closely related to aging, and its incidence increases with age. Laboratory tests for osteoporosis mainly include the following: serum calcium, serum phosphorus, and alkaline phosphatase levels are usually normal in primary osteoporosis, but alkaline phosphatase may be elevated within months after a fracture; serum parathyroid hormone testing is used to rule out secondary osteoporosis, and the hormone level may be normal or elevated in primary patients; bone turnover markers (including bone-specific alkaline phosphatase, osteocalcin, and type I procollagen peptides reflecting bone formation, as well as tartrate-resistant acid phosphatase, urinary pyridinium and deoxypyridinium, and NC-terminal cross-linked peptides of type I collagen reflecting bone resorption) can assess bone metabolism status; the normal morning urine calcium / creatinine ratio is approximately 0.13±0.01, and an increased ratio suggests excessive urinary calcium excretion, which may reflect an increased bone resorption rate.

[0003] In terms of auxiliary examinations, bone imaging (X-ray) can detect fractures and lesions such as osteoarthritis, intervertebral disc disease, and anterior displacement of the spine. When bone mass is reduced, increased bone translucency, reduced trabeculae with widened gaps, disappearance of transverse trabeculae, and blurred bone structure can be observed. However, it is usually necessary for bone mass to decrease by more than 30% to be observed. The typical manifestation is biconcave deformation of the vertebral body or collapse of the anterior edge into a wedge shape (i.e., compression fracture, commonly seen in the 11th and 12th thoracic vertebrae and the 1st and 2nd lumbar vertebrae). Bone mineral density testing is a predictive indicator of fracture risk. Measuring any part can assess the overall fracture risk, while measuring a specific part can help predict the local fracture risk of that part.

[0004] In recent years, research on osteoporosis detection has become increasingly in-depth. Currently, the main methods for detecting osteoporosis are bone mineral density, including dual-energy X-ray absorptiometry (DXA), ultrasound bone mineral density, and quantitative CT of the lumbar spine (QCT). There is no direct method for detecting plasma in clinical practice. Therefore, it is necessary to develop a kit for osteoporosis detection to overcome the problems that urgently need to be solved in the existing technology and achieve high-throughput, high-sensitivity, high-specificity, and low-cost detection of multiple biomarkers. Summary of the Invention

[0005] The purpose of this invention is to provide a kit for osteoporosis detection that can rapidly, simply, and accurately determine the expression levels of 6-26 protein biomarkers in a single experiment, without requiring expensive detection instruments. Specifically, this invention provides a combined osteoporosis detection kit and detection method. The combined detection kit contains reagents that specifically detect the concentrations of the protein biomarkers FST, NRIP1, OSBPL11, Cystatin SN, CEBPD, and NACC1. The false positive results obtained using this kit are extremely low.

[0006] Furthermore, this invention combines protein chip detection with a machine learning model, outputting the disease probability through a random forest model. The detection threshold is clearly defined (0.628), and the process is highly operable. The kit of this invention enables high-throughput detection, allowing for the simultaneous detection of multiple biomarkers in a single experiment, with a small sample volume, making it suitable for large-scale screening.

[0007] In this regard, the present invention includes, but is not limited to, the following: In some aspects, the present invention provides the use of reagents for the specific detection of protein biomarkers in the preparation of kits for the detection of primary osteoporosis, said protein biomarkers including: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD and NACC1.

[0008] In some aspects, the present invention also provides the use of reagents for the specific detection of protein biomarkers in the preparation of kits for use in methods for detecting primary osteoporosis in test subjects, said protein biomarkers including: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1, and said methods comprising: (1) The test subject obtains the sample to be tested; (2) Mix the sample to be tested with reagents for the specific detection of protein markers; and (3) Obtain the concentration of each protein marker in the sample to be tested, calculate the probability of disease, and when the probability of disease is greater than or equal to the threshold, it indicates that the sample to be tested comes from an osteoporosis patient or has an osteoporosis risk.

[0009] In some specific aspects, the present invention calculates the probability of disease by incorporating the concentration of protein biomarkers into a random forest model.

[0010] In some specific aspects, the threshold described in this invention is obtained by using the concentration data of these six protein markers (FST, NRIP1, OSBPL11, Cystatin SN, CEBPD, and NACC1) detected in samples (e.g., 500 samples in Example 2) as initial parameters for model training. In more specific aspects, the threshold is 0.628, with a probability of 0.628 as the cutoff value: when the model outputs a disease probability ≥ 0.628, it is determined to be high-risk / having osteoporosis; when the probability < 0.628, it is determined to be healthy.

