AI sarcopenia long-term risk prediction model based on proteomics data and application thereof

Abstract

The invention discloses an AI sarcopenia long-term risk prediction system and method based on proteomics data. According to the method, a multi-step feature selection strategy is adopted, finally seven proteins such as IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL are locked to serve as core prediction factors, clinical variables (such as age, gender and BMI) are fused, and an individualized 10-year sarcopenia risk prediction model is trained and generated through machine learning methods such as integrated learning and cross validation. The model and the prediction system based on the model not only have high prediction accuracy and good model calibration degree, but also realize model interpretability through a Shapley value (SHAP) interpretation algorithm, can reveal association strength and action mechanism of each protein and sarcopenia risk, and provide scientific basis for pathogenesis research and accurate intervention of diseases.

Description

TECHNICAL FIELD

[0001] The application relates to an AI sarcopenia long-term risk prediction model based on proteomics data and application thereof, and belongs to the technical field of biomedical detection and diagnosis. BACKGROUND

[0002] With the acceleration of global population aging, sarcopenia, as an important disease affecting the health and quality of life of the elderly, has increasingly become a problem that needs to be paid attention to in the field of global public health. However, the existing technology has limited methods for early prediction and risk assessment of sarcopenia, especially in large-scale population, long-term follow-up and comprehensive analysis of multi-omics data. Traditional prediction methods mostly rely on clinical symptoms or a single biomarker, which is difficult to achieve high-precision, personalized and long-term prediction of the risk of sarcopenia, limiting the implementation of early intervention and individualized health management of the disease. SUMMARY

[0003] The application aims at the deficiencies of the prior art, and provides an AI sarcopenia long-term risk prediction model based on proteomics data and a construction method and application thereof. The application adopts a multi-step feature selection strategy, finally screens seven proteins IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL as core prediction factors, and fuses clinical variables (such as age, gender, BMI, etc.), trains and generates an individualized 10-year sarcopenia risk prediction model (SarcoPro and SarcoPro+ model) through machine learning methods such as ensemble learning and cross-validation. The model not only has high prediction accuracy and good model calibration, but also realizes model interpretability through Shapley value (SHAP) explanation algorithm, can reveal the correlation strength and mechanism of each protein and sarcopenia risk, and provides a scientific basis for the pathogenesis research and precise intervention of the disease.

[0004] In order to achieve the above-mentioned purpose, the application adopts the following technical scheme:

[0005] In a first aspect, the application provides a construction method of an AI sarcopenia long-term risk prediction model based on proteomics data, comprising the following steps:

[0006] Step 1) Obtain proteomics data, and screen characteristic proteins as key factors for risk prediction;

[0007] Step 2) Obtain clinical variables;

[0008] Step 3) Establish a risk prediction model based on the characteristic proteins or the characteristic proteins combined with the clinical variable information;

[0009] The characteristic proteins include 7 proteins of IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL.

[0010] The clinical variables include age, gender, education, race, baseline grip strength and BMI.

[0011] Preferably, the method of screening characteristic proteins in step 1) comprises: training a preliminary classifier using a gradient boosting decision tree (LightGBM), ranking protein importance according to information gain, and introducing proteins step by step using a sequential forward selection method (SFS) until the AUC stops rising (no improvement for three consecutive times of DeLong test).

[0012] Preferably, the SarcoPro model is constructed based on the characteristic proteins using a LightGBM algorithm in step 3), or the SarcoPro+ model is constructed based on the characteristic proteins and clinical variable information using a LightGBM algorithm.

[0013] In a second aspect, the present application provides a sarcopenia risk prediction model, which is constructed according to the method of the first aspect of the present application.

[0014] In a third aspect, the present application provides any one of computer readable storage media:

[0015] 1) which stores a program for executing the method of the first aspect of the present application;

[0016] 2) which stores a program for executing the risk prediction model of the second aspect of the present application.

[0017] In a fourth aspect, the present application provides any one of electronic devices:

[0018] 1) which comprises a processor and a memory, the processor being configured to execute an executable program stored in the memory, the executable program being configured to run the method of the first aspect of the present application;

[0019] 2) which comprises a processor and a memory, the processor being configured to execute an executable program stored in the memory, the executable program being configured to run the risk prediction model of the second aspect of the present application.

[0020] Preferably, the electronic device further comprises an input device connected to the processor, configured to accept input information of a user.

[0021] Preferably, the electronic device further comprises a display device, configured to display the risk prediction result according to the instructions of the input device and the processor.

[0022] In a fifth aspect, the present application provides a sarcopenia prediction device, which comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the risk prediction model of the second aspect of the present application.

