System for constructing a cerebral amyloid angiopathy recognition model, storage medium, and kit

Abstract

The application belongs to the technical field of biomedicine, and particularly relates to a construction system of a cerebral amyloid angiopathy recognition model, a storage medium and a kit. Based on plasma proteomic data of cerebral amyloid angiopathy patients and normal old controls, six key protein markers closely related to cerebral amyloid angiopathy are screened out. The application also provides a model for recognizing cerebral amyloid angiopathy based on a combination of six plasma protein biomarkers. The model shows excellent performance in multiple independent data sets, can accurately distinguish cerebral amyloid angiopathy patients from normal old controls, and further distinguish cerebral amyloid angiopathy patients with high-risk cerebral hemorrhage risk from those with low-risk cerebral hemorrhage risk. The application provides a new tool for early diagnosis and hemorrhage risk assessment of cerebral amyloid angiopathy.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to a system for constructing a model for identifying cerebral amyloid angiopathy, a storage medium, and a reagent kit. Background Technology

[0002] Cerebral amyloid angiopathy (CAA) is an age-related small vessel disease of the brain, a significant cause of primary non-traumatic intracerebral hemorrhage and vascular cognitive impairment in the elderly. The pathological feature of CAA is primarily the deposition of β-amyloid protein (Aβ) in the walls of small and medium-sized arteries within the brain. Meta-analyses suggest that approximately 60% of lobar hemorrhages are associated with moderate to severe CAA. With the increasing trend of population aging, the incidence of CAA-related intracerebral hemorrhage has significantly increased, and its high recurrence and disability rates further exacerbate the burden on patients' families and society. However, there are currently no effective treatments for CAA, largely due to the lack of early diagnostic methods for the disease.

[0003] Currently, the clinical diagnosis of CAA primarily relies on the Boston Diagnostic Criteria, version 1.5, which assesses patients based on clinical presentations (such as lobar hemorrhage or cognitive decline) and imaging markers (such as lobar hemorrhage, microbleeds, and cortical iron deposits). However, the presence of these hemorrhagic imaging markers often indicates severe damage to small blood vessels in the brain, at which point patients are unlikely to benefit from treatment. To further improve the sensitivity of early diagnosis, the Boston Diagnostic Criteria, version 2.0, added non-hemorrhagic imaging markers (multiple punctate subcortical white matter hyperintensities and perivascular spaces in the semioval region). However, these imaging indicators have low specificity, easily leading to misdiagnosis or overdiagnosis, which in turn negatively impacts treatment decisions.

[0004] Therefore, identifying biomarkers that can aid in the early diagnosis of CAA and predict bleeding risk is of significant clinical importance. Previous studies have shown that Aβ in cerebrospinal fluid... 40 Aβ 42 Cerebrospinal fluid (CSF) levels and total Tau protein may be closely related to the occurrence and progression of cardiac aneurysm (CAA). However, due to the invasive nature of CSF sample collection, which requires lumbar puncture and other procedures, its clinical application is limited, making it difficult to meet the needs of large-scale screening. Compared to CSF, peripheral blood biomarkers have gradually become a research hotspot due to their convenient and non-invasive nature. Recent studies have revealed that serum neurofilament light chain protein, matrix metalloproteinases, and chitinase-3-like protein 1 are significantly abnormally expressed in CAA patients and are associated with recurrent cerebral hemorrhage. However, most of these studies focus on the analysis of single or a few protein biomarkers, making it difficult to comprehensively reveal the complex pathological features of CAA.

[0005] Therefore, although existing technologies have made some progress in the diagnosis of CAA and in the study of bleeding risk prediction, there are still many shortcomings and limitations, including the following:

[0006] 1. Deficiencies in existing imaging diagnosis: Although the Boston diagnostic criteria version 2.0 introduced non-hemorrhagic imaging markers (such as high signal intensity in subcortical white matter and perivascular space in the semioval region), the inclusion of these markers increased the risk of overdiagnosis and limited the accuracy of clinical diagnosis.

[0007] 2. Limitations of invasive procedures: Although cerebrospinal fluid markers have shown high diagnostic sensitivity and specificity, their clinical applicability and widespread use are limited due to the invasive nature of lumbar puncture.

[0008] 3. Lack of systematic protein expression profile analysis: Existing peripheral blood biomarker studies mostly focus on a few proteins, which are difficult to fully reflect the multidimensional pathological characteristics of CAA. Omics studies are needed to fully reveal the changes in the protein profile related to the disease.

[0009] 4. Lack of risk assessment models: Existing biomarker studies have limited ability to dynamically monitor disease progression and stratify and predict the risk of cerebral hemorrhage.

[0010] The aforementioned problems limit the widespread application of existing technologies in CAA disease screening and prognosis prediction. Therefore, there is an urgent need for a more efficient, non-invasive, and comprehensive diagnostic technology to support disease management and personalized treatment. This invention, through systematic analysis based on peripheral plasma proteomics, screens key protein biomarkers closely related to CAA and constructs diagnostic and risk prediction models, effectively addressing the limitations of existing technologies. Summary of the Invention

[0011] The purpose of this invention is to provide a system for constructing a model for identifying cerebral amyloid angiopathy, a storage medium, and a reagent kit.

[0012] To address the above problems, this invention provides a system for constructing a model for identifying cerebral amyloid angiopathy, comprising:

[0013] The classification module is used to acquire subjects and divide them into non-overlapping generation and validation queues;

[0014] The acquisition module is used to collect plasma samples from the generation queue, perform non-targeted proteomics detection on the plasma samples from the generation queue to screen key proteins, and obtain the quantitative value of each protein in each plasma sample from the generation queue.

[0015] The training module is used to screen key protein biomarkers based on the quantitative values ​​of each protein in each plasma sample of the generation cohort using differential protein analysis and unsupervised hierarchical clustering algorithms; and to perform modeling and validation based on the quantitative values ​​of key proteins in each plasma sample of the generation cohort to obtain an optimized cerebral amyloid angiopathy model.

[0016] The validation module is used to collect plasma samples from the validation cohort, perform targeted proteomics detection on the plasma samples to validate key proteins, and obtain quantitative values ​​of key proteins in each plasma sample of the validation cohort. Based on the quantitative values ​​of key proteins in each plasma sample of the validation cohort, the optimized cerebral amyloid angiopathy model is input for further validation to obtain the final cerebral amyloid angiopathy model.

[0017] Furthermore, in the above system, the acquisition module is used for:

[0018] Collect plasma samples from subjects in the production cohort;

[0019] Each plasma sample was processed to remove high-abundance proteins, resulting in a low-abundance protein solution.

[0020] Protein concentration and peptide preparation were performed on the low-abundance protein solution of each plasma sample to obtain the peptide solution of each plasma sample.

[0021] Each peptide solution that generated the cohort was subjected to high-pH reverse-phase fractionation, and the fractions were collected and combined for subsequent mass spectrometry detection.

[0022] Using data-dependent acquisition (DDA) and data-independent acquisition (DIA) modes, mass spectrometry was performed on each peptide solution of the merged components in the generation cohort to obtain quantitative values ​​of proteins and peptides in each plasma sample in the generation cohort.

[0023] Furthermore, in the above system, the mass spectrometry detection includes data-dependent acquisition (DDA) using a timsTOF Pro mass spectrometer to construct a spectral database; and data-independent acquisition (DIA) using an Orbitrap Exploris 480 mass spectrometer, outputting raw mass spectrometry data which is then compared with the spectral database using Spectronaut to output quantitative values ​​for each protein and peptide in each plasma sample from the production queue.