[0011] In some specific aspects, the core formula of the random forest model of this invention is as follows:

[0012] in: T Number of decision trees (set to 500 in this model); I t ( x ): Whether the t-th tree classifies sample x as osteoporosis (yes = 1, no = 0); P ( Y =1): The probability of disease after multiple trees vote.

[0013] In other aspects, the present invention provides a kit for detecting primary osteoporosis, the kit comprising reagents for specifically detecting protein biomarkers including: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1.

[0014] In some aspects, the reagent for specifically detecting protein biomarkers described in this invention is an antibody that specifically binds to the protein biomarker.

[0015] In some aspects, the protein biomarkers described in this invention also include COG8, CORO1A, CREBBP, ELMO1, EphA5, ITGA7, PDIA2, SIX2, or TSKU, or any combination thereof.

[0016] In some aspects, the protein biomarkers described in this invention include Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, NACC1, COG8, CORO1A, CREBBP, ELMO1, EphA5, ITGA7, PDIA2, SIX2, and TSKU.

[0017] In some aspects, the protein markers described in this invention are Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, NACC1, COG8, CORO1A, CREBBP, ELMO1, EphA5, ITGA7, PDIA2, SIX2, and TSKU.

[0018] In some aspects, the protein biomarkers described in this invention are selected from multiple or all of the following 26 osteoporosis biomarker-related antibodies: CEBPD, Cystatin SN, FST, NACC1, NRIP1, OSBPL11, LY6E, NDUFB1, COG8, CORO1A, CREBBP, ELMO1, EphA5, ITGA7, PDIA2, SIX2, TSKU, HTRA4, METTL3, MUC2, NIPBL, OVOL1, PKD2L1, RIN3, SOX8, and ST14.

[0019] In some respects, in the method described in this invention, in step (1), the sample to be tested is plasma, serum, urine, somatic cell culture or tissue; And / or, in step (2), the capture antibody for the specific detection protein marker is immobilized onto the chip to prepare an antibody chip; And / or, in step (3), a biotin-conjugated detection antibody and a fluorescein-conjugated streptavidin are used to detect the protein markers conjugated on the antibody chip.

[0020] In some respects, the sample to be tested is plasma.

[0021] In some respects, in the method of the present invention, in step (1), the sample to be tested is plasma; And / or, in step (2), capture antibodies that specifically detect each protein marker are immobilized onto the chip at a concentration of 0.01~2 ng; And / or, in step (3), the fluorescein is selected from Cy3, Cy5 or FITC.

[0022] In some respects, the capture antibody and / or detection antibody of the present invention are each independently selected from the antibodies listed in Table 5.

[0023] In other aspects, the present invention provides the application of protein biomarkers in constructing predictive models for primary osteoporosis, said protein biomarkers including: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD and NACC1.

[0024] In other aspects, the present invention also provides the use of protein biomarkers in the preparation of diagnostic reagents for primary osteoporosis, said protein biomarkers including: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD and NACC1.

[0025] In some aspects, the detection reagent products of the present invention include equipment for analyzing the content of the protein markers.

[0026] In some aspects, the primary osteoporosis described in this invention is postmenopausal osteoporosis and / or senile osteoporosis. In particular, senile osteoporosis.

[0027] Postmenopausal osteoporosis is a systemic metabolic bone disease that occurs in women after natural or induced menopause. Due to ovarian failure and a sharp decline in estrogen levels, bone metabolism becomes imbalanced, with bone resorption far exceeding bone formation. This leads to rapid bone loss, damage to bone microstructure, and a significant increase in bone fragility, ultimately increasing the risk of fractures. At this time, bone metabolism is in a high-turnover state, with overall bone metabolism being more active. Both bone resorption and bone formation levels are increased, but the increase in bone resorption is more significant, exceeding the increase in bone formation, thus causing bone loss.

[0028] Osteoporosis in the elderly is a systemic metabolic bone disease fundamentally based on natural aging. Accompanying age, multiple factors such as degenerative changes in overall bodily function, declining bone formation capacity, impaired calcium / vitamin D absorption, hormonal imbalances, and insufficient exercise and nutrition work together to lead to a sustained and slow decrease in bone mass, degeneration of the entire skeletal structure, decreased bone strength, and an increased risk of fractures. At this stage, bone metabolism is in a low-turnover state, with overall bone metabolism being low. Bone formation capacity is significantly impaired, while bone resorption capacity is slightly increased. Insufficient bone regeneration capacity results in age-related bone degeneration.

