Molecular model for auxiliary diagnosis of diabetic nephropathy established by detecting serum specimens of patients with diabetic nephropathy and non-diabetic nephropathy based on proteomics as well as construction method and application of molecular model

Abstract

The invention relates to a molecular model for auxiliary diagnosis of diabetic nephropathy based on proteomics detection of serum specimens of patients with diabetic nephropathy and non-diabetic nephropathy as well as a construction method and application of the molecular model, which can effectively screen out biomarkers for auxiliary diagnosis model construction and combine with important clinical variables to construct a clinical auxiliary diagnosis model. According to the technical scheme, the molecular model is formed by combining proteins MTTP, TNNI1 and ANGPTL8, the MTTP ratio of diabetic nephropathy patients to non-diabetic kidney disease patients is 8-12.5, the TNNI1 ratio is 7-15, and the ANGPTL8 ratio is 8-14, the protein composition in a sample is detected through a high-throughput mass spectrum means, and the molecular model has the advantages that the molecular model is simple in structure, convenient to operate and high in sensitivity, and the detection accuracy is high. The difference of a diabetic nephropathy patient group and a non-diabetic nephropathy group in the proteomics level is compared, the biomarker for auxiliary diagnosis model construction is screened out, and a new scheme is provided for auxiliary diagnosis of diabetic nephropathy.

Description

Technical Field

[0001] The present invention relates to the field of biomedicine technology, and in particular to a molecular model for establishing auxiliary diagnosis of diabetic nephropathy based on proteomics detection of serum samples of diabetic nephropathy and non-diabetic nephropathy patients, as well as a construction method and application thereof. Background Art

[0002] Diabetic nephropathy (DN) is one of the most common and severe microvascular complications of diabetes and has become the leading cause of chronic kidney disease in my country. Due to the complex pathogenesis of DN, the lack of specific intervention targets, and extremely limited treatment options, the incidence of chronic kidney disease continues to rise.

[0003] Clinical proteomics is a discipline that applies proteomics technology to clinical diagnosis, treatment, prognosis, and etiology research. By combining proteomics with clinical epidemiology, medical statistics, bioinformatics, artificial intelligence, and molecular biology, we explore the molecular effects and mechanisms of disease-related proteins with the goal of preventing disease, improving diagnosis and treatment, accelerating disease recovery, and promoting human health.

[0004] Proteomic diagnostic research provides a basis for disease diagnosis. Based on proteomic technology and research strategies, new disease diagnostic markers can be discovered at the protein expression level, or existing disease diagnostic systems can be optimized. Through the research logic of evidence-based medicine and clinical epidemiology, proteomic technology can be combined with various medical diagnostic technologies such as experiments and imaging to diagnose whether a disease is present and differentiate between different diseases. This can then be compared with the "gold standard" to evaluate the authenticity, reliability, and practicality of diagnostic tests. Therefore, it is imperative to develop a molecular model for auxiliary diagnosis based on proteomic detection of diabetic nephropathy serum specimens. Summary of the Invention

[0005] In view of the above situation, in order to solve the defects of the existing technology, the purpose of the present invention is to provide a method for constructing a molecular model for auxiliary diagnosis of diabetic nephropathy based on proteomics detection of serum samples of diabetic nephropathy patients and non-diabetic kidney disease patients. The method can effectively detect the protein composition in the sample by high-throughput mass spectrometry, compare the differences between the diabetic nephropathy group and the non-diabetic kidney disease group, or the samples of different disease types that need to be identified at the proteomic level, screen out biomarkers for the construction of auxiliary diagnosis models, and combine important clinical variables to construct clinical auxiliary diagnosis models.

[0006] The technical solution provided by the present invention is that the molecular model is composed of a combination of proteins MTTP, TNNI1, and ANGPTL8, wherein the MTTP ratio of diabetic nephropathy patients and non-diabetic kidney disease patients is 8-12.5, the TNNI1 ratio is 7-15, and the ANGPTL8 ratio is 8-14.