[0024] Furthermore, in the above system, the acquisition module randomly divides the generated queue into a first subject group and a second subject group that do not overlap, in a 7:3 ratio. The first subject group and the second subject group each include non-overlapping patients with cerebral amyloid angiopathy (CAA) and normal elderly controls, respectively. The quantitative values ​​of each protein in each plasma sample of the first subject group are used as the first set, and the quantitative values ​​of each protein in each plasma sample of the second subject group are used as the test set.

[0025] In the first set, biomarkers for six key proteins associated with cerebral amyloid angiopathy were screened using differential protein analysis and unsupervised hierarchical clustering. The screening criteria for differential proteins were FDR-adjusted p-value <0.05 and fold change >1.2 or <0.83.

[0026] The verification module is used for:

[0027] Plasma samples were collected from subjects in the validation cohort. Each plasma sample was processed to remove high-abundance proteins, resulting in a low-abundance protein solution. The protein concentration of the low-abundance protein solution was determined using the BCA method. Based on the determined protein concentration, an equal volume of low-abundance protein was taken from the low-abundance protein solution of each plasma sample. This low-abundance protein solution was then subjected to dithiothreitol reduction, iodoacetamide alkylation, trypsin digestion, and peptide desalting, followed by redissolution with 0.1% formic acid to obtain a peptide solution for each plasma sample. Targeted proteomics analysis was performed on the peptide solution of each plasma sample in the validation cohort. Quantitative validation of six key proteins was conducted using parallel reaction monitoring (PRM) mode, yielding quantitative values ​​for the key proteins in each plasma sample of the validation cohort.

[0028] Furthermore, in the above system, the cerebral amyloid angiopathy diagnostic model is used to distinguish between patients with cerebral amyloid angiopathy and normal elderly controls; in the training of the cerebral amyloid angiopathy diagnostic model, the age, gender and standardized final quantitative values ​​of key proteins of the subjects in the training set are used as input variables, and the subjects in the training set are either patients with cerebral amyloid angiopathy or normal elderly controls as labels.

[0029] The hemorrhage risk prediction model is used to distinguish between high-risk and low-risk cerebral hemorrhage patients among those with cerebral amyloid angiopathy. During the training of the hemorrhage risk prediction model, the age, sex, and standardized final quantitative values ​​of key proteins of all cerebral amyloid angiopathy patients in the first set are used as input variables, and cerebral amyloid angiopathy patients in the first set who are at high or low risk of cerebral hemorrhage are used as labels.

[0030] Furthermore, in the above system, the training module is used to: divide the first set into a non-overlapping training set and a validation set; based on the quantitative values ​​of six key proteins in the training set, validation set, and test set, construct and train a cerebral amyloid angiopathy diagnostic model using the Extreme Gradient Boosting (XGBoost) algorithm to obtain an optimized cerebral amyloid angiopathy diagnostic model; wherein the training set, validation set, and test set each include non-overlapping cerebral amyloid angiopathy patients and normal elderly controls;

[0031] The verification module is used to: calculate the key performance indicators of the optimized cerebral amyloid angiopathy diagnostic model, including: area under the receiver operating curve (AUC), accuracy, sensitivity and specificity. If the key performance indicators of the cerebral amyloid angiopathy diagnostic model meet the requirements, the optimized cerebral amyloid angiopathy diagnostic model is verified based on the quantitative values ​​of 6 key proteins in the verification cohort to obtain the final cerebral amyloid angiopathy diagnostic model.

[0032] The training module is used to: construct and train a hemorrhage risk prediction model based on a first set and a test set using the Extreme Gradient Boosting (XGBoost) algorithm to obtain an optimized hemorrhage risk prediction model; wherein the first set and the test set each include non-overlapping patients with high-risk cerebral hemorrhage and patients with low-risk cerebral hemorrhage.

[0033] The validation module is used to: calculate the key performance indicators of the optimized bleeding risk prediction model, including: area under the receiver operating curve (AUC), accuracy, sensitivity and specificity. If the key performance indicators of the optimized bleeding risk prediction model meet the requirements, the optimized bleeding risk prediction model is validated based on the quantitative values ​​of 6 key proteins in the validation cohort to obtain the final bleeding risk prediction model.

[0034] Furthermore, in the above system, the training module is used for:

[0035] Based on the quantitative values ​​of six key proteins generated from the cohort, an extreme gradient boosting (XGBoost) algorithm was used to construct a diagnostic model for cerebral amyloid angiopathy and a hemorrhage risk prediction model.

[0036] For the diagnostic model of cerebral amyloid angiopathy, the XGBoost algorithm and 10-fold cross-validation were used to construct the diagnostic model in the first set. The first set was randomly divided into a non-overlapping training set and a validation set in a 7:3 ratio. The age, gender, and standardized final quantification of key proteins of the subjects in the training set were used as input variables, and the subjects in the training set who were cerebral amyloid angiopathy patients or normal elderly controls were used as labels. The key performance indicators of the diagnostic model in the validation set were calculated, including the area under the receiver operating characteristic (AUC), accuracy, sensitivity, and specificity, to ensure that the key performance indicators of the validation set met the requirements. The XGBoost algorithm relied on the xgboost package in R software, and the hyperparameters were set as follows: learning rate eta = 0.3, minimum weight of all observations in the subset min_child_weight = 1, maximum tree depth max_depth = 6, maximum number of iterations nrounds = 200, minimum objective function reduction required for further branching at the leaf nodes of the tree gamma = 0, and sampling rate subsample = ... 1. When constructing each tree, the feature sampling rate colsample_bytree = 1, the L1 regularization weight alpha = 0, and the L2 regularization weight lambda = 1;

[0037] Alternatively, for the hemorrhage risk prediction model, the XGBoost algorithm and 10-fold cross-validation are used to construct a hemorrhage risk prediction model for cerebral amyloid angiopathy in the first set. The age, sex, and standardized final quantitative values ​​of key proteins of all cerebral amyloid angiopathy patients in the first set are used as input variables, and the cerebral amyloid angiopathy patients in the first set who are at high risk or low risk of cerebral hemorrhage are used as labels. The XGBoost algorithm relies on the xgboost package in R software, and the hyperparameters are set as follows: learning rate eta = 0.01, minimum weight of all observations in the subset min_child_weight = 1, maximum tree depth max_depth = 6, maximum number of iterations nrounds = 200, minimum objective function reduction required for further branching at the leaf nodes of the tree gamma = 0, sampling rate subsample = 0.8 when constructing each tree, feature sampling rate colsample_bytree = 0.8 when constructing each tree, L1 regularization weight alpha = 0, and L2 regularization weight lambda = 1.

[0038] The key performance indicators of the diagnostic model and the hemorrhage risk prediction model in the first set are calculated. If the key performance indicators of the model in the first set meet the requirements, the quantitative values ​​of 6 key proteins in the test set are input into the cerebral amyloid angiopathy diagnostic model and the hemorrhage risk prediction model. The second key performance indicator of the model is calculated. If the second key performance indicator of the model meets the requirements, the quantitative values ​​of 6 key proteins in the validation queue are input into the cerebral amyloid angiopathy diagnostic model and the hemorrhage risk prediction model. The third key performance indicator of the model is calculated. If the third key performance indicator of the model meets the requirements, the optimized cerebral amyloid angiopathy diagnostic model and the hemorrhage risk prediction model are used as the final cerebral amyloid angiopathy identification model.

[0039] Furthermore, in the above system, the targeted proteomics detection includes:

[0040] The parallel reaction monitoring (PRM) mode was used on a timsTOF Pro2 mass spectrometer to quantitatively validate the selected key proteins in peptide solutions from each plasma sample in the validation cohort.

[0041] Furthermore, in the above system, the biomarkers of the key protein include:

[0042] Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE 4 and PREP.

