Diabetes cognitive impairment prediction marker combination based on blood proteome risk scoring model and application of diabetes cognitive impairment prediction marker combination

Abstract

The invention discloses a diabetes cognitive impairment prediction marker combination based on a blood proteome risk scoring model and application of the diabetes cognitive impairment prediction marker combination, and the marker combination is composed of seven proteins, namely NEFL, WFDC2, PTX3, EXTL1, FGF23, ENO3 and PPME1. The marker combination disclosed by the invention is high in specificity for recognizing cognitive impairment of T2D people and excellent in prediction performance. Compared with PET (Polyethylene Terephthalate) inspection, expensive instruments and equipment are not needed; compared with cerebrospinal fluid examination, risks caused by lumbar puncture can be avoided. The method has extremely high reference value for prediction and targeted therapy intervention of T2D cognitive impairment, and is suitable for popularization and application.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a combination of predictive markers for diabetic cognitive impairment based on a blood proteome risk scoring model and its application. Background Art

[0002] Cognitive impairment has become a major public health issue. Type 2 diabetes (T2D) is a significant risk factor for cognitive impairment, nearly doubling the risk. Effective treatments are currently lacking for this high-risk population, prompting early detection and timely intervention. However, the underlying mechanisms of cognitive impairment in patients with T2D remain unclear. Therefore, elucidating the specific mechanisms of cognitive impairment in T2D patients and identifying markers that reflect these specific pathological changes will greatly facilitate early diagnosis and targeted intervention for this high-risk population.

[0003] Previous studies have established an A / T / N biomarker framework based on the characteristic manifestations of cognitive impairment, including markers of amyloid β (Aβ), tau protein (T), and neuronal damage (N). However, the pathological mechanisms of cognitive impairment in T2D are complex, involving multiple pathological changes such as insulin resistance, dyslipidemia, and chronic inflammation. This suggests that this biomarker framework, primarily focused on Alzheimer's disease (AD), cannot fully reflect the pathological changes of diabetes-related cognitive impairment. Indeed, recent studies have shown that the relationship between Aβ and tau protein and cognitive impairment in T2D is controversial. Currently, there is a lack of blood biomarkers that characterize cognitive impairment in T2D. There are also few reports on targeted therapeutic agents for patients with cognitive impairment in T2D.

[0004] Therefore, identifying pathological markers of cognitive impairment that are specific to the T2D population will provide an effective tool for early identification and targeted intervention of people with diabetic cognitive impairment. Summary of the Invention

[0005] The present invention aims to provide a set of blood markers for accurately predicting diabetic cognitive impairment. This set of blood markers has high predictive efficacy for diabetic cognitive impairment, is easy to detect, and is suitable for widespread application.

[0006] This study, based on quantitative plasma proteomics, employed a Cox proportional hazards regression model to identify proteins in blood that are associated with cognitive impairment in patients with T2D. Of the 2,923 plasma proteins detected, 83 were associated with the risk of cognitive impairment in patients with T2D. Through machine learning feature screening, seven proteins were identified as predictive markers for cognitive impairment in T2D: NEFL (neurofilament light chain), WFDC2 (WAPFour-disulfide core domain protein 2), PTX3 (Pentraxin 3), EXTL1 (exostosin like glycosyltransferase 1), FGF23 (fibroblast growth factor 23), ENO3 (enolase 3), and PPME1 (protein phosphatase methylesterase 1). Kaplan–Meier survival curve analysis showed that higher levels of NEFL, WFDC2, FGF23, and PTX3 were significantly (positively) associated with an increased risk of cognitive impairment in T2D patients. Decreased levels of EXTL1 and ENO3 were associated with an increased risk of cognitive impairment in T2D patients (negatively). PPME1 was less strongly associated with the risk of cognitive impairment in T2D. For ease of application, we integrated these seven blood protein markers to construct a comprehensive proteomic risk score (ProRS). This model demonstrated excellent predictive performance for cognitive impairment in T2D, with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.93. Further incorporating age, sex, education level, and cognitive function score into the model slightly improved its predictive performance, reaching an AUC of 0.94. This model demonstrated robustness in predicting cognitive impairment in T2D across diverse populations.

[0007] As one aspect of the present invention, the present invention provides a marker combination for predicting diabetic cognitive impairment based on a blood proteomic risk scoring model, wherein: the marker combination consists of 7 proteins, namely: NEFL, WFDC2, PTX3, EXTL1, FGF23, ENO3 and PPME1.

[0008] As another aspect of the present invention, the present invention also provides the use of the diabetic cognitive impairment prediction marker combination based on the blood proteome risk scoring model in the preparation of a reagent for identifying diabetic cognitive impairment.