[0007] The method for constructing the molecular model comprises the following steps: 1) Preparation of protein-enriched nanomagnetic bead samples Take a diabetic serum sample, dilute it with loading buffer, mix it evenly with the nanomagnetic bead suspension, incubate it at 37°C for 1 hour, wash the magnetic beads twice with loading buffer, and then wash the magnetic beads once with loading buffer without CHAPS. Adsorb the magnetic beads on a magnetic stand and discard the supernatant to obtain a protein-enriched nanomagnetic bead sample. 2) Preparation of peptide samples Lysis buffer was added to the protein-enriched nanomagnetic bead sample, and the sample was incubated at 60°C for 30 min for reductive alkylation. An equal volume of ddH2O was added to dilute the SDC concentration to below 0.5%. Trypsin was added and the sample was incubated at 37°C with shaking overnight for enzymatic digestion. TFA was added the next day to terminate the enzymatic digestion. The supernatant was desalted on an SDB-RPS desalting column, vacuum-dried, and frozen at -20°C to obtain the peptide sample for later use. 3) Mass spectrometry detection of peptide samples The peptide samples prepared in step 2) were detected by mass spectrometry using an UltiMate 3000 RSLCnano nanoliter liquid chromatography (Thermo) coupled to a timsTOF Pro mass spectrometer (Bruker). Samples were injected via an autosampler, bound to a C18 trapping column, and then separated on the analytical column. An analytical gradient was established using mobile phase A with 0.1% formic acid and mobile phase B with 0.1% formicacid in ACN. The flow rate was set to 300 nL / min, and mass spectrometry data were acquired in diaPASEF mode. The diaPASEF acquisition window was set using timsControl software based on the mass-to-charge ratio-ion mobility distribution. The collision energy was set to increase linearly from 59 eV (1 / K0 = 1.6 Vs / cm^2) to 20 eV (1 / K0 = 0.6 Vs / cm^2) based on ion mobility to obtain DIA raw data. 4) Analyze DIA raw data; The DIA raw data obtained in step 3) were analyzed using the library-free mode of DIA-NN (1.8.1) software. The database used for the search was the HUMAN protein sequence database (20220209) downloaded from Uniprot. The search parameters were set to default. The protein quantitative information was normalized using the MaxLFQ algorithm to obtain proteomics data. 5) Proteomics data preprocessing The proteomics data obtained in step 4) were preprocessed. First, contaminating proteins were removed and logarithmic transformation was performed. Then, missing values ​​were estimated using values ​​representing a normal distribution around the detection limit of the mass spectrometer. Specifically, the mean and standard deviation of the actual quantitative values ​​were calculated. A new dataset was created with a mean of the original mean minus 1.8 original standard deviations and a standard deviation of 0.25 original standard deviations. Missing values ​​were filled using this new dataset. Finally, proteins with a missing ratio of more than 50% in both groups were removed. The remaining proteins were selected for subsequent analysis. 6) Statistical analysis For the remaining proteins in step 5), continuous variables that conform to the normal distribution are described using mean and standard deviation. Continuous variables that do not conform to the normal distribution are described using quartiles. Categorical variables are described using percentages. For the test between two groups of data, the t-test is used for data that conform to the normal distribution, and the rank sum test is used for data that do not conform to the normal distribution. For the screening of biomarkers, the difference fold (FC) and statistical test (p ≤ 0.05) are selected as the initial screening basis. LASSO regression and the machine learning method extreme gradient boosting (XGBoost) are used for further biomarker screening, and finally protein markers are screened; 7) Use the final screened protein markers to build a diagnostic model The XGBoost machine learning model was used to screen important proteins from the protein markers screened in step 6), and then the LASSO regression model was used for five-fold cross-validation to further screen differential protein features. Next, the differential proteins were randomly combined and screened through three-fold cross-validation. The optimal combination was selected as the optimal diagnostic molecular model based on the AUC value indicator.

[0008] In step 1), the composition of the loading buffer is: 10mM Tris-Cl, 1mM EDTA, 150mM KCl, 0.05% CHAPS; the composition of the loading buffer without CHAPS is: 10mM Tris-Cl, 1mM EDTA, 150mM KCl.

[0009] In step 2), the lysis buffer composition is: 1% SDC / 100 mM Tris-HCl, pH=8.5 / 10 mM TCEP / 40 mM CAA.

[0010] In step 3), the C18 trapping column parameters are: 75 µm*2 cm, 3 µm particle size, 100 Å pore size, Thermo; the analytical column parameters are: 75 µm*25 cm, 1.6 µm particle size, 100 Å pore size, IonOpticks; the capillary voltage is set to 1400 V, the scan range of the MS1 and MS2 spectra is set to 100-1700 m / z, the ion mobility range is set to 0.6-1.6 Vs / cm2, and the accumulation time and ramp time are set to 100 ms.

[0011] In step 4), the search parameters were as follows: enabling the Precursor ion generation option for theoretical spectral library prediction; using Trypsin / P, allowing up to two missed cleavage sites; setting carbamidomethyl (C) modification as fixed modification; setting oxidaton (M) and acetylation (protein N-terminal) as variable modifications; setting the MS1 and MS2 mass tolerance to 15 ppm; enabling MBR; enabling heuristic protein inference; and setting the FDR to 1%.