[0043] According to another aspect of the present invention, a computer-readable storage medium is also provided, having stored thereon computer-executable instructions, wherein when executed by a processor, the computer-executable instructions cause the processor to:

[0044] Acquire subjects and divide them into non-overlapping generation and validation queues;

[0045] Plasma samples were collected from the generation cohort, and non-targeted proteomics detection was performed on the plasma samples from the generation cohort to screen key proteins, and the quantitative value of each protein in each plasma sample of the generation cohort was obtained.

[0046] Based on the quantitative values ​​of each protein in each plasma sample from the generation cohort, key protein biomarkers were screened using differential protein analysis and unsupervised hierarchical clustering algorithms. Based on the quantitative values ​​of key proteins in each plasma sample from the generation cohort, modeling and validation were performed to obtain an optimized cerebral amyloid angiopathy model.

[0047] Plasma samples were collected from the validation cohort, and targeted proteomics detection was performed on the plasma samples to validate key proteins, obtaining quantitative values ​​of key proteins in each plasma sample of the validation cohort. Based on the quantitative values ​​of key proteins in each plasma sample of the validation cohort, the optimized cerebral amyloid angiopathy model was input for further validation to obtain the final cerebral amyloid angiopathy model.

[0048] According to another aspect of the present invention, a detection kit for identifying cerebral amyloid angiopathy is also provided, the detection kit comprising reagents for detecting biomarkers of the following key proteins: Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE 4, and PREP.

[0049] This invention aims to address the shortcomings of existing technologies, such as insufficient imaging diagnostics, limitations of invasive procedures, and a lack of systematic analysis and risk assessment models. It develops a system, storage medium, and reagent kit for constructing a brain amyloid angiopathy (CAA) identification model based on plasma proteomics technology. By constructing a diagnostic and prognostic prediction model containing six key protein biomarkers, it provides technology for non-invasive diagnosis and precision medicine of CAA. The detection method of this invention features high sensitivity, strong specificity, non-invasiveness, and ease of clinical application.

[0050] This invention uses high-throughput proteomics analysis (liquid chromatography-mass spectrometry, LC-MS / MS) combined with unsupervised clustering algorithms to screen six key protein biomarkers closely related to CAA from peripheral plasma samples, including Pro-CTSH (procathepsin S precursor), USP15 (ubiquitin-specific protease 15), ApoA-IV (apolipoprotein A-IV), Fibulin-5 (fibular protein-5), RNASE 4 (ribonuclease 4), and PREP (prolyl endopeptidase).

[0051] This invention also provides a model for CAA identification based on a combination of six peripheral blood protein biomarkers, which exhibits excellent performance and has been validated on multiple datasets. Accurate CAA identification is achieved through machine learning of plasma protein biomarkers. For diagnosing CAA, the model achieves an area under the receiver operating curve (AUC) of 1.000 in the training set (first set), with 100% accuracy, 100% sensitivity, and 100% specificity; an AUC of 0.985 in the validation set (first set), with 93.3% accuracy, 91.7% sensitivity, and 95.2% specificity; an AUC of 0.961 in the test set, with 89.2% accuracy, 97.1% sensitivity, and 80.0% specificity; and an AUC of 0.942 in an independent validation cohort, with 91.7% accuracy, 86.7% sensitivity, and 96.7% specificity. The CAA diagnostic model provided by this invention demonstrates good robustness.

[0052] For predicting the risk of bleeding from CAA, the AUC was 0.977 in the first set, with an accuracy of 96.3%, sensitivity of 95.4%, and specificity of 100%; the AUC was 0.808 in the test set, with an accuracy of 77.1%, sensitivity of 80.8%, and specificity of 66.7%; and the AUC was 0.740 in the independent validation cohort, with an accuracy of 86.2%, sensitivity of 88.0%, and specificity of 75.0%.

[0053] Compared with the prior art, the present invention has the following significant advantages:

[0054] 1. Beneficial technological effects

[0055] 1) Non-invasive diagnostic methods:

[0056] This invention uses peripheral blood as the test sample, avoiding invasive procedures such as lumbar puncture, significantly reducing the physical burden and psychological stress on patients, and improving the clinical applicability of the method.

[0057] 2) High sensitivity and specificity:

[0058] Based on six key protein biomarkers (Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE4, and PREP), the constructed diagnostic and bleeding risk prediction models demonstrated excellent performance across multiple datasets. In an independent validation cohort, the model achieved an AUC of 0.942 for diagnosing CAA, with an accuracy of 91.7%, sensitivity of 86.7%, and specificity of 96.7%, outperforming existing diagnostic methods. For predicting bleeding risk in CAA, the model achieved an AUC of 0.740, with an accuracy of 86.2%, sensitivity of 88.0%, and specificity of 75.0%.

[0059] 3) The scientific basis of differential protein screening:

[0060] Non-targeted proteomics detection and targeted proteomics validation ensured the comprehensiveness and reliability of differential protein screening, laying a solid foundation for model construction.

[0061] 4) The robustness and generalization ability of the model:

[0062] By designing and modeling with machine learning on multiple datasets (using the XGBoost algorithm), the model demonstrates good robustness and generalization ability in both the generation and validation queues, and is generalizable.

[0063] 2. Beneficial economic effects

[0064] 1) Simplify the testing process:

[0065] The standardized plasma testing and targeted proteomics testing process is efficient and convenient, reducing the technical barriers to laboratory and clinical operations.

[0066] 2) Cost reduction:

[0067] Compared with traditional cerebrospinal fluid testing or imaging biomarker analysis, the blood testing method of this invention is lower in cost and more suitable for large-scale screening and early diagnosis.

[0068] 3. Beneficial effects on society

[0069] 1) Promote the development of precision medicine:

[0070] Based on proteomics technology and machine learning methods, this invention provides a technical foundation for personalized diagnosis and management of CAA, and helps to implement the concept of precision medicine.

[0071] 2) Reduce the social medical burden:

[0072] Early detection and intervention in CAA patients can reduce the consumption of medical resources caused by disease progression and alleviate the economic burden on families and society.

[0073] In summary, this invention has significant advantages in terms of technology, economy, and society, and can provide important support for the early and accurate diagnosis and personalized management of cerebral amyloid angiopathy, with broad clinical application prospects and promotional value. Attached Figure Description

[0074] Figure 1A This is a differential expression map of key proteins in the first set of the generation queue according to an embodiment of the present invention;

[0075] Figure 1B This is a differential expression map of key proteins in the test set of the generation queue according to an embodiment of the present invention;

[0076] Figure 1C This is a graph showing differential protein expression in a validation queue according to an embodiment of the present invention;

[0077] Figure 2 This is a schematic diagram of the AUC of a key protein in an embodiment of the present invention, which identifies CAA patients in the training and validation sets of the generated cohort.

[0078] Figure 3 This is a schematic diagram of the confusion matrix of the key protein in an embodiment of the present invention in recognizing CAA patients in the training and validation sets of the generated cohort;

[0079] Figure 4 This is a schematic diagram of the AUC of a key protein in an embodiment of the present invention, which identifies CAA patients in a test set that generates a cohort.

[0080] Figure 5 This is a schematic diagram of the confusion matrix of a key protein in an embodiment of the present invention, which identifies CAA patients in a test set that generates a cohort;

[0081] Figure 6 This is a schematic diagram of the AUC of a key protein in a validation cohort of the present invention for recognizing the performance of CAA patients in one embodiment of the invention;

[0082] Figure 7 This is a schematic diagram of the confusion matrix of a key protein in a validation cohort recognizing the performance of CAA patients according to an embodiment of the present invention;

[0083] Figure 8 This is a schematic diagram of the AUC of a key protein in an embodiment of the present invention, showing how it distinguishes between CAA patients at high and low risk of intracerebral hemorrhage in the first set of the generation cohort.

[0084] Figure 9 This is a schematic diagram of the confusion matrix of a key protein in an embodiment of the present invention, showing how it distinguishes between CAA patients at high and low risk of intracerebral hemorrhage in the first set of the generation cohort.