[0029] The osteoporosis detection chip kit of this invention uses a modified sandwich enzyme-linked immunosorbent assay (ELISA) to determine osteoporosis biomarkers. The specific antibody binds to the target antigen (i.e., the protein biomarker) in the sample, and the screened detection antibody binds to another region of the target antigen to form a stable compound. This kit can simultaneously detect 6 to 26 osteoporosis-related biomarker proteins. The combined antibody chip provides better detection results than using a single biomarker.

[0030] The slide also includes positive control wells and negative control wells; the positive control wells contain biotinylated IgG antibodies of corresponding specific antibodies at corresponding concentrations, and the concentration of biotinylated IgG antibodies in each reaction is consistent to facilitate standardized detection; the negative control wells contain biotinylated irrelevant antibodies of corresponding concentrations, and the concentration of biotinylated irrelevant antibodies in each reaction is consistent to facilitate standardized detection.

[0031] The kit described in this invention uses fluorescein as a detection signal to determine the concentration of the target antigen in the sample by comparing the signals of an unknown sample and a standard sample.

[0032] The kit described in this invention uses Genepix Pro 6.1 software to obtain the signal intensity of all points on the chip, and uses a self-developed spreadsheet-based analysis software to calculate the expression level and standard curve of each protein. The positive control signal is standardized, and the background threshold is set to the average intensity of the control group plus 2 times the standard deviation (SD) for statistical analysis. Proteins with expression levels higher than the background plus 2xSD are used for corrected T-test analysis.

[0033] The following is an introduction to 15 protein biomarkers: CEBPD (CCAAT / enhancer-binding protein δ): uniprot ID: P49716, as a key bone formation promoting factor, is at the core of the transcriptional network regulating osteoblast differentiation. Studies have shown that upregulation of CEBPD is necessary for human osteoblast differentiation and bone formation, and its dysfunction may lead to insufficient bone formation.

[0034] Cystatin SN (CST1, cysteine ​​protease inhibitor SN): uniprot ID: P01037, encoded by the CST1 gene, belongs to the type 2 cystatins family. It inhibits cysteine ​​protease activity and is mainly found in saliva, tears, urine, and semen, exerting a protective function.

[0035] FST (follicle-stabilizing hormone): uniprot ID: P19883, as an endogenous antagonist of the TGF-β superfamily, in animal models, molecularly designed based on FST dose-dependently increases trabecular bone mass (up to 42%) and muscle mass in mice without affecting erythrocytes, showing the potential for simultaneous muscle and bone growth -2.

[0036] NACC1 (nucleus accumbens-associated protein 1): uniprot ID: Q96RE7, is crucial for maintaining normal skeletal morphology and development. Animal experiments have shown that NACC1 deficiency leads to spinal morphological defects in mice (such as lumbar sacralization). The mechanism may be related to NACC1's regulation of chondrocyte migration and the expression of cartilage matrix proteins (such as matrixin-3).

[0037] NRIP1 (Nuclear Receptor Interacting Protein 1): uniprot ID: P48552, a key promoter of osteoclast differentiation. Human gene association studies have confirmed its association with osteoporosis. In monocytes / macrophages, NRIP1 regulates osteoclast differentiation, thereby affecting bone homeostasis. OSBPL11 (Oxysterol-Binding Protein-Like 11): uniprot ID: Q9BXB4, encoded by the OSBPL11 gene located on chromosome 3 (3q21.2), belongs to the oxysterol-binding protein (OSBP) family, a group of intracellular lipid receptors.

[0038] COG8 (Oligopoisole Golgi Golgi Complex Subunit 8): uniprot ID: Q96MW5, encodes the 8th subunit-2 of the COG complex, a Golgi transport complex. The COG complex is a peripheral membrane protein complex located on the cis-side of the Golgi apparatus and participates in vesicle retrograde transport and glycosylation within the Golgi apparatus.

[0039] CORO1A (uniprot ID: P31146) is a negative regulator of osteoclast function. It negatively regulates osteoclast bone resorption activity by inhibiting the secretion of lysosomal enzymes such as cathepsin K. Therefore, enhancing CORO1A function may be an innovative strategy to inhibit bone loss.

[0040] CREBBP (CBP protein): uniprot ID: Q92793 is a histone acetyltransferase and transcriptional coactivator that plays a key role in linking amino acid metabolism with hepatic gluconeogenesis.