[0012] In step 7), the AUC value index is: AUC>=0.90, the model has excellent discrimination ability; 0.90>AUC>=0.75, the model has good discrimination ability; 0.75>AUC>=0.60, the model has certain discrimination ability, but is not recommended; 0.60>AUC, the model has poor discrimination ability.

[0013] The present invention provides an application of a molecular model for auxiliary diagnosis of diabetic nephropathy based on proteomics detection of serum samples from diabetic nephropathy patients and non-diabetic kidney disease patients in the preparation of a kit for detecting diabetic nephropathy.

[0014] The present invention provides a method for establishing an auxiliary diagnostic molecular model based on proteomic detection of serum specimens from patients with diabetic nephropathy. The present invention detects the protein composition in the sample by high-throughput mass spectrometry, compares the differences between the diabetic nephropathy group and the non-diabetic nephropathy group at the proteomic level, screens out biomarkers for the construction of auxiliary diagnostic models, and combines important clinical variables to construct a clinical auxiliary diagnostic model and evaluate the model. The above results show that an auxiliary diagnostic molecular model can be established based on proteomic detection of serum specimens from patients with diabetic nephropathy, providing a new solution for the auxiliary diagnosis of diabetic nephropathy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Workflow constructed for the diagnostic model of the present invention.

[0016] Figure 2 Differentially expressed proteins of the present invention: (A, B) quantitative distribution of (A) proteins and (B) peptides, (C) volcano plot of differentially expressed proteins, (D) proteomic analysis, PCA plot, each point represents a protein, (E) quantitative heat map of differentially expressed proteins.

[0017] Figure 3 shows the proteins related to GO, KEGG and DO enrichment analysis of the present invention: (A) GO-based DEP enrichment analysis (two-tailed hypergeometric test; p≤0.05), GO entries are sorted by p-value, and the top 5 entries in each category are selected for display; (B) KEGG-based DEP enrichment analysis (two-tailed hypergeometric test; p≤0.05), KEGG entries are sorted by p-value, and the top 15 entries are selected for display; (C) DO-based DEP enrichment analysis (two-tailed hypergeometric test; p≤0.05), DO entries are sorted by p-value, and the top 15 entries are selected for display.

[0018] Figure 4 shows the PPI interaction network analysis of the present invention: (A) protein interaction network of related proteins, where dots represent proteins, lines between dots represent protein interaction relationships, and line width represents interaction strength, PPI network diagrams of core proteins obtained by four algorithms, (B) DEGREE, (C) DMNC, (D) MCC, (E) MNC, and (F) Venn diagrams of core proteins.

[0019] Figure 5 The model of the present invention was constructed as follows: (A) protein importance map, predicted relative importance of proteins, (B, C) LASSO regression analysis, and (D) logistic regression model.

[0020] Figure 6 Evaluation of the model of the present invention: (A) receiver operating characteristic (ROC) curve, (B) confusion matrix, (C) calibration curve, (D) DCA curve, and (E) nomogram. DETAILED DESCRIPTION

[0021] The specific embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Example

[0022] The present invention, when implemented, specifically comprises the following steps: 1) Preparation of protein-enriched nanomagnetic bead samples Take 20 μL of diabetic serum sample, dilute it with loading buffer (10 mM Tris-Cl, 1 mM EDTA, 150 mM KCl, 0.05% CHAPS), mix it evenly with 1 mg of nanomagnetic bead suspension, incubate it at 37 ° C for 1 hour, wash the magnetic beads twice with loading buffer, and then wash the magnetic beads once with loading buffer without CHAPS (10 mM Tris-Cl, 1 mM EDTA, 150 mM KCl). Adsorb the magnetic beads on a magnetic stand and discard the supernatant to obtain the protein-enriched nanomagnetic bead sample; 2) Preparation of peptide samples Lysis buffer (1% SDC / 100 mM Tris-HCl, pH=8.5 / 10 mM TCEP / 40 mM CAA) was added to the protein-enriched nanomagnetic bead sample and incubated at 60°C for 30 min for reductive alkylation reaction. An equal volume of ddH2O was added to dilute the SDC concentration to below 0.5%. 1 μg of trypsin was added and incubated at 37°C with shaking overnight for enzymatic digestion. TFA was added the next day to terminate the enzymatic digestion. The supernatant was desalted on an SDB-RPS desalting column, vacuum-evacuated, and frozen at -20°C to obtain the peptide sample for later use. 3) Mass spectrometry detection of peptide samples The peptide samples prepared in step 2) were detected by mass spectrometry using an UltiMate 3000 RSLCnano nanoliter liquid chromatography (Thermo) coupled to a timsTOF Pro mass spectrometer (Bruker). Samples were injected via an autosampler and bound to a C18 trapping column (75 µm*2 cm, 3 µm particle size, 100 Å pore size, Thermo). Separation followed an analytical column (75 µm*25 cm, 1.6 µm particle size, 100 Å pore size, IonOpticks). An analytical gradient was established using mobile phase A (0.1% formic acid) and mobile phase B (0.1% formic acid in ACN). The flow rate was set at 300 nL / min, and mass spectrometry data were acquired in diaPASEF mode. The capillary voltage was set to 1400 V, the scan range of MS1 and MS2 spectra was set to 100–1700 m / z, the ion mobility range was set to 0.6–1.6 Vs / cm2, and the accumulation time and ramp time were set to 100 ms.