[0085] Figure 10This is a schematic diagram of the AUC of a key protein in an embodiment of the present invention, used to distinguish between CAA patients at high and low risk of intracerebral hemorrhage in a test set that generates a cohort.

[0086] Figure 11 This is a schematic diagram of the confusion matrix of a key protein in an embodiment of the present invention, showing how it distinguishes between CAA patients at high and low risk of intracerebral hemorrhage in a test set that generates a cohort.

[0087] Figure 12 This is a schematic diagram of the AUC of a key protein in an embodiment of the present invention, used to distinguish between CAA patients at high and low risk of intracerebral hemorrhage in a validation cohort.

[0088] Figure 13 This is a schematic diagram of the confusion matrix of a key protein in an embodiment of the present invention, showing how it distinguishes between CAA patients at high and low risk of intracerebral hemorrhage in a validation cohort. Detailed Implementation

[0089] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0090] With the rapid development of proteomics technology, researchers can comprehensively analyze peripheral blood protein expression profiles using high-throughput analytical methods (such as liquid chromatography-mass spectrometry, LC-MS / MS), providing an important opportunity for the discovery of new biomarkers. Systematic proteomics-based analysis can not only reveal the characteristics of protein changes related to CAA (causing acute exacerbation of angina pectoris) but also screen for combinations of multiple biomarkers to improve diagnostic accuracy. Compared to single biomarkers, combinations of multiple biomarkers can more comprehensively reflect different pathological stages of the disease, thereby achieving higher sensitivity and specificity.

[0091] This invention aims to overcome the shortcomings of existing technologies and develop a system, storage medium, and reagent kit for constructing a model for identifying cerebral amyloid angiopathy (CAA). Through systematic analysis of plasma protein expression profiles in CAA patients, a set of key protein biomarkers was screened and validated using an unsupervised clustering algorithm. These biomarkers exhibit significant diagnostic value. Furthermore, this invention constructed a diagnostic model and a bleeding risk prediction model based on plasma key protein biomarkers, and validated their performance in an independent validation cohort. The development of this invention provides a new solution for the non-invasive diagnosis and prognostic assessment of CAA, and has significant clinical application value.

[0092] like Figure 2 As shown, this invention provides a system for constructing a model for recognizing cerebral amyloid angiopathy, a storage medium, and a reagent kit, comprising:

[0093] Step S1, Design and partitioning of the research dataset:

[0094] The classification module study included 274 prospectively enrolled subjects from two centers, and the subjects were divided into a non-overlapping generation cohort and a validation cohort; the validation cohort was used to evaluate the generalization performance of the model.

[0095] Specifically, this invention was approved by the ethics committees of a first-level hospital and a second-level hospital, and all subjects or their legal representatives signed written informed consent forms. The study strictly followed the Declaration of Helsinki.

[0096] 1. Case inclusion criteria

[0097] 1) Patients diagnosed with "pathologically confirmed" or "probable cerebral amyloid angiopathy (CAA)" according to Boston Criteria Version 1.5;

[0098] 2) Age ≥ 50 years;

[0099] 3) Available plasma samples;

[0100] 4) Exclude patients with other neurological diseases or serious systemic diseases.

[0101] 2. Inclusion criteria for the control group

[0102] 1) A community resident population that matches age and gender;

[0103] 2) No history of stroke or dementia;

[0104] 3) There are available plasma samples.

[0105] 3. Study cohort division

[0106] To construct and validate a diagnostic model based on key protein combinations, the subjects were divided into the following two study cohorts:

[0107] 1) The cohort (the first center, including 116 CAA patients and 118 controls) included CAA patients prospectively recruited from a primary hospital between January 2015 and July 2020, and control groups recruited from the Shanghai Aging Study during the same period.

[0108] 2) Validation cohort (second center, including 30 CAA patients and 30 controls): CAA patients recruited from a second hospital between October 2017 and July 2023, and control groups recruited from a research project during the same period.

[0109] Step S2, collection and testing of plasma samples:

[0110] To screen for differentially expressed proteins associated with cerebral amyloid angiopathy (CAA), plasma samples from a generation cohort were collected using a collection module. Non-targeted proteomics analysis was then performed on these plasma samples to screen for key proteins, obtaining quantitative values ​​for the key proteins in each plasma sample from the generation cohort. The specific steps are as follows:

[0111] Step A1: Sample collection: Peripheral blood samples were collected from the forearms of subjects who generated the cohort. After EDTA anticoagulation, the plasma samples were separated twice by centrifugation at 1000g for 10 minutes at 4°C and stored at -80°C.

[0112] Step A2: Sample processing:

[0113] Step A21: Each separated plasma sample from the generated queue is thawed on ice. After enriching blood low-abundance proteins using the Deep Low Abundance Protein Enrichment and Pretreatment Kit, a low-abundance protein solution is obtained for each plasma sample. The protein concentration of the low-abundance protein solution of each plasma sample is determined using the BCA method to obtain the low-abundance protein solution.

[0114] Step A22: Based on the measured protein concentration, take a certain volume of low-abundance protein solution from the low-abundance protein solution of each plasma sample according to an equal amount of low-abundance protein, and then perform dithiothreitol reduction, iodoacetamide alkylation, trypsin digestion and peptide desalting in sequence, and then redissolve it with 0.1% formic acid to obtain the peptide solution of each plasma sample.

[0115] In step A23, each peptide solution from the generated cohort was fractionated using an Agilent 1260 Infinity II HPLC system at high pH reverse phase, and 36 fractions were collected. After freeze-drying, the fractions were reconstituted with 0.1% formic acid and combined into 6 fractions for subsequent mass spectrometry detection.

[0116] Step A3, LC-MS / MS detection:

[0117] Using data-dependent acquisition (DDA) and data-independent acquisition (DIA) modes, LC-MS / MS was used to detect the six components obtained from each peptide solution of the generated protein, and the quantitative values ​​of each protein and peptide in each plasma sample of the generating cohort were obtained.

[0118] Step A31, DDA mode is used to build the spectral library:

[0119] Step A311, Liquid Chromatography Separation: Take 200 ng of peptide solution from each of the six components of the peptide solution in each plasma sample and separate them using a NanoElute UHPLC system (Bruker, Bremen, Germany). The mobile phase in the NanoElute UHPLC system is solvent A and solvent B; gradient elution is performed from 3% to 100% of solvent B at a flow rate of 300 nL / min to obtain the first peptide solution after separation. Solvent A is a 0.1% formic acid aqueous solution, and solvent B is a 0.1% formic acid acetonitrile solution (acetonitrile is 100%).

[0120] Step A312, Mass Spectrometry Detection: Input the separated first peptide solution into the mass spectrometer. The mass spectrometer is a TimsTOFPro (Bruker, Bremen, Germany). The detection mode is positive ion, the precursor ion scan range is 100-1700 m / z, the ion mobility range is 0.75-1.4 V·s / cm², the capillary voltage is 1500 V, the total cycle time for 10 MS / MS scans is 1.16 s, the charge range is 0-5, the dynamic exclusion time is 0.4 min, the target ion intensity is 20000, and the ion intensity threshold is 2500. The first raw mass spectrometry data is then output.

[0121] Step A313: The first raw mass spectrometry data was analyzed using Spectronaut (Biognosys AG, Switzerland). The DDA file (the first raw mass spectrometry data) was matched with the human UniProt FASTA database. The enzyme used for digestion was Trypsin / P (allowing two missed cleavage sites), the fixed modification was Carbamidomethyl (C), and the variable modifications included Oxidation (M) and Acetyl (Protein N-term). 1% FDR filtration was applied at both the protein and precursor levels to obtain the spectral database and ensure the reliability of the results.