[0009] As one preferred embodiment of the application of the present invention, the method for constructing the marker combination comprises the following steps: Step 1: Detect plasma proteins in patients with type 2 diabetes using quantitative proteomics; Step 2: Use Cox proportional hazards regression model to identify proteins associated with the risk of cognitive impairment in type 2 diabetes; Step 3: Train a lightweight gradient boosting model to screen a combination of predictive markers for cognitive impairment in type 2 diabetes; Step 4: Fit the logistic regression model and estimate the area under the curve of the model.

[0010] As one of the preferred embodiments of the present invention, step 3 includes training a lightweight gradient boosting model using the proteins associated with the risk of cognitive impairment in type 2 diabetes obtained in step 2; clarifying the importance of proteins based on the cross-entropy of the model; selecting the most important proteins and adding them to the training of another lightweight gradient boosting model to predict whether diabetic cognitive impairment occurs; gradually adding the remaining proteins with the highest significance to repeatedly train the lightweight gradient boosting model, and recording the model's predictive performance for cognitive impairment in type 2 diabetes; using Delong's test to test whether the effect of each protein on the model performance is significant; and selecting proteins that can significantly improve the model's predictive performance as a marker combination.

[0011] As one of the preferred embodiments of the present invention, step 2 includes fitting a right-truncated Cox proportional hazard model to each protein obtained in step 1, recording the protein hazard ratio and significance; sorting the proteins according to the P value from small to large; correcting the P value for multiple testing using the Benjamini-Hochberg method; and screening proteins with an FDR < 0.05 as proteins associated with the risk of cognitive impairment in type 2 diabetes.

[0012] As one of the preferred solutions of the present invention, in step 3, the boosting type of the lightweight gradient boosting model adopts a gradient boosting decision tree.

[0013] As one of the preferred solutions of the present invention, in step 3, the maximum number of leaf nodes is 31, the maximum tree depth is not limited, the learning rate is 0.1, the number of trees is 100, and the learning target is binary classification.

[0014] As one of the preferred solutions of the present invention, in step 3, the L2 regularization term is set to 0.1, the classification weight is set to unbalanced, and the importance evaluation index is set to cross entropy.

[0015] The present invention has the following beneficial effects: The marker combination has high specificity and excellent predictive performance for identifying cognitive impairment in people with T2D. Compared with PET scans, it does not require expensive equipment; compared with cerebrospinal fluid (CSF) examinations, it avoids the risks associated with lumbar punctures. This method has high reference value for both the prediction of cognitive impairment in T2D and targeted therapeutic intervention, making it suitable for widespread application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments, wherein: Figure 1 It is a protein associated with the risk of cognitive impairment in T2D patients.

[0017] Figure 2 It is a predictive marker for cognitive impairment in T2D.

[0018] Figure 3 Kaplan–Meier survival curve analysis.

[0019] Figure 4 This is the AUC curve analysis. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with specific embodiments. Example 1:

[0021] Among 2,707 cognitively normal patients with T2D, 350 developed cognitive impairment after a mean follow-up of 13.2 years. The following series of analyses screened this cohort and identified a set of specific predictive markers for cognitive impairment in T2D.

[0022] Step 1: Data preprocessing: Step 1.1: Extract quantitative proteomic data, first diagnosis data, and covariate data (age, sex, and education level) from the cohort. Step 1.2: Identify patients with diabetic cognitive impairment and the duration of their illness based on ICD-10 codes.

[0023] Step 2: Identify risk-associated proteins: Step 2.1: For each protein, fit a right-censored Cox proportional hazards model and record the protein hazard ratio and significance; Step 2.2: Based on the DeLong test, sort the proteins according to the probability of making a type I error (P value) in the AUC significance test from small to large; Step 2.3: P values ​​were corrected for multiple testing using the Benjamini-Hochberg method; Step 2.4: Screen proteins with FDR (False Discovery Rate) < 0.05 as the risk protein set.

[0024] A total of 83 plasma proteins were identified to be associated with the risk of cognitive impairment in T2D patients ( Figure 1 ).

[0025] Step 3: Develop a blood marker panel: Step 3.1: Select all proteins obtained in step 2.4 and train a lightweight gradient boosting model; The lightweight gradient boosting model uses a traditional gradient boosted decision tree, with a maximum number of leaf nodes of 31, an unlimited maximum tree depth, a learning rate of 0.1, a bin size of 200,000 samples, and 100 trees. The learning objective is binary classification, with all samples used for fitting in each training run. The L2 regularization term is set to 0.1, and the classification weights are set to unbalanced. The importance evaluation metric is set to cross entropy.

[0026] Information Gain = entropy (parent) – [average entropy(children)] Entropy = Σ i (-p i log2p i ) Step 3.2: Based on the cross entropy of the model in step 3.1, determine the importance of the protein; Step 3.3: Select the most important proteins and add them to the training of another lightweight gradient boosting model with the same training parameters as step 3.1 to predict whether diabetic cognitive impairment occurs.