[0023] The diaPASEF acquisition window was set using timsControl software based on the mass-to-charge ratio-ion mobility distribution. The collision energy was linearly increased from 59 eV (1 / K0 = 1.6 Vs / cm²) to 20 eV (1 / K0 = 0.6 Vs / cm²) based on ion mobility, generating DIA raw data. 4) Analyze DIA raw data; The DIA raw data obtained in step 3) were analyzed using the library-free mode of DIA-NN (1.8.1) software. The database used for the search was the HUMAN protein sequence database (20220209) downloaded from Uniprot. The search parameters were mainly set to default settings. The key parameters are described as follows: the Precursor ion generation option was enabled for theoretical spectrum library prediction; Trypsin / P was used, allowing up to two missed cleavage sites; Carbamidomethyl (C) modification was set as fixed modification; Oxidaton (M) and Acetylation (protein N-terminal) were set as variable modifications; MS1 and MS2 mass tolerance was set to 15 ppm; MBR was enabled; Heuristic protein inference was enabled; and FDR was set to 1%. Protein quantitative information was normalized using the MaxLFQ algorithm to obtain proteomics data. 5) Proteomics data preprocessing The proteomics data obtained in step 4) were preprocessed. First, contaminating proteins were removed and logarithmic transformation was performed. Then, missing values ​​were estimated using values ​​representing a normal distribution around the detection limit of the mass spectrometer. In short, the mean and standard deviation of the actual quantitative values ​​were calculated, and a new data set was created with a mean of the original mean minus 1.8 original standard deviations and a standard deviation of 0.25 original standard deviations. Missing values ​​were filled using this new data set. Finally, proteins with missing ratios exceeding 50% in both groups were removed, and the remaining proteins were selected for subsequent analysis. 6) Statistical analysis For the remaining proteins in step 5), continuous variables that conform to the normal distribution are described using mean and standard deviation. Continuous variables that do not conform to the normal distribution are described using quartiles. Categorical variables are described using percentages. For the test between two groups of data, the t-test is used for data that conform to the normal distribution, and the rank sum test is used for data that do not conform to the normal distribution. For the screening of biomarkers, the difference fold (FC) and statistical test (p ≤ 0.05) are selected as the initial screening basis. LASSO regression and the machine learning method extreme gradient boosting (XGBoost) are used for further biomarker screening, and finally protein markers are screened; 7) Use the final screened protein markers to build a diagnostic model The XGBoost machine learning model was used to screen important proteins from the protein markers screened in step 6), and then the LASSO regression model was used for five-fold cross-validation to further screen differential protein features. Next, the differential proteins were randomly combined and screened through three-fold cross-validation. The optimal combination was selected as the optimal diagnostic molecular model based on indicators such as the AUC value.

[0024] This study established a molecular model for auxiliary diagnosis based on proteomics testing of serum samples from patients with diabetic nephropathy and non-diabetic kidney disease diagnosed by renal puncture. The study provided a proteomics-based approach and research strategy, using high-throughput mass spectrometry to detect the protein composition of the samples. The proteomic differences between the diabetic nephropathy group and the non-diabetic kidney disease group were compared, and biomarkers for diagnostic model construction were screened. Furthermore, the model was constructed and evaluated in combination with important clinical variables. The relevant experimental data are as follows: 1. Sample information A total of 24 samples were divided into two groups: those diagnosed with diabetic nephropathy (G1) by renal biopsy pathology, with a sample outcome of Group 1 (13 cases); and those diagnosed with non-diabetic nephropathy (G0), with a sample outcome of Group 0 (11 cases).