[0122] Step A32, DIA mode is used for protein quantification analysis:

[0123] Step A321, Liquid Phase Separation: Take 1 μg of peptide solution from each of the 6 components of the peptide solution in each plasma sample and separate them using an Easy nLC system (Thermo Fisher Scientific). The mobile phase in the Easy nLC system is solvent A and solvent B; gradient elution is performed from 2% to 100% of solvent B at a flow rate of 300 nL / min to obtain the second peptide solution after separation. Solvent A is a 0.1% formic acid aqueous solution, and solvent B is a 0.1% formic acid acetonitrile solution (acetonitrile is 80%).

[0124] Step A322, Mass Spectrometry Detection: Input the separated second peptide solution into the mass spectrometer. The mass spectrometer is an OrbitrapExploris 480 (Thermo Fisher Scientific). The detection mode is positive ion, the precursor ion scan range is 350-1200 m / z, the resolution is 60,000 for full scan and 30,000 for tMS2 scan, the collision energy is 30%, and the DIA window is set to 40 variable windows to improve detection depth and sensitivity, so as to output the second raw data of mass spectrometry.

[0125] Step A323: The second raw mass spectrometry data and the spectral database are compared using Spectronaut (Biognosys AG, Switzerland) software to obtain quantitative results. Spectronaut software dynamically determines the ideal extraction window based on dynamic iRT calibration and gradient stability, and performs quantitative analysis based on precursors and proteins filtered by FDR to output quantitative values ​​for each protein and peptide in each plasma sample.

[0126] Step A33, Data Cleaning: For the quantitative values ​​of each protein in each plasma sample in the generation queue, remove proteins with more than 50% missing values; imput the missing values ​​of the remaining proteins with less than or equal to 50% missing values ​​using the randomForest package in R software to output the final quantitative value of each protein in each plasma sample in the generation queue.

[0127] Step S3: Screen key plasma proteins using the training module, and perform machine learning modeling and validation.

[0128] Step B1, Differential Protein Analysis and Key Protein Screening:

[0129] Step B11, further dividing the generation cohort: The generation cohort is randomly divided into a first subject group and a second subject group with a ratio of 7:3, which are non-overlapping. Both the first and second subject groups include non-overlapping CAA patients and normal elderly controls. The quantitative values ​​of each protein in each plasma sample of the first subject group are used as the first set, and the quantitative values ​​of each protein in each plasma sample of the second subject group are used as the test set.

[0130] Step B12, Differential Protein Analysis: In the first set of the generated cohort, according to the screening criteria of FDR-adjusted p-value < 0.05 and fold change (FC) > 1.2 or < 0.83, proteins that meet the final quantitative values ​​of each protein in each plasma sample are selected as differentially expressed proteins; wherein, the FDR-adjusted p-value is used as the q-value;

[0131] For example, in one embodiment of the present invention, a total of 952 plasma proteins were detected, and 166 of them were identified as proteins that showed differential expression between CAA patients and controls (differential protein definition: q value < 0.05, and FC > 1.2 or < 0.83).

[0132] Step B13, Key Protein Screening: For the screened differentially expressed proteins, first calculate the Pearson correlation coefficient to generate a correlation matrix; based on the correlation matrix, perform unsupervised hierarchical clustering using the Ward.D2 method, and determine the optimal number of clusters using the Gap statistic, for example, 6 clusters can be obtained; in each cluster, representative proteins are screened based on the minimum q value of the proteins in each cluster, and defined as key proteins.

[0133] For example, in one embodiment of the present invention, key protein screening is performed as follows: For the differentially expressed proteins, the Pearson correlation coefficient is first calculated to generate a correlation matrix; based on the correlation matrix, unsupervised hierarchical clustering is performed using the Ward.D2 method, and the optimal number of clusters is determined to be 6 using the Gap statistic; in each cluster, representative proteins are selected based on the minimum q value and defined as key proteins; finally, the following six proteins are selected: Pro-CTSH (procathepsin S), USP15 (ubiquitin-specific protease 15), ApoA-IV (apolipoprotein A-IV), Fibulin-5 (fibular protein-5), RNASE 4 (ribonuclease 4), and PREP (prolyl endopeptidase); among them, Pro-CTSH, USP15, Fibulin-5, RNASE 4, and PREP are significantly highly expressed in CAA patients, while ApoA-IV is significantly lowly expressed in CAA.

[0134] Here, as Figure 1A The image shows a differential expression map of key proteins in the first set of generated queues; as shown... Figure 1B The image shows a differential expression map of key proteins in the test set that generated the queue; p<0.05, p<0.01, p<0.001;

[0135] Preferred, six key proteins can be finally screened out: Pro-CTSH (procathepsin S precursor), USP15 (ubiquitin-specific protease 15), ApoA-IV (apolipoprotein A-IV), Fibulin-5 (fibulin-5), RNASE 4 (ribonuclease 4), and PREP (prolyl endopeptidase).

[0136] Step B2, Model Building and Training:

[0137] Step B21: Based on the quantitative values ​​of six key proteins generated from the queue, the Extreme Gradient Boosting (XGBoost) algorithm is used to construct a diagnostic model for cerebral amyloid angiopathy and a hemorrhage risk prediction model, respectively.

[0138] Here, based on the generated queue, an optimized cerebral amyloid angiopathy identification model is trained, including: a cerebral amyloid angiopathy diagnostic model and a hemorrhage risk prediction model;

[0139] Specifically, step B21 includes: for the cerebral amyloid angiopathy diagnostic model, using the XGBoost algorithm and 10-fold cross-validation, constructing the cerebral amyloid angiopathy diagnostic model in the first set; randomly dividing the first set into a non-overlapping training set (70% of the first set) and a validation set (30% of the first set) in a 7:3 ratio; the training set and validation set are used for model construction and parameter optimization; using the age, gender, and standardized final quantification of key proteins of the subjects in the training set as input variables, and using cerebral amyloid angiopathy patients or normal elderly controls in the training set as labels; calculating the key performance indicators of the diagnostic model in the validation set, including the area under the receiver operating system (AUC), accuracy, sensitivity, and specificity, so that the key performance indicators of the validation set meet the requirements; wherein, the XGBoost algorithm relies on the xgboost package of R software, and the hyperparameters are set as learning rate eta = 0.3, minimum weight of all observations in the subset min_child_weight = 1, maximum tree depth max_depth = 6, and maximum number of iterations nrounds = 200, the minimum objective function reduction required to further branch at the leaf nodes of the tree, gamma = 0, the sampling rate of samples when building each tree, subsample = 1, the feature sampling rate when building each tree, colsample_bytree = 1, L1 regularization weight alpha = 0, L2 regularization weight lambda = 1;

[0140] For the hemorrhage risk prediction model, the XGBoost algorithm and 10-fold cross-validation were used to construct a hemorrhage risk prediction model for cerebral amyloid angiopathy in the first set. The age, sex, and standardized final quantitative values ​​of key proteins of all cerebral amyloid angiopathy patients in the first set were used as input variables, and cerebral amyloid angiopathy patients with high-risk or low-risk cerebral hemorrhage in the first set were used as labels. The XGBoost algorithm relied on the xgboost package of R software, and the hyperparameters were set as follows: learning rate eta = 0.01, minimum weight of all observations in the subset min_child_weight = 1, maximum tree depth max_depth = 6, maximum number of iterations nrounds = 200, minimum objective function reduction required for further branching at the leaf nodes of the tree gamma = 0, sampling rate subsample = 0.8 when constructing each tree, feature sampling rate colsample_bytree = 0.8 when constructing each tree, L1 regularization weight alpha = 0, and L2 regularization weight lambda = 1.

[0141] Here, the cerebral amyloid angiopathy (CAA) identification model may include a cerebral amyloid angiopathy diagnostic model and / or a cerebral amyloid angiopathy hemorrhage risk prediction model.