[0041] ELMO1 (phagocytic and cellular motility protein 1): uniprot ID: Q92556, a core signaling hub in the bone resorption process. ELMO1 deficiency significantly improves bone loss in various osteoporosis and arthritis models. It promotes bone resorption by regulating osteoclast function, influencing its "sealed zone" formation, and protease distribution.

[0042] EphA5 (Ephrin receptor A5): uniprot ID: P54756, is an inhibitor of osteogenic differentiation. Studies have found that EphA5 is a downstream target of miR-499. When an elevated level of a long non-coding RNA called ZFAS1 is reached, it adsorbs miR-499, leading to upregulation of EphA5 expression. This, in turn, inhibits osteogenic differentiation of bone marrow mesenchymal stem cells, promotes adipogenic differentiation, and ultimately results in bone loss.

[0043] ITGA7 (integrin α7), uniprot ID: Q13683, is a receptor protein on the cell surface. Its main function is to help cells attach firmly to the basement membrane. It binds to an extracellular matrix protein called laminin and is highly expressed in skeletal muscle and cardiac muscle. It is essential for maintaining the structural integrity and normal contractile function of muscle tissue.

[0044] PDIA2 (protein disulfide isomerase A2): uniprot ID: D3Z6P0, a member of the PDI family, catalyzes protein folding and disulfide bond formation in the endoplasmic reticulum. It possesses thiol isomerase, oxidase, and reductase activities; its key function is estradiol binding activity, regulating intracellular estradiol levels.

[0045] SIX2 (uniprot ID: Q9NPC8) belongs to the SIX homebox transcription factor family. It is a key regulator of embryonic development, playing a central role, especially in kidney development. Its predicted interacting proteins include EYA1, WT1, PAX2, etc.

[0046] TSKU (Tsukushi, small leucine-rich proteoglycan): uniprot ID: Q8WUA8, belongs to the small leucine-rich proteoglycan (SLRP) family, whose members typically function in the extracellular matrix. It is predicted to bind to TGF-β (transforming growth factor-β), participating in cholesterol efflux, cholesterol homeostasis, and nervous system development; and negatively regulate the WNT signaling pathway-5.

[0047] The kit described in this invention has the following advantages: This invention is the first to discover that a combination of six proteins—FST, NRIP1, OSBPL11, Cystatin SN, CEBPD, and NACC1—can serve as a plasma biomarker for osteoporosis. This invention combines protein chip detection with a machine learning model, using a random forest model to output the probability of disease prevalence. The detection threshold is clearly defined (0.628), and the method is highly operable. The kit of this invention enables high-throughput detection, allowing for the simultaneous detection of multiple biomarkers in a single experiment, with a small sample volume, making it suitable for large-scale screening. Clinical sample validation shows that this kit has a sensitivity of over 75.58% and a specificity of over 76.67% for osteoporosis.

[0048] Through two rounds of protein chip screening, we obtained 15 biomarkers. Based on this, we conducted bioinformatics analysis and statistical analysis, and finally prepared 6 of these biomarkers into a detection kit. Attached Figure Description

[0049] Figure 1 Showing two comparisons of protein chip results. Detailed Implementation

[0050] Example 1: Validation and Screening of Biomarkers First, differentially expressed proteins were initially screened using plasma samples from 26 patients in the osteoporosis group and 14 healthy individuals (35 women and 4 men, mean age 68 years). Specifically, the basic statistical method used for significance analysis was modified t-statistic. The analysis results included the log2 fold change, p-value, and corrected p-value for each protein in the inter-group comparisons. The definition criteria for differentially expressed proteins were: a corrected p-value less than 0.05 and a fold change greater than 1.2 or less than 0.83 (i.e., an absolute logFC value greater than 0.263). A total of 1340 differentially expressed proteins met these criteria, mainly enriched in biological processes such as extracellular matrix, cell adhesion, osteogenic differentiation, and osteoclast regulation, as well as classic bone metabolism-related signaling pathways such as osteoclast differentiation, PI3K-Akt, MAPK, and TNF. These results indicate that the development of osteoporosis is a complex regulatory network involving multiple pathways and stages. Twenty-six differentially expressed proteins were screened and further validated to serve as biomarkers for the detection of primary osteoporosis.

[0051] The second round of screening involved 400 patients in the osteoporosis group and 100 healthy patients (159 males, 341 females, mean age 67.76 years). A custom-designed microarray kit was used to further screen the 26 proteins identified in the first round. The experimental results are shown in Tables 1 and 2 below. Figure 1 As shown: Table 1. 26 protein biomarkers selected in the first screening

[0052] Note: In Table 1, NS indicates no significant difference in plasma protein markers between the osteoporosis group and the healthy group; Down indicates that plasma protein markers were significantly downregulated in the osteoporosis group compared to the healthy group; Up indicates that plasma protein markers were significantly upregulated in the osteoporosis group compared to the healthy group.