[0025] 2. Experimental Methods 2.1 Sample processing Samples were enriched using superparamagnetic iron oxide nanoparticles. A 20 μL sample was diluted with loading buffer (10 mM Tris-Cl, 1 mM EDTA, 150 mM KCl, 0.05% CHAPS) and mixed with 1 mg of nanoparticle suspension. The suspension was incubated at 37°C for 1 hour. The beads were washed twice with loading buffer and once with loading buffer without CHAPS (10 mM Tris-Cl, 1 mM EDTA, 150 mM KCl). The beads were adsorbed on a magnetic stand and the supernatant was discarded to obtain the protein-enriched nanoparticles. Lysis buffer (1% SDC / 100 mM Tris-HCl, pH 8.5 / 10 mM TCEP / 40 mM CAA) was added to the sample and incubated at 60°C for 30 min for reductive alkylation. Add an equal volume of ddH2O to dilute the SDC concentration to below 0.5%. Add 1 μg of trypsin and incubate at 37°C with shaking overnight for enzymatic digestion. The next day, add TFA to terminate the digestion. Remove the supernatant and desalt it using an SDB-RPS desalting column. Vacuum dry and freeze at -20°C until further use.

[0026] 2.2 Mass spectrometry detection --- LC-MS / MS analysis Mass spectrometry was performed on an UltiMate 3000 RSLCnano nanoliter liquid chromatography (Thermo) coupled to a timsTOFPro mass spectrometer (Bruker). Peptide samples were injected via an autosampler, bound to a C18 trapping column (75 µm x 2 cm, 3 µm particle size, 100 Å pore size, Thermo), and then separated on an analytical column (75 µm x 25 cm, 1.6 µm particle size, 100 Å pore size, IonOpticks). The analytical gradient consisted of mobile phase A (0.1% formalin) and mobile phase B (0.1% formalin in ACN). The flow rate was set at 300 nL / min. Mass spectrometry data were acquired in diaPASEF mode. The capillary voltage was set at 1400 V. The scan range for MS1 and MS2 spectra was set from 100 to 1700 m / z. The ion mobility range was set to 0.6–1.6 Vs / cm2. Accumulation time and ramp time were set to 100 ms.

[0027] The diaPASEF acquisition window was set using timsControl software based on the mass-to-charge ratio-ion mobility distribution. The collision energy was linearly increased from 59 eV (1 / K0 = 1.6 Vs / cm²) to 20 eV (1 / K0 = 0.6 Vs / cm²) based on ion mobility.

[0028] 2.3 Search Library DIA raw data were analyzed using DIA-NN (1.8.1) software in library-free mode. The search database used was the HUMAN protein sequence database (20220209) downloaded from Uniprot. The search parameters were primarily default, with key parameters described as follows: Precursor ion generation was enabled for theoretical spectral library prediction; Trypsin / P was used, with a maximum of two missed cleavage sites allowed; Carbamidomethyl (C) modification was set as fixed; Oxidaton (M) and Acetylation (protein N-terminal) were set as variable; MS1 and MS2 mass tolerances were set to 15 ppm; MBR was enabled; Heuristic protein inference was enabled; and the FDR was set to 1%. Protein quantification information was normalized using the MaxLFQ algorithm.

[0029] 2.4 Data Processing For proteomics data processing, contaminating proteins were first removed and logarithmic transformation was performed. Then, missing values ​​were imputed using values ​​representing a normal distribution around the mass spectrometer detection limit. Briefly, the mean and standard deviation of the actual quantitative values ​​were calculated, and a new dataset with a mean equal to the original mean minus 1.8 original standard deviations and a standard deviation equal to 0.25 original standard deviations was created. Missing values ​​were then imputed using this new dataset. Finally, proteins with a missing value ratio exceeding 50% in both groups were removed, and the remaining proteins were selected for subsequent analysis.

[0030] 2.5 Statistical analysis Descriptive statistical analysis used mean and standard deviation for continuous variables that followed a normal distribution, quartiles for continuous variables that did not follow a normal distribution, and percentages for categorical variables. For comparisons between two groups, the t-test was used for data that followed a normal distribution, and the rank-sum test for data that did not follow a normal distribution. For biomarker screening, the fold difference (FC) and statistical tests (p ≤ 0.05) were used as initial criteria. LASSO regression and the machine learning method Extreme Gradient Boosting (XGBoost) were used for further biomarker screening. Finally, the selected biomarkers were modeled using appropriate modeling methods based on their specific circumstances. For modeling data processing, the data was split into 70% for training and 30% for testing. Model evaluation was conducted using three key aspects: discrimination using the c-index, receiver operating characteristic (ROC) area (AUC), or c-statistics; calibration and consistency using calibration curves; and clinical efficacy using DCA clinical decision curve analysis or nomograms. Statistical analysis and model building methods were performed using R or Python.

[0031] 3. Data Results 3.1 Patients, Data Collection, and Study Design This study aims to construct a clinical model to distinguish patients with or without disease or to perform differential diagnosis and auxiliary diagnosis. The study population was participants from the First Affiliated Hospital of Zhengzhou University: (1) the training cohort consisted of 17 samples; (2) the validation cohort consisted of 7 samples. We used all samples for bioinformatics analysis and used a machine learning-based biomarker candidate panel to diagnose diabetic nephropathy and evaluate the diagnostic efficacy of the biomarkers. The validation cohort was used to evaluate the performance of the model in distinguishing different populations.