[0142] Step B22: Calculate the key performance indicators of the cerebral amyloid angiopathy diagnostic model and the hemorrhage risk prediction model in the first set. If each model meets the requirements for key performance indicators in the first set, input the quantitative values ​​of 6 key proteins in the test set into the cerebral amyloid angiopathy diagnostic model and the hemorrhage risk prediction model respectively, and calculate the second key performance indicator of the model. If each model meets the requirements for the second key performance indicator, input the quantitative values ​​of 6 key proteins in the validation queue into the cerebral amyloid angiopathy diagnostic model and the hemorrhage risk prediction model, and calculate the third key performance indicator of each model. If each model meets the requirements for the third key performance indicator, the optimized cerebral amyloid angiopathy diagnostic model and the hemorrhage risk prediction model are used as the final cerebral amyloid angiopathy identification model.

[0143] Preferably, the key performance indicators include: AUC value, accuracy, sensitivity, specificity, etc.

[0144] For example, such as Figures 2 to 5As shown, in one embodiment of the present invention, the cerebral amyloid angiopathy diagnostic model has an AUC value of 1.000, accuracy of 100%, sensitivity of 100%, and specificity of 100% in the training set; an AUC value of 0.985 in the validation set, accuracy of 93.3%, sensitivity of 91.7%, and specificity of 95.2%; and an AUC value of 0.961 in the test set, accuracy of 89.2%, sensitivity of 97.1%, and specificity of 80.0%.

[0145] like Figures 8 to 11 As shown, in one embodiment of the present invention, the bleeding risk prediction model has an AUC value of 0.977 in the first set, an accuracy of 96.3%, a sensitivity of 95.4%, and a specificity of 100%; and an AUC value of 0.808 in the test set, with an accuracy of 77.1%, a sensitivity of 80.8%, and a specificity of 66.7%.

[0146] Table 1 shows the performance of the CAA diagnostic model built based on age, sex, and six protein biomarkers on the training, validation, and test sets.

[0147] Table 1

[0148]

[0149] Table 2 shows the performance of the CAA bleeding risk prediction model built based on age, sex, and six protein biomarkers in the first set and test set.

[0150] Table 2

[0151]

[0152] Preferably, in step B22, based on the validation queue, the optimized cerebral amyloid angiopathy (CAA) identification model is further validated to obtain the final cerebral amyloid angiopathy (CAA) model, including:

[0153] Step B221: Targeted detection and validation of key plasma proteins in the validation cohort:

[0154] The optimized cerebral amyloid angiopathy (CAA) identification model was applied to the validation cohort. To verify the performance of the optimized CAA identification model, targeted proteomics detection was used to quantify key proteins such as Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE4, and PREP in the validation cohort. The optimized CAA model was then applied to the quantitative detection results to calculate the model's third key performance indicator. If the model's third key performance indicator met the requirements, the optimized CAA model was adopted as the final CAA model.

[0155] The verification process in this step can be implemented through a verification module.

[0156] A better approach is step B221, which can be implemented as follows:

[0157] Step C1, Sample collection: Plasma samples from subjects in the validation cohort were anticoagulated with EDTA, centrifuged twice at 1000g for 10 minutes at 4°C to obtain the separated plasma samples, and stored at -80°C.

[0158] Step C2, Sample Processing:

[0159] Step C21: Thaw the separated plasma sample on ice, enrich the blood low-abundance proteins using the Deep Low Abundance Protein Enrichment and Pretreatment Kit to obtain a low-abundance protein solution, and determine the protein concentration of the low-abundance protein solution using the BCA method.

[0160] Step C22: Based on the measured protein concentration, take a certain volume of low-abundance protein solution from the low-abundance protein solution of each plasma sample according to an equal amount of low-abundance protein, and then perform dithiothreitol reduction, iodoacetamide alkylation, trypsin digestion and peptide desalting in sequence, and then redissolve it with 0.1% formic acid to obtain the peptide solution of each plasma sample.

[0161] Step C3, LC-MS / MS detection:

[0162] Step C31, Liquid Chromatography Separation: Take 200 ng of peptide solution from each plasma sample and separate it using a NanoElute UHPLC system (Bruker, Bremen, Germany). The mobile phase in the NanoElute UHPLC system is solvent A and solvent B. Gradient elution is performed from 2% to 80% of solvent B at a flow rate of 300 nL / min to obtain the second peptide solution after separation. Solvent A is a 0.1% formic acid aqueous solution, and solvent B is a 0.1% formic acid acetonitrile solution (acetonitrile is 100%).

[0163] Step C32, Mass Spectrometry Detection: Input the separated second peptide solution into the mass spectrometer. The mass spectrometer is a TimsTOFPro2 (Bruker, Bremen, Germany). The detection mode is positive ion, the precursor ion scan range is 100-1700 m / z, the ion mobility range is 0.85-1.3 V·s / cm², the capillary voltage is 1500 V, the total cycle time for 4 MS / MS scans is 0.53 s, the charge range is 0-5, the dynamic exclusion time is 0.4 min, the target ion intensity is 10000, and the ion intensity threshold is 1500. The third raw mass spectrometry data is then output.

[0164] Step C33: Based on the third raw data of mass spectrometry, obtain the mass-to-charge ratio and signal intensity of fragment ions after peptide fragmentation as secondary spectra; search and compare the secondary spectra generated by mass spectrometry with the theoretical secondary spectra constructed from preset key protein sequences, and obtain the correctly matched theoretical peptide sequences after algorithm scoring and filtering.

[0165] Here, based on the third raw data of mass spectrometry, the mass-to-charge ratio and signal intensity of peptides in the sample (first-order spectrum), as well as the mass-to-charge ratio and signal intensity of fragment ions after peptide fragmentation (second-order spectrum) can be obtained.

[0166] Preferably, the FragPipe (v20.0) search software can be used to generate a theoretical secondary spectrum library based on the protein sequence of the key protein for protein identification. More preferably, the search parameters are as follows: the database is the MWY-23-2196-a_human_iRT.fasta database (82,504 sequences in total), with the addition of a reverse library and a contamination library to control the false positive rate caused by random matching and eliminate the influence of contaminating proteins; the mass tolerance for both precursor ions and fragment ions is 20 ppm; the digestive enzyme is stricttrypsin (allowing 2 maximum missed cleavages); the fixed modification is Carbamidomethyl (C); the variable modifications are Oxidation (M) and Acetyl (Protein N-term) (maximum 3 variable modifications); and the FDR is controlled within 1%. During PRM analysis, the secondary spectrum generated from the third raw mass spectrometry data is searched and compared with the theoretical secondary spectrum using Skyline (v23.1) software. After algorithmic scoring and filtering, the correctly matching theoretical peptide sequence is obtained.

[0167] Step C4, protein quantification:

[0168] Based on the theoretical peptide sequences, Skyline (v23.1) software was used to quantify the peptides and key proteins. The quantification rules were as follows: the peptide quantification value was the sum of the areas of the screened daughter ions, and the protein quantification value was the sum of the areas of the specific peptides, so as to obtain the quantification value of each key protein and peptide. After exporting the data, the quantification value of each key protein was factor-corrected according to the internal reference protein, that is, standardized, to obtain the standardized quantification value of the key protein.

[0169] like Figure 1C The image shows a differential expression map of key proteins in the validation cohort. p<0.05, p<0.01, p<0.001;

[0170] Step C5, Model Validation:

[0171] The optimized cerebral amyloid angiopathy (CAA) identification model was applied to the validation cohort. The age, sex, and standardized quantitative values ​​of key proteins of the subjects in the validation cohort were used as input variables to calculate the third key performance indicator of the model. If the third key performance indicator of the model met the requirements, the optimized cerebral amyloid angiopathy (CAA) model was used as the final cerebral amyloid angiopathy (CAA) identification model.