[0027] Step 3.4: Gradually add the remaining proteins with the highest significance, repeat step 3.1, and record the model's predictive performance on the patient's risk of cognitive impairment; Step 3.5: Use Delong's test to test whether the addition of each protein has a significant effect on model performance; Step 3.6: Select all proteins that can significantly improve the diagnostic performance of the model after being added to the model to form a blood marker panel.

[0028] The panel includes 7 proteins that are predictive markers of cognitive impairment in T2D, namely NEFL, WFDC2, PTX3, EXTL1, FGF23, ENO3 and PPME1 ( Figure 2). Kaplan–Meier survival curve analysis showed that higher levels of NEFL (HR = 2.48, P = 1.12×10 -12 ), WFDC2 (HR = 2.19, P = 5.96 × 10 -12 ), FGF23(HR = 2.08, P = 2.70 × 10⁻ 7 ), and PTX3 (HR = 2.00, P = 5.47 × 10 -9 ) was significantly associated with an increased risk of cognitive impairment in T2D patients. EXTL1 (HR = 1.64, P = 3.15 × 10 -5 ) and ENO3 (HR = 2.11, P = 7.05 × 10 -8 ) levels were associated with an increased risk of cognitive impairment in T2D patients. PPME1 was weakly associated with the risk of cognitive impairment in T2D (HR = 1.37, P =8.37 × 10⁻²)( Figure 3 ).

[0029] Step 4: Build and evaluate the performance of the comprehensive protein panel assessment model: Step 4.1: Fit a logistic regression model using protein levels; P(y= 1|X)=σ(w T X+b), where X is the protein level matrix and σ is the Sigmoid function; The loss function uses the logarithmic loss function to fit the logistic regression model.

[0030] Step 4.2: Stepwise add covariate data and estimate the area under the curve of the model; The covariate [XC] was added, the model was refitted, and the receiver operating curve method was used to evaluate the model performance.

[0031] The results showed that the model had excellent predictive performance for T2D cognitive impairment, with an area under the curve (AUC) of 0.93. Further incorporating age, gender, education level, and cognitive function score into the model slightly improved the predictive performance, reaching an AUC of 0.94 ( Figure 4 ). The markers of the present invention provide a reliable tool for predicting cognitive dysfunction in T2D population.

[0032] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A combination of predictive markers for diabetic cognitive impairment based on a blood proteome risk scoring model, characterized by: The marker combination consists of 7 proteins, namely: NEFL, WFDC2, PTX3, EXTL1, FGF23, ENO3 and PPME1.

2. Use of the combination of predictive markers for diabetic cognitive impairment based on the blood proteomic risk scoring model according to claim 1 in the preparation of a reagent for identifying diabetic cognitive impairment.

3. The use according to claim 2, characterized in that: The method for constructing the marker combination comprises the following steps: Step 1: Detect plasma proteins in patients with type 2 diabetes using quantitative proteomics; Step 2: Use Cox proportional hazards regression model to identify proteins associated with the risk of cognitive impairment in type 2 diabetes; Step 3: Train a lightweight gradient boosting model to screen a combination of predictive markers for cognitive impairment in type 2 diabetes; Step 4: Fit the logistic regression model and estimate the area under the curve of the model.

4. The use according to claim 3, characterized in that: Step 3 includes training a lightweight gradient boosting model using the risk-associated proteins for cognitive impairment in type 2 diabetes obtained in step 2; clarifying the importance of proteins based on the cross-entropy of the model; selecting the most important proteins and adding them to the training of another lightweight gradient boosting model to predict whether diabetic cognitive impairment occurs; gradually adding the remaining proteins with the highest significance to repeatedly train the lightweight gradient boosting model, and recording the model's predictive performance for cognitive impairment in type 2 diabetes; using Delong's test to test whether the impact of each protein on the model performance is significant; and selecting proteins that can significantly improve the model's predictive performance as a marker combination.

5. The use according to claim 3 or 4, characterized in that: Step 2 included fitting a right-truncated Cox proportional hazards model for each protein obtained in step 1, recording the protein hazard ratio and significance; sorting the proteins according to P value from small to large; correcting the P value for multiple testing using the Benjamini-Hochberg method; and screening proteins with an FDR < 0.05 as proteins associated with the risk of cognitive impairment in type 2 diabetes.

6. The use according to claim 3 or 4, characterized in that: In step 3, the lightweight gradient boosting model uses a gradient boosting decision tree as the boosting type.

7. The use according to claim 3 or 4, characterized in that: In step 3, the maximum number of leaf nodes is 31, the maximum tree depth is unlimited, the learning rate is 0.1, the number of trees is 100, and the learning objective is binary classification.

8. The use according to claim 7, characterized in that: In step 3, the L2 regularization term is set to 0.1, the classification weight is set to unbalanced, and the importance evaluation index is set to cross entropy.

Images