[0032] 3.2 Mass spectrometry data Mass spectrometry data were acquired using a timsTOF Pro LC / MS system. Serum / plasma proteome profiles were acquired for all participants (n=24) using a data-independent acquisition (DIA) strategy. Furthermore, we analyzed the proteomes of 24 participants from the entire cohort using the DIA approach, quantifying a total of 2,081 proteins and 21,897 peptides ( Figure 2 A, 2B). The raw proteomics data were analyzed using DIA-NN software (1.8.1), and the final results were filtered at the precursor ion and protein levels with an FDR of 1%.

[0033] 3.3 Differential protein analysis Differential abundance analysis of the entire cohort revealed that 41 proteins were significantly dysregulated in the diabetic nephropathy G1 group (|log2FC| ≥ 0.26, p ≤ 0.05). Across the entire population, 15 proteins were upregulated and 26 proteins were downregulated in diabetic nephropathy compared with the non-diabetic nephropathy group (Figure 2C). See Results. / output / 4.fold_change / summary.compare.xlsx for details.

[0034] The principal component analysis (PCA) diagram was drawn based on the intensity calculation of the 41 differentially expressed proteins found. The results showed that the diabetic nephropathy G1 group and the non-diabetic kidney disease G0 group could be clearly distinguished by the differentially expressed proteins (Figure 2D). The quantitative information of the differentially expressed proteins was used to draw a heat map ( Figure 2 E), we found that the expression of differential proteins was significantly different between the two groups.

[0035] 3.4 Functional enrichment analysis To investigate the biological functions of these proteins, we performed functional enrichment analysis on the 41 proteins identified, including Gene Ontology (GO, http: / / www.geneontology.org / ), Kyoto Encyclopedia of Genes and Genomes (KEGG, https: / / www.kegg.jp / kegg / ), pathway, and Disease Ontology (DO, https: / / disease-ontology.org / ). Based on the GO database, we further analyzed the biological process (BP), molecular function (MF), and cellular composition (CC) of these proteins using R software.

[0036] In this study, only GO terms with statistically significant enrichment analysis (hypergeometric test, p ≤ 0.05) were considered ( Figure 3 A), KEGG pathway (hypergeometric test, p≤0.05) ( Figure 3 B) and DO items (hypergeometric test, p ≤ 0.05) ( Figure 3 C).

[0037] Based on the STRING database, we used CytoScape software to map the protein-protein interaction (PPI) network of differentially expressed proteins (Figure 4A). Furthermore, using the "CytoHubba" plugin, we further calculated the core proteins for four algorithms (DEGREE, DMNC, MCC, and MNC) (Figures 4B-E). The top 10 proteins from each algorithm were selected as key proteins for cross-analysis and Venn diagram construction. Ultimately, the gene names corresponding to the core proteins were identified as AGRN, APOC3, TNNT1, TNNI1, BGN, SDC4, THBS4, MYH10, MTTP, and TNNC1 (Figure 4F).

[0038] 3.5 Disease Risk Prediction First, we used the XGBoost machine learning model to screen out 11 important proteins from the 41 differentially expressed proteins in the above differential analysis. Figure 5 A shows the predicted importance of proteins in the training cohort ( Figure 5 A).

[0039] Then, a LASSO regression model was used for five-fold cross-validation to further screen out 10 differential protein features (Figures 5B and 5C). Next, the proteins were randomly combined and screened using three-fold cross-validation. The optimal combination was selected using indicators such as AUC values ​​to construct a logistic regression model. The optimal variable combination was: MTTP, TNNI1, and ANGPTL8. Finally, the optimal variable combination was used as the optimal diagnostic molecular model. Figure 5 D shows the coefficients of the optimal model protein.

[0040] 3.6 Model Evaluation Finally, the model was evaluated. The ROC curve is used to evaluate the discrimination of the model. The area under the ROC curve (AUC) is generally used to evaluate the model prediction performance. The AUC (95% CI) of the training set is 0.944 (0.843-1), and the AUC (95% CI) of the test set is 0.917 (0.686-1). AUC>=0.90, the model has excellent discrimination ability; 0.90>AUC>=0.75, the model has good discrimination ability; 0.75>AUC>=0.60, the model has certain discrimination ability, but it is not recommended; 0.60>AUC, the model has poor discrimination ability ( Figure 6 A).