[0172] Preferably, the key performance indicators include: AUC value, accuracy, sensitivity, specificity, etc.

[0173] For example, such as Figure 6 and 7 As shown, in one embodiment of the present invention, the final diagnostic model obtained in an independent validation queue has an AUC value of 0.942, an accuracy of 91.7%, a sensitivity of 86.7%, and a specificity of 96.7%, indicating that the CAA diagnostic model provided by the present invention has good robustness; as Figure 12 and 13 As shown, in one embodiment of the present invention, the final bleeding risk prediction model has an AUC value of 0.740, an accuracy of 86.2%, a sensitivity of 88.0%, and a specificity of 75.0% in an independent validation queue.

[0174] Table 3 shows the performance of the diagnostic model in an independent validation queue.

[0175] Table 3

[0176]

[0177] Table 4 shows the performance of the bleeding risk prediction model in an independent validation cohort.

[0178] Table 4

[0179]

[0180] In the validation cohort, to validate the performance of the diagnostic model, targeted proteomics assays were used to quantify six key proteins (Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE 4, and PREP).

[0181] When conducting subsequent tests on the subjects, peripheral blood samples can be collected from the forearms of the subjects as in C1 to C5 for analysis to obtain quantitative values ​​of key proteins. These quantitative values ​​of key proteins are then input into the final cerebral amyloid angiopathy diagnostic model to determine whether the subject is a patient with cerebral amyloid angiopathy. Subsequently, the quantitative values ​​of key proteins of subjects identified as having cerebral amyloid angiopathy can be input into the hemorrhage risk prediction model to predict whether the future risk of cerebral hemorrhage in patients with cerebral amyloid angiopathy is high or low.

[0182] According to another aspect of the present invention, a computer-readable storage medium is also provided, having stored thereon computer-executable instructions, wherein when executed by a processor, the computer-executable instructions cause the processor to:

[0183] Acquire subjects and divide them into non-overlapping generation and validation queues;

[0184] Plasma samples were collected from the generation cohort, and non-targeted proteomics detection was performed on the plasma samples from the generation cohort to screen key proteins, and the quantitative value of each protein in each plasma sample of the generation cohort was obtained.

[0185] Based on the quantitative values ​​of each protein in each plasma sample from the generation cohort, key protein biomarkers were screened using differential protein analysis and unsupervised hierarchical clustering algorithms. Based on the quantitative values ​​of key proteins in each plasma sample from the generation cohort, modeling and validation were performed to obtain an optimized cerebral amyloid angiopathy model.

[0186] Plasma samples were collected from the validation cohort, and targeted proteomics detection was performed on the plasma samples to validate key proteins, obtaining quantitative values ​​of key proteins in each plasma sample of the validation cohort. Based on the quantitative values ​​of key proteins in each plasma sample of the validation cohort, the optimized cerebral amyloid angiopathy model was input for further validation to obtain the final cerebral amyloid angiopathy model.

[0187] According to another aspect of the present invention, a detection kit for identifying cerebral amyloid angiopathy is also provided, the detection kit comprising reagents for detecting biomarkers of the following key proteins: Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE 4, and PREP.

[0188] This invention aims to address the shortcomings of existing technologies, such as insufficient imaging diagnostics, limitations of invasive procedures, lack of systematic analysis, and lack of risk assessment models. It develops a system, storage medium, and reagent kit for constructing a brain amyloid angiopathy (CAA) identification model based on plasma proteomics technology. By constructing a diagnostic and prognostic prediction model containing six key protein biomarkers, it provides technology for non-invasive diagnosis and precision medicine of CAA. The detection method of this invention features high sensitivity, strong specificity, non-invasiveness, and ease of clinical application.

[0189] This invention uses high-throughput proteomics analysis (liquid chromatography-mass spectrometry, LC-MS / MS) combined with unsupervised clustering algorithms to screen six key protein biomarkers closely related to cerebral amyloid angiopathy (CAA) from peripheral plasma samples. These biomarkers include Pro-CTSH (procathepsin S), USP15 (ubiquitin-specific protease 15), ApoA-IV (apolipoprotein A-IV), Fibulin-5 (fibular protein-5), RNASE 4 (ribonuclease 4), and PREP (prolyl endopeptidase).

[0190] This invention also provides a model for CAA identification based on a combination of six peripheral blood protein biomarkers, which exhibits excellent performance and has been validated on multiple datasets. Accurate CAA identification is achieved through machine learning of plasma protein biomarkers.

[0191] For diagnosing CAA, the area under the receiver operating characteristic (AUC) in the training set of the first set is 1.000, with an accuracy of 100%, sensitivity of 100%, and specificity of 100%; in the validation set, the AUC is 0.985, with an accuracy of 93.3%, sensitivity of 91.7%, and specificity of 95.2%; in the test set, the AUC is 0.961, with an accuracy of 89.2%, sensitivity of 97.1%, and specificity of 80.0%; and in the independent validation cohort, the AUC is 0.942, with an accuracy of 91.7%, sensitivity of 86.7%, and specificity of 96.7%. The CAA diagnostic model provided by this invention exhibits good robustness.

[0192] For predicting the risk of bleeding from CAA, the AUC was 0.977 in the first set, with an accuracy of 96.3%, sensitivity of 95.4%, and specificity of 100%; the AUC was 0.808 in the test set, with an accuracy of 77.1%, sensitivity of 80.8%, and specificity of 66.7%; and the AUC was 0.740 in the independent validation cohort, with an accuracy of 86.2%, sensitivity of 88.0%, and specificity of 75.0%.

[0193] This invention innovatively develops a construction system, storage medium, and reagent kit for identifying cerebral amyloid angiopathy by combining non-targeted proteomics detection and targeted protein validation. Compared with existing technologies, it has the following significant advantages:

[0194] 1. Beneficial technological effects

[0195] 1) Non-invasive diagnostic methods:

[0196] This invention uses peripheral blood as the test sample, avoiding invasive procedures such as lumbar puncture, significantly reducing the physical burden and psychological stress on patients, and improving the clinical applicability of the method.

[0197] 2) High sensitivity and specificity:

[0198] Based on six key protein biomarkers (Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE4, and PREP), the constructed diagnostic and bleeding risk prediction models demonstrated excellent performance across multiple datasets. In an independent validation cohort, the model achieved an AUC of 0.942 for diagnosing CAA, with an accuracy of 91.7%, sensitivity of 86.7%, and specificity of 96.7%, outperforming existing diagnostic methods. For predicting bleeding risk in CAA, the model achieved an AUC of 0.740, with an accuracy of 86.2%, sensitivity of 88.0%, and specificity of 75.0%.

[0199] 3) The scientific basis of differential protein screening:

[0200] Non-targeted proteomics detection and targeted proteomics validation ensured the comprehensiveness and reliability of differential protein screening, laying a solid foundation for subsequent model construction.

[0201] 4) The robustness and generalization ability of the model:

[0202] Through design and machine learning modeling on multiple datasets (XGBoost), the model demonstrates good robustness and generalization ability in both the generation and validation queues, and is generalizable.

[0203] 2. Beneficial economic effects

[0204] 3) Simplify the testing process:

[0205] The standardized peripheral blood testing and targeted proteomics validation process is efficient and convenient, reducing the technical barriers to laboratory and clinical operations.

[0206] 4) Cost reduction:

[0207] Compared to traditional cerebrospinal fluid testing or complex imaging biomarker analysis, the blood testing method of this invention is lower in cost and more suitable for large-scale screening and early diagnosis.

[0208] 3. Beneficial effects on society

[0209] 3) Promote the development of precision medicine:

[0210] Based on proteomics technology and machine learning methods, this invention provides a technical foundation for personalized diagnosis and management of CAA, and helps to implement the concept of precision medicine.