[0053] Table 2. Plasma differential protein levels between the osteoporosis group and healthy controls in the second screening.

[0054] Table 2 shows that, with a P-value < 0.05 as the limiting condition, a total of 8 differentially expressed plasma proteins were found between the osteoporosis group and healthy controls. Among them, 4 proteins had FC > 1.2 or < 0.8 and the trend was consistent with the screening results, 4 proteins had FC between 0.8 and 1.2, and the trends of NDUFB1 and LY6E were opposite to the first time.

[0055] Figure 1The results show two comparisons of protein chip screening experiments. Yellow indicates a consistent trend (15 / 26); pink indicates a significant difference regardless of trend (8 / 26); yellow + pink indicates a consistent trend with a significant difference (6 / 26). The detection threshold was subsequently calculated using these six protein markers (i.e., FST, NRIP1, OSBPL11, Cystatin SN, CEBPD, and NACC1).

[0056] Example 2: Machine Learning Model Construction and Optimization 3.1 Data Preparation and Preprocessing Six proteins (FST, NRIP1, OSBPL11, Cystatin SN, CEBPD, and NACC1) obtained from the second round of screening were used as detection biomarkers. The concentration data of these six protein biomarkers detected in 500 samples from the second round were used as the initial parameters for the machine learning model. Before calculating the concentration of protein biomarkers in different samples, two identical positive control points on each chip were used as a reference for data standardization. The signal values ​​of the two positive controls differed by approximately four times. Before standardization, the positive control value POS for each chip was calculated as POS1 + 4. POS2) / 2. Then use this value to standardize all the data: Correction value = Original value (Sample average POS) / POS. The positive control is a specific antibody with a corresponding concentration of biotinylated IgG antibody. The concentration of biotinylated IgG antibody is consistent in each reaction to facilitate standardized detection. 3.1.1 Table 3 Core Dependency Package Description

[0057] 3.1.2 Code Workflow The core steps and overall logic are as follows: Data preparation → Data preprocessing → Imbalanced data balancing → Multi-model training → Threshold optimization → Model evaluation → Optimal model selection → Result output 3.1.3 Detailed Analysis i. Stratified sampling: Stratify according to the proportion of Group labels, dividing the dataset into 70% training set and 30% test set. ii. Core Algorithm: Balance the training set using the ROSE algorithm (an oversampling method based on synthetic samples). iii. Train four classic classification models based on the balanced training set to adapt to different data features: Logistic Regression (LR): A linear classification model that outputs the probability of disease occurrence and is highly interpretable. Specifically, logistic regression is a generalized linear classification model that uses the sigmoid function to map linear combinations to probability values ​​between 0 and 1. It is the most commonly used benchmark model for clinical prediction.

[0058] Prediction formula: z=β0+β1 x1+β2 x2+...+βn xn

[0059] in: P(Y=1): The probability that a sample has osteoporosis. z: The linearly weighted score calculated for all 350 training samples; directly substitute into the Sigmoid function to calculate the probability. β0: Intercept, -0.5510329 β1...βn: Feature weight coefficients, CystatinSN: 9.574946e-06, FST: 0.0001685209, OSBPL11: 2.507354e-05, CEBPD: -3.911261e-05, NACC1: 7.017791e-06, NRIP1: -6.893954e-05 x1...xn: Biomarker characteristics (CystatinSN, FST, OSBPL11, CEBPD, NACC1, NRIP1) Judgment rules: Threshold → Determined as osteoporosis (1); Threshold → Determined as healthy (0) Random Forest (RF): An ensemble learning model that is resistant to overfitting and can output feature importance. Specifically, the Random Forest model constructs multiple decision trees and outputs the final classification result through voting.

[0060] Core formula (classification probability):

[0061] in: T Number of decision trees (set to 500 in this model) I t ( x ): Whether the t-th tree classifies sample x as osteoporosis (yes = 1, no = 0) P ( Y =1): Probability of disease after multiple trees vote Features: It has strong nonlinear fitting ability; It can output the importance of features; It is resistant to overfitting and has high stability.

[0062] Support Vector Machine (SVM): Employs a linear kernel function to output probabilities, suitable for high-dimensional biological data. Specifically, SVM finds the optimal class hyperplane to maximize the class margin; this model uses a linear kernel function.