[0023] Preferably, the prediction device comprises a data acquisition module and a risk prediction module; the data acquisition module is used to acquire the content detection indicators of the 7 characteristic proteins of the subject or the content detection indicators and clinical variable indicators; the risk prediction module is connected to the data acquisition module and is used to input each of the indicators into the pre-trained risk prediction model to predict the probability of the occurrence of the sarcopenia risk of the subject for the reference of the clinical diagnosis of the doctor.

[0024] Preferably, the prediction device further comprises a feature extraction and selection module, an AI model construction module, and a result output and display module.

[0025] In a sixth aspect, the present application provides an application of a reagent for detecting biomarkers in a sample in the preparation of a product for predicting sarcopenia, wherein the biomarkers are IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL.

[0026] The reagent is a probe for recognizing the biomarker gene, a primer for amplifying the biomarker gene, or a binding agent (such as an antibody) for binding the biomarker protein.

[0027] Preferably, the sample is selected from a peripheral blood sample.

[0028] Preferably, the product is a nucleic acid membrane strip, a chip or a kit.

[0029] Preferably, the chip is a gene chip or a protein chip, the gene chip comprises oligonucleotide probes for the biomarker genes for detecting the transcription level of the biomarker genes, and the protein chip comprises specific binding agents for the biomarker proteins; the kit comprises a gene detection kit or a protein detection kit, the gene detection kit comprises reagents or chips for detecting the transcription level of the biomarker genes, and the protein detection kit comprises reagents or chips for detecting the expression level of the biomarker proteins.

[0030] Preferably, the kit is a qPCR kit, an immunoblotting detection kit, an immunochromatography detection kit, a flow cytometry analysis kit, an immunohistochemistry detection kit, an ELISA kit or an electrochemiluminescence detection kit.

[0031] In a seventh aspect, the present application provides an application of biomarkers in constructing a sarcopenia risk prediction model, wherein the biomarkers are IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL.

[0032] Compared with the prior art, the present application has the following beneficial effects:

[0033] 1) The present application introduces high-throughput proteomic data, combines advanced machine learning algorithms (such as LightGBM), automatically screens and integrates key predictors from thousands of protein markers, and establishes a high-performance long-term risk prediction model based on the screened predictors. Compared with the traditional method relying only on clinical information or a single biomarker, the model is significantly superior to the prior art in terms of discrimination accuracy (AUC, sensitivity, specificity, etc.) and risk calibration ability (Brier score, calibration slope, etc.), and can identify high-risk individuals earlier and more accurately, providing a scientific basis for early intervention and precision health management.

[0034] 2) Strong model interpretability, promoting clinical landing and trust improvement

[0035] The present application integrates explainable algorithms such as SHAP, which not only quantifies the marginal contribution of each protein to the prediction result, but also generates risk decomposition maps for individuals, visually displaying the positive and negative effects of each factor on individual sarcopenia risk. This technical feature overcomes the limitations of traditional "black box" AI models, greatly improving the trust and acceptability of the model in clinical practical application, and facilitating doctors and patients to understand and adopt the prediction results.

[0036] 3) Multi-modal data fusion, high-throughput automatic processing, improving efficiency and suitable for large-scale screening

[0037] The present application not only integrates proteomic data, but also combines multi-modal clinical information such as population sociology, lifestyle, physical indicators, etc., to achieve more comprehensive and personalized risk prediction. In addition, the present application uses automatic quality control and machine learning algorithms to automatically process large-scale proteomic data and missing information, greatly improving data processing efficiency, reducing errors caused by manual intervention and subjective judgment, and being suitable for application in large-scale population screening and follow-up cohort, with good scalability and economic benefits.

[0038] 4) Online and convenient deployment, promoting clinical transformation and health management landing

[0039] The system can be deployed as a web-based online application, and users only need to input key data to obtain individualized 10-year sarcopenia risk and visual interpretation results; this feature greatly reduces the application threshold of the model, facilitating its promotion in multiple scenarios such as physical examination centers, primary medical care, and health management institutions, and providing strong technical support for public health management and medical decision-making.

[0040] 5) The present application provides a sarcopenia long-term risk prediction system and method based on proteomics data and artificial intelligence algorithms, which can realize efficient, accurate and interpretable prediction of the risk of sarcopenia in individuals in the next 10 years, filling the technical gap in the related field and providing an innovative tool for precise prevention and control of senile diseases and health management; the model and system of the present application can be used as a clinical auxiliary decision-making tool to support large-scale population health screening, individualized risk early warning and follow-up monitoring; compared with the prior art, the present application breaks through the limitations of single biomarker prediction, realizes the deep integration of multi-omics data and AI algorithms, and significantly improves the sensitivity and specificity of long-term risk prediction of sarcopenia. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A schematic diagram for gradually introducing proteins by sequential forward selection (SFS) in the embodiments of the present application.