[0211] 4) Reduce the social medical burden:

[0212] Early detection and intervention in CAA patients can reduce the consumption of medical resources caused by disease progression and alleviate the economic burden on families and society.

[0213] In summary, this invention has significant advantages in terms of technology, economy, and society, and can provide important support for the early and accurate diagnosis and personalized management of cerebral amyloid angiopathy, with broad clinical application prospects and promotional value.

[0214] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0215] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0216] Obviously, those skilled in the art can make various modifications and variations to the invention without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of the invention and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A system for constructing a model for identifying cerebral amyloid angiopathy, characterized in that, include: The classification module is used to obtain plasma samples from subjects and divide them into non-overlapping production and validation queues. The acquisition module is used to collect plasma samples from the generation queue, perform non-targeted proteomics detection on the plasma samples from the generation queue to screen key proteins, and obtain the quantitative value of each protein in each plasma sample from the generation queue. The training module is used to screen key protein biomarkers based on the quantitative values ​​of each protein in each plasma sample of the generation cohort using differential protein analysis and unsupervised hierarchical clustering algorithms; and to perform modeling and validation based on the quantitative values ​​of key proteins in each plasma sample of the generation cohort to obtain an optimized cerebral amyloid angiopathy model. The validation module is used to collect plasma samples from the validation cohort, perform targeted proteomics detection on the plasma samples to validate key proteins, and obtain quantitative values ​​of key proteins in each plasma sample of the validation cohort. Based on the quantitative values ​​of key proteins in each plasma sample of the validation cohort, the optimized cerebral amyloid angiopathy model is input for further validation to obtain the final cerebral amyloid angiopathy model. The training module is used to: divide the first set into a non-overlapping training set and a validation set; based on the quantitative values ​​of six key proteins in the training set, validation set, and test set, construct and train a diagnostic model for cerebral amyloid angiopathy using an extreme gradient boosting algorithm to obtain an optimized diagnostic model for cerebral amyloid angiopathy; wherein the training set, validation set, and test set each include non-overlapping patients with cerebral amyloid angiopathy and normal elderly controls. The verification module is used to: calculate the key performance indicators of the optimized cerebral amyloid angiopathy diagnostic model, including: area under the receiver operating curve, accuracy, sensitivity and specificity. If the key performance indicators of the optimized cerebral amyloid angiopathy diagnostic model meet the requirements, the optimized cerebral amyloid angiopathy diagnostic model is verified based on the quantitative values ​​of 6 key proteins in the verification cohort to obtain the final cerebral amyloid angiopathy diagnostic model. The training module is used to: construct and train a hemorrhage risk prediction model based on a first set and a test set using an extreme gradient boosting algorithm to obtain an optimized hemorrhage risk prediction model; wherein the first set and the test set each include non-overlapping patients with high-risk cerebral hemorrhage and patients with low-risk cerebral hemorrhage. The verification module is used to: calculate the key performance indicators of the optimized bleeding risk prediction model, including: area under the receiver operating curve, accuracy, sensitivity and specificity; if the key performance indicators of the optimized bleeding risk prediction model meet the requirements, then the optimized bleeding risk prediction model is verified based on the quantitative values ​​of 6 key proteins in the verification cohort to obtain the final bleeding risk prediction model. Biomarkers for the key proteins include: Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE 4 and PREP.

2. The system for constructing a brain amyloid angiopathy identification model as described in claim 1, characterized in that, The acquisition module is used for: Collect plasma samples from subjects in the production cohort; Each plasma sample was processed to remove high-abundance proteins, resulting in a low-abundance protein solution. Protein concentration and peptide preparation were performed on the low-abundance protein solution of each plasma sample to obtain the peptide solution of each plasma sample. Each peptide solution that generated the cohort was subjected to high-pH reverse-phase fractionation, and the fractions were collected and combined for subsequent mass spectrometry detection. Using both data-dependent and data-independent acquisition modes, mass spectrometry was performed on each peptide solution of the merged components in the generation queue to obtain quantitative values ​​of proteins and peptides in each plasma sample in the generation queue.

3. The system for constructing a brain amyloid angiopathy identification model as described in claim 2, characterized in that, The acquisition module is used for: The generated cohorts were randomly divided into a first cohort and a second cohort of non-overlapping subjects at a ratio of 7:

3. The first cohort and the second cohort each included non-overlapping patients with cerebral amyloid angiopathy and normal elderly controls, respectively. The quantitative values ​​of each protein in each plasma sample from the first cohort were used as the first set, and the quantitative values ​​of each protein in each plasma sample from the second cohort were used as the test set. In the first set, biomarkers for six key proteins associated with cerebral amyloid angiopathy were screened using differential protein analysis and unsupervised hierarchical clustering. The screening criteria for differential proteins were FDR-adjusted p-value < 0.05 and fold change > 1.2 or < 0.

83. The verification module is used for: Plasma samples were collected from subjects in the validation cohort. Each plasma sample was processed to remove high-abundance proteins, resulting in a low-abundance protein solution. The protein concentration of the low-abundance protein solution was determined using the BCA method. Based on the determined protein concentration, an equal volume of low-abundance protein was taken from the low-abundance protein solution of each plasma sample and subjected to dithiothreitol reduction, iodoacetamide alkylation, trypsin digestion, and peptide desalting. The peptide solution was then redissolved in 0.1% formic acid to obtain a peptide solution for each plasma sample. Targeted proteomics detection was performed on the peptide solution of each plasma sample in the validation cohort. Quantitative validation of six key proteins was conducted using parallel reaction monitoring mode, obtaining the quantitative values ​​of the key proteins in each plasma sample of the validation cohort.

4. The system for constructing a brain amyloid angiopathy identification model as described in claim 1, characterized in that, The diagnostic model for cerebral amyloid angiopathy is used to distinguish between patients with cerebral amyloid angiopathy and normal elderly controls. During the training of the diagnostic model for cerebral amyloid angiopathy, the age, sex, and standardized final quantitative values ​​of key proteins of the subjects in the training set are used as input variables, and patients with cerebral amyloid angiopathy or normal elderly controls in the subjects in the training set are used as labels. The hemorrhage risk prediction model is used to distinguish between high-risk and low-risk cerebral hemorrhage patients among those with cerebral amyloid angiopathy. During the training of the hemorrhage risk prediction model, the age, sex, and standardized final quantitative values ​​of key proteins of all cerebral amyloid angiopathy patients in the first set are used as input variables, and cerebral amyloid angiopathy patients in the first set who are at high or low risk of cerebral hemorrhage are used as labels.

5. The system for constructing a brain amyloid angiopathy identification model as described in claim 2, characterized in that, The mass spectrometry detection includes: Data-dependent acquisition was performed using a timsTOF Pro mass spectrometer. Data acquisition was performed using an Orbitrap Exploris 480 mass spectrometer.

6. The system for constructing a brain amyloid angiopathy identification model as described in claim 3, characterized in that, The targeted proteomics detection includes: The parallel reaction monitoring mode was used on a timsTOF Pro2 mass spectrometer to quantitatively validate the selected key proteins in peptide solutions from each plasma sample in the validation cohort.

7. A computer-readable storage medium having stored thereon computer-executable instructions, wherein, When the computer-executable instructions are executed by the processor, the processor causes the processor to: execute the system for constructing a cerebral amyloid angiopathy identification model as described in any one of claims 1 to 6.

8. A detection kit for identifying cerebral amyloid angiopathy, characterized in that, The detection kit, employing the system for constructing a brain amyloid angiopathy identification model as described in any one of claims 1 to 6, comprises reagents for detecting biomarkers of the following key proteins: Pro-CTSH, USP15, ApoA-IV, Fibulin-5, RNASE 4 and PREP.

Images