[0063] Hyperplane formula:

[0064] Classification decision function:

[0065] Probability output (Platt scaling):

[0066] in: ω: Hyperplane normal vector b: Bias term A, B: Probability calibration parameters P ( Y =1): Disease probability output by SVM K-Nearest Neighbors (KNN): Data needs to be standardized (centered and scaled) before classification based on distance. Specifically, KNN classifies data based on a distance metric; this model uses Euclidean distance with K=5.

[0067] Distance formula:

[0068] Classification rules:

[0069] Probability output:

[0070] in: N K ( x ): The set of 5 training samples with the smallest distance, obtained by sorting by distance in learn$x. y i learn$y is the label of its 5 nearest neighbors. I ( y i = c ): Indicator function, counts as 1 if the nearest neighbor label equals category c, otherwise counts as 0. argmax: Counts the number of "1" and "0" cells, and selects the class with the larger number of cells as the final predicted classification. iv. Evaluation of core indicators: Accuracy: The percentage of correct predictions overall. AUC: Area under the ROC curve, measures the overall discrimination ability of the model. Sensitivity: The proportion of real patients correctly predicted (a core clinical indicator). Specificity: The proportion of real healthy people who are correctly predicted. v. Optimal threshold filtering The `find_best_threshold` function filters for the optimal threshold based on the Youden Index: The Youden index = sensitivity + specificity - 1. Iterates through the thresholds in the range of 0.5 to 0.8 with a step size of 0.001 to calculate the Youden index at each threshold. The maximum value occurs when the threshold is 0.628, at which point the sensitivity is 0.7583, the specificity is 0.7667, and the Youden index is 0.525.

[0071] vi. Batch evaluation of models The algorithm iterates through the four models, automatically calculates the optimal threshold, re-evaluates the model using the optimal threshold, outputs the optimized performance metrics, merges all model results, and generates a standardized performance comparison table.

[0072] vii. Optimal Model Selection: Selection Rule: Select the optimal model based on the largest AUC value.

[0073] 3.2 Table 4 Comparison of four models (Logistic Regression, Random Forest, SVM, KNN)

[0074] 3.3 Model Performance Evaluation: Through comparison and validation with four machine learning models—logistic regression, random forest, support vector machine (SVM), and K-nearest neighbors—the random forest model demonstrated the most balanced overall performance, achieving an accuracy of 0.76, sensitivity of 0.7583, and specificity of 0.7667. All indicators remained above 0.75, with no significant weaknesses. It performed best, particularly in balancing disease detection rate (sensitivity) and correct identification rate in healthy individuals (specificity), demonstrating the highest clinical practical value. Therefore, this application preferentially selects random forest as the core model for osteoporosis prediction.

[0075] 3.4 Determination of Optimal Model and Threshold: Based on the principle of maximizing the Youden exponent of the ROC curve, the optimal classification threshold for the random forest model was determined to be 0.628. Using a probability of 0.628 as the cutoff value: when the model outputs a disease probability ≥ 0.628, the sample is classified as high-risk / having osteoporosis; when the probability < 0.628, the sample is classified as healthy. This threshold achieves an optimal balance in distinguishing between diseased and healthy samples, maintaining high sensitivity (0.7583) while obtaining good specificity (0.7667), making it suitable for auxiliary screening scenarios for osteoporosis.

[0076] Example 3: Preparation of the reagent kit of the present invention I. Components of the Reagent Kit 1. Solid-phase support: The reaction chamber is formed by a chip, a soft silicone pad, a rigid frame, and a U-shaped frame clamp. The rigid frame is divided into 2x8 or 4x16 wells, forming a 16- or 64-well frame. The size of the soft silicone pad corresponds to the rigid frame and the standard glass slide. Each small compartment of the rigid frame forms a small reaction well, and the chip in each small reaction well is used to couple a specific capture antibody against the protein marker (see Table 5).

[0077] 2. Washing solution: 20X concentrated washing solution containing Tween 20. 1X washing solution is pH 7.2, containing 0.1% Tween 20 and 0.1 mol / L phosphate buffer.

[0078] 3. Sample diluent: 1 bottle of 15ml 5X concentrated diluent B for diluting samples, protein marker-specific monoclonal antibodies and Cy3-streptavidin.

[0079] The 1X sample dilution B is a 15 mM, pH 7.4 PBS buffer. The solutes and their mass, molar or volume concentrations in the dilution B are as follows: 0.5% casein, 2-4% sucrose, 150 mM NaCl.