[0041] The confusion matrix represents the classification of the test set, with the horizontal axis representing the actual category and the vertical axis representing the predicted category. It can show the predictive ability of the model to a certain extent (Figure 6B). The calibration curve is used to evaluate the calibration of the model, that is, the degree of consistency between the probability of the predicted outcome and the actual observed frequency of occurrence. The dotted line is the ideal curve, the red line is the actual curve of the model, and the blue line is the curve after resampling calibration. The closer the calibration curve is to the ideal curve, the better the model effect is ( Figure 6 C); The decision curve is used to evaluate the clinical utility of a model. The net benefit rate reflects the model's clinical value; the larger the net benefit rate, the better the model's effectiveness. A black horizontal line represents a scenario where no one receives treatment, resulting in a net benefit of 0. A gray diagonal line represents a scenario where everyone receives treatment, resulting in a negative net benefit. The red curve represents the net benefit curve for the current model. Within a certain range, the higher the net benefit rate of a model, the greater its clinical utility. Experimental results demonstrate that assessing a patient's condition based on this molecular model can assist physicians in making diagnoses.

[0042] The nomogram visualizes the model results, facilitating risk prediction. The sum of the individual scores (points) for each variable at different values ​​yields a total score (Total Points), which is then converted to a Risk Score, representing the probability of an event occurring (Figure 6E), or the probability of developing diabetic nephropathy.

[0043] 4. Conclusion This invention is the first to detect the protein composition in the sample by high-throughput mass spectrometry, compare the differences in serum markers at the proteomic level between the diabetic nephropathy group and the non-diabetic kidney disease group, screen out biomarkers for the construction of auxiliary diagnosis models, and combine important clinical variables to construct a clinical auxiliary diagnosis model and evaluate the model. The above results show that an auxiliary diagnosis molecular model can be established based on proteomic detection of serum samples of diabetic nephropathy patients, providing a new technical means for the auxiliary diagnosis of diabetic nephropathy.

[0044] It should be pointed out that the above is only a preferred embodiment of the present invention and does not impose any form of limitation on the present invention. Any technician familiar with this profession can make changes or modify the technical content disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention, and all of them fall within the scope of protection of the present invention.

Claims

1. A molecular model for assisting the diagnosis of diabetic nephropathy is established based on proteomics detection of serum samples from patients with diabetic nephropathy and patients without diabetic kidney disease, characterized in that: It is composed of the proteins MTTP, TNNI1, and ANGPTL8, among which the MTTP ratio between patients with diabetic nephropathy and patients without diabetic kidney disease is 8-12.5, the TNNI1 ratio is 7-15, and the ANGPTL8 ratio is 8-14.

2. The method for constructing a molecular model for auxiliary diagnosis of diabetic nephropathy based on proteomics detection of serum specimens of diabetic nephropathy patients and non-diabetic kidney disease patients according to claim 1, characterized in that: The following steps are involved: 1) Preparation of protein-enriched nanomagnetic bead samples Take a diabetic serum sample, dilute it with a loading buffer, mix it evenly with the nanomagnetic bead suspension, incubate it at 37°C for 1 hour, wash the magnetic beads twice with the loading buffer, and then wash the magnetic beads once with a loading buffer without CHAPS, adsorb the magnetic beads on a magnetic stand, and discard the supernatant to obtain a protein-enriched nanomagnetic bead sample; 2) Preparation of peptide samples Add lysis buffer to the protein-enriched nanomagnetic bead sample, incubate at 60 °C for 30 min for reductive alkylation reaction, add an equal volume of ddH2O, dilute the SDC concentration to less than 0.5%, add trypsin, incubate at 37 °C with shaking overnight for enzyme digestion, add TFA to terminate the enzyme digestion the next day, take the supernatant for desalting on SDB-RPS desalting column, vacuum dry and freeze at -20 °C to obtain peptide samples for later use; 3) Mass spectrometry detection of peptide samples The peptide samples prepared in step 2) were detected by mass spectrometry using an UltiMate 3000 RSLCnano nanoliter liquid phase tandem timsTOF Pro mass spectrometer. The samples were injected through an automatic sampler, bound to a C18 trapping column, and then separated in an analytical column. The mobile phase A 0.1% formic acid and the mobile phase B 0.1% formic acid in ACN were used to establish an analytical gradient. The flow rate of the analysis was set to 300 nL / min. The mass spectrometer was used for data acquisition in diaPASEF mode. According to the distribution law of mass-to-charge ratio-ion mobility, the diaPASEF acquisition window was set using timsControl software. The collision energy was set to increase linearly from 59 eV at 1 / K0=1.6 Vs / cm^2 to 20 eV at 1 / K0=0.6 Vs / cm^2 according to the ion mobility to obtain DIA raw data. 4) Analyze DIA raw data; The DIA raw data obtained in step 3) were analyzed using the library-free mode of the DIA-NN (1.8.1) software. The database used for the search was the HUMAN protein sequence database downloaded from Uniprot. The search parameters were set by default. The protein quantitative information was normalized using the MaxLFQ algorithm to obtain the proteomics data. 5) Proteomics data preprocessing The proteomics data obtained in step 4) were preprocessed. First, contaminating proteins were removed and logarithmic transformation was performed. Then, the missing values ​​were estimated using values ​​representing the normal distribution around the detection limit of the mass spectrometer. Specifically, the mean and standard deviation of the actual quantitative values ​​were calculated, and a new data set was created, whose mean was the original mean minus 1.8 original standard deviations and whose standard deviation was 0.25 original standard deviations. The missing values ​​were filled using the new data set. Finally, proteins with missing ratios exceeding 50% in both groups were removed, and the remaining proteins were selected for subsequent analysis. 6) Statistical analysis For the continuous variables in the remaining proteins in step 5) that conform to the normal distribution, the mean and standard deviation are used to describe them. For the continuous variables that do not conform to the normal distribution, the quartiles are used to describe them. For the test between the two groups of data, the t test is used for the data that conform to the normal distribution, and the rank sum test is used for the data that do not conform to the normal distribution. For the screening of biomarkers, the difference multiple and the statistical test p≤0.05 are selected as the basis for the initial screening. LASSO regression and the machine learning method extreme gradient boosting method are used for further screening of biomarkers, and finally the protein markers are screened out; 7) Use the final screened protein markers to build a diagnostic model The XGBoost machine learning model was used to screen out important proteins from the protein markers screened in step 6), and then the LASSO regression model was used for five-fold cross-validation to further screen out differential protein features; then, the differential proteins were randomly combined and screened by three-fold cross-validation, and the optimal combination was selected as the optimal diagnostic molecular model through the AUC value indicator.