[0080] 4. Detection antibodies: Biotinylated detection antibody mixture (monoclonal antibodies against 6 protein markers, each antibody concentration of 0.1-50 ng / ml), the source of each monoclonal antibody is shown in Table 5 below.

[0081] Table 5. Source information of capture and detection antibodies for the six protein biomarkers.

[0082] 5. 200µl of 300X concentrated Cy3-streptavidin solution (3 mg / ml).

[0083] 6. Sample processing solution: 10 ml of 2X cell lysis buffer (RIPA buffer: 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% NP-40 or Triton X-100, 0.5% - 1% sodium deoxycholate and 0.1% SDS).

[0084] 7. Standards: Includes a mixed dry powder of standards for 6 proteins.

[0085] To detect the presence of corresponding bone metabolism-related cytokines in samples, kits were prepared containing immobilized antibodies against specific proteins including Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1.

[0086] Example 4: Experiment on the quantitative detection of osteoporosis biomarkers using the kit of this application. 1. Complete drying of glass slide chips: Take the aldehyde-modified glass chips out of the box, equilibrate at room temperature for 20-30 minutes, open the packaging bag, remove the seal strip, and then place the chips in a vacuum desiccator or dry at room temperature for 1-2 hours.

[0087] 2. Antibody chips were prepared using a fully automated spotting instrument. The specific method is as follows: 1) Monoclonal antibodies that specifically bind to the six markers are mixed with pH 7.4 phosphate buffer (1X antibody dilution (blocking solution), 100 μL) containing 1.4-2% casein to form a mixture of the six antibodies.

[0088] 2) In the antibody mixture, each antibody content of 0.01~2ng is immobilized on the chip in the reaction well, and 2 to 4 replicates are set for each antibody.

[0089] 3. Sample testing Disease samples and control samples were added to a chip immobilized with antibodies corresponding to six protein biomarkers for testing. The specific testing procedure is as follows: (1) Chip operation procedure: Add 50-100µl of sample or standard to each reaction well containing the antibody chip. If it is a tissue sample, the tissue needs to be lysed first, and the obtained tissue lysate should be added at a concentration of 50-500ug / ml after protein concentration determination; if it is plasma or plasma sample, dilute it 2-10 times with sample diluent before adding. Standards should be diluted according to different gradients.

[0090] (2) Cleaning: Remove the sample from each reaction well, wash with washing solution 5 times, shake on a shaker at room temperature for 5 minutes each time, use 150µl of 1× washing solution per well, remove the washing solution completely after each wash, and dilute the washing solution with deionized water to 20×.

[0091] (3) Incubation of biotinylated detection antibody mixture: Centrifuge the detection antibody mixture tube, then add 1.4 ml of sample diluent, mix well, and centrifuge again quickly. Add 80 µl of detection antibody to each well and incubate on a shaker at room temperature for 2 hours.

[0092] (4) Washing: Remove the detection antibody from each well, wash 5 times with 1× washing buffer, shake on a shaker at room temperature for 5 minutes each time, and use 150µl of washing buffer per well. After each wash, remove all the washing buffer.

[0093] (5) Incubation of 1X Cy3-streptavidin: Centrifuge the Cy3-streptavidin tube, then add 1.4 ml of sample diluent, mix well and centrifuge again quickly. Add 80 µl of Cy3-streptavidin to each well, cover the slide with aluminum foil and incubate in the dark on a shaker at room temperature for 1 hour.

[0094] (6) Cleaning: Remove Cy3-streptavidin from each well, wash 5 times with washing solution, shake on a shaker at room temperature for 5 minutes each time, using 150µl of washing solution per well, and remove the washing solution completely after each wash.

[0095] (7) Fluorescence detection Step 1: Remove the chip frame, being careful not to touch the side of the chip with the conjugated antibody.

[0096] Step 2: Place the chip in a chip cleaning tube, add about 30ml of cleaning solution to completely cover the chip, shake on a shaker at room temperature for 15 minutes, and then discard the cleaning solution.

[0097] Step 3: Remove residual cleaning solution from the chip. Place the chip in a chip cleaning / drying tube, without capping, and centrifuge at 1000 rpm for 3 minutes.

[0098] Step 4: Scan the signal using a laser scanner such as Axon GenePix, using Cy3 or the green channel (excitation frequency = 532nm).

[0099] 4. Data extraction from the chip and data analysis using analysis software to obtain the concentration of each protein biomarker. Specifically, Genepix Pro 6.1 software was used to acquire the signal intensity of all points on the chip, calculate the expression level and standard curve of each protein, standardize the signal using positive control signals, and set a background threshold. The average intensity of the control group was added to 2 times the standard deviation (SD) for statistical analysis. Proteins with expression levels higher than the background plus 2xSD were used for corrected T-test analysis. The positive control wells contained biotinylated IgG antibodies of the corresponding specific antibody at the appropriate concentration. The concentration of biotinylated IgG antibodies was consistent in each reaction to facilitate standardized detection. The negative control wells contained biotinylated irrelevant antibodies of the corresponding concentration. The concentration of biotinylated irrelevant antibodies was consistent in each reaction to facilitate standardized detection.

[0100] The concentrations of the six protein biomarkers in the detected samples were input into the random forest model established in Example 2 to obtain the disease probability, and the disease probability was compared with the optimal classification threshold of 0.628 to determine whether the patient had the disease.

[0101] Example 5: Application Validation of the Reagent Kit Independent samples were collected: plasma from 71 patients in the osteoporosis group and 60 healthy patients (93 women and 38 men, with a mean age of 68.6 years).

[0102] The kit of this invention, containing specific antibodies for detecting six protein biomarkers (Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1), was used for detection, and the signal scanning results were obtained (see Table 6 below). The detection results based on the random forest model in Example 2 are as follows: the true positive rate (i.e., sensitivity, true / (true+false)) can reach 77.1%, indicating that the detection model established for the six protein biomarkers in this application is suitable for clinical auxiliary diagnosis (>70%). The kit of this invention has good detection efficacy for osteoporosis, is easy to operate, has high throughput, and is suitable for auxiliary screening of osteoporosis.

[0103] Table 6. Detection results of 131 samples

[0104] Note: FALSE indicates that the prediction result is inconsistent with the actual situation; TRUE indicates that the prediction result is consistent with the actual situation.

Claims

1. Use of a reagent for specifically detecting a protein marker in the manufacture of a kit for detecting primary osteoporosis, characterized in that, The protein biomarkers include: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1.

2. Use of a reagent for specifically detecting a protein marker for the manufacture of a kit for use in a method of detecting primary osteoporosis in a test subject, characterized in that, The protein biomarkers include: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1, and the method includes: (1) The test subject obtains the sample to be tested; (2) Mix the sample to be tested with reagents for the specific detection of protein markers; and (3) Obtain the concentration of each protein marker in the sample to be tested, calculate the probability of disease, and when the probability of disease is greater than or equal to the threshold, it indicates that the sample to be tested comes from an osteoporosis patient or has an osteoporosis risk.

3. A kit for detecting primary osteoporosis, characterized by comprising the antibody of claim 1 or 2. The kit includes reagents for the specific detection of protein biomarkers, including Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1.

4. Use according to claim 1 or 2 or kit according to claim 3, characterized in that, The reagent for specifically detecting protein biomarkers is an antibody that specifically binds to the protein biomarker.

5. Use according to claim 1 or 2 or kit according to claim 3, characterized in that, The protein biomarkers also include COG8, CORO1A, CREBBP, ELMO1, EphA5, ITGA7, PDIA2, SIX2, or TSKU, or any combination thereof.

6. Use according to claim 1 or 2 or kit according to claim 3, characterized in that, The protein biomarkers include Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, NACC1, COG8, CORO1A, CREBBP, ELMO1, EphA5, ITGA7, PDIA2, SIX2, and TSKU.

7. Use according to claim 4, characterized in that, In step (1), the sample to be tested is plasma, serum, urine, somatic cell culture or tissue; And / or, in step (2), the capture antibody for the specific detection protein marker is immobilized onto the chip to prepare an antibody chip; And / or, in step (3), a biotin-conjugated detection antibody and a fluorescein-conjugated streptavidin are used to detect the protein markers conjugated on the antibody chip.

8. Use according to claim 7, characterized in that, In step (1), the sample to be tested is plasma; And / or, in step (2), capture antibodies that specifically detect each protein marker are immobilized onto the chip at a concentration of 0.01~2 ng; And / or, in step (3), the fluorescein is selected from Cy3, Cy5 or FITC.

9. Use of protein markers in constructing a prediction model for primary osteoporosis, characterized in that, The protein biomarkers include: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1.

10. Use of a protein marker for the manufacture of a diagnostic test product for primary osteoporosis, characterized in that, The protein biomarkers include: Cystatin SN, FST, NRIP1, OSBPL11, CEBPD, and NACC1.

Images