3. The method for constructing a molecular model for auxiliary diagnosis based on proteomics detection of diabetic nephropathy serum specimens according to claim 2, characterized in that: In the step 1), the composition of the loading buffer is: 10mM Tris-Cl, 1mMEDTA, 150mM KCl, 0.05% CHAPS; the composition of the loading buffer without CHAPS is: 10mM Tris-Cl, 1mMEDTA, 150mM KCl.

4. The method for constructing a molecular model for auxiliary diagnosis based on proteomics detection of diabetic nephropathy serum specimens according to claim 2, characterized in that: In the step 2), the lysis solution composition is: 1% SDC / 100 mM Tris-HCl, pH=8.5 / 10 mM TCEP / 40 mM CAA.

5. The method for constructing an auxiliary diagnosis molecular model based on proteomics detection of diabetic nephropathy serum specimens according to claim 2, characterized in that: In the step 3), the parameters of the C18 trapping column are: 75 µm*2 cm, 3 µmparticle size, 100 Å pore size, Thermo, and the parameters of the analytical column are: 75 µm*25 cm, 1.6 µmparticle size, 100 Å pore size, IonOpticks; the capillary voltage is set to 1400 V, the scanning range of the MS1 and MS2 spectra is set to 100-1700 m / z, the ion mobility range is set to 0.6-1.6 Vs / cm2, and the accumulation time and ramp time are set to 100 ms.

6. The method for constructing a molecular model for auxiliary diagnosis based on proteomics detection of diabetic nephropathy serum specimens according to claim 2, characterized in that: In step 4), the search parameters are as follows: enable the Precursor ion generation option to predict the theoretical spectral library; use Trypsin / P, and allow up to 2 missed cleavage sites; Carbamidomethyl (C) modification is set as fixed modification; Oxidaton (M) and Acetylation (protein N-terminal) are set as variable modifications; MS1 and MS2 mass tolerances are set to 15 ppm; MBR is enabled; Heuristic protein inference is enabled; and FDR is set to 1%.

7. The method for constructing an auxiliary diagnosis molecular model based on proteomics detection of diabetic nephropathy serum specimens according to claim 2, characterized in that: In the step 7), the AUC value index is: AUC>=0.90, the model has excellent distinguishing ability; 0.90>AUC>=0.75, the model has good distinguishing ability; 0.75>AUC>=0.60, the model has certain distinguishing ability, but it is not recommended; 0.60>AUC, the model has poor distinguishing ability.

8. Use of the molecular model for auxiliary diagnosis of diabetic nephropathy based on proteomics detection of serum specimens of diabetic nephropathy patients and non-diabetic kidney disease patients as described in any one of claims 1 to 7 in the preparation of a kit for detecting diabetic nephropathy.

Images