[0042] Figure 2 A schematic diagram for 5-fold nested cross-validation in the embodiments of the present application.

[0043] Figure 3 A ROC curve graph for comparing the SarcoPro and SarcoPro+ models with the conventional logistic regression model in the embodiments of the present application.

[0044] Figure 4 A PRC curve graph for comparing the SarcoPro and SarcoPro+ models with the conventional logistic regression model in the embodiments of the present application.

[0045] Figure 5 A calibration curve graph for the SarcoPro and SarcoPro+ models in the embodiments of the present application.

[0046] Figure 6 A decision curve (DCA) graph for the SarcoPro and SarcoPro+ models in the embodiments of the present application.

[0047] Figure 7 A SHAP method is used to analyze the marginal contribution of each protein to the model output in the embodiments of the present application.

[0048] Figure 8 An example graph of an online web application in the embodiments of the present application. DETAILED DESCRIPTION

[0049] In order to make the present application more apparent, the preferred embodiments are described in detail below with reference to the accompanying drawings.

[0050] The present application provides a method for constructing an AI sarcopenia long-term risk prediction model based on proteomics data, comprising the following steps:

[0051] Step 1) Obtain proteomics data, and screen characteristic proteins as risk prediction key factors;

[0052] Step 2) Obtain clinical variables;

[0053] Step 3) Establish a risk prediction model based on the characteristic proteins or in combination with the characteristic proteins and clinical variable information;

[0054] The characteristic proteins include 7 proteins of IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL;

[0055] The clinical variables include age, gender, education, race, baseline grip strength and BMI.

[0056] In some embodiments of the present application, the method for screening characteristic proteins in step 1) comprises: first training a preliminary classifier using gradient boosting decision tree (LightGBM), ranking protein importance according to information gain, and gradually introducing proteins using sequential forward selection method (SFS) until AUC stops rising (no improvement for three consecutive times of DeLong test).

[0057] In some embodiments of the present application, the step 3) comprises constructing a SarcoPro model based on the characteristic proteins using LightGBM algorithm or constructing a SarcoPro+ model in combination with the characteristic proteins and clinical variable information using LightGBM algorithm.

[0058] The present application also provides a sarcopenia risk prediction model, which is constructed according to the method of the first aspect of the present application.

[0059] The present application also provides any one of the following computer readable storage media:

[0060] 1) which stores a program for executing the method of the first aspect of the present application;

[0061] 2) which stores a program for executing the risk prediction model of the second aspect of the present application.

[0062] The present application also provides any one of the following electronic devices:

[0063] 1) the electronic device comprises a processor and a memory, the processor is configured to execute an executable program stored in the memory, the executable program is configured to run the method of the first aspect of the present application;

[0064] 2) the electronic device comprises a processor and a memory, the processor is configured to execute an executable program stored in the memory, the executable program is configured to run the risk prediction model of the second aspect of the present application.

[0065] In some embodiments of the present application, the electronic device further comprises an input device connected to the processor, configured to accept the input information of the user.

[0066] In some embodiments of the present application, the electronic device further comprises a display device configured to display the risk prediction result according to the instruction of the input device and the processor.

[0067] The present application also provides a sarcopenia prediction device, the risk prediction device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor executing the risk prediction model of the second aspect of the present application.

[0068] In some embodiments of the present application, the prediction device comprises a data acquisition module and a risk prediction module; the data acquisition module is configured to obtain the content detection index of the 7 characteristic proteins of the subject or the content detection index and the clinical variable index; the risk prediction module is connected to the data acquisition module and is configured to input each index into the pre-trained risk prediction model to predict the probability of the occurrence of the sarcopenia risk of the subject, so as to provide a reference for the clinical diagnosis of the doctor.

[0069] In some embodiments of the present application, the prediction device further comprises a feature extraction and selection module, an AI model construction module, and a result output and display module.

[0070] The present application also provides the use of a reagent for detecting a biomarker in a sample in the preparation of a product for predicting sarcopenia, the biomarker being IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL;

[0071] The reagent is a probe for recognizing the biomarker gene, a primer for amplifying the biomarker gene, or a binding agent (such as an antibody, etc.) for binding the biomarker protein.

[0072] In some embodiments of the present application, the sample is selected from a peripheral blood sample.

[0073] In some embodiments of the present application, the product is a nucleic acid film strip, a chip or a kit.

[0074] In some embodiments of the present application, the chip is a gene chip comprising oligonucleotide probes for the biomarker genes for detecting the transcription level of the biomarker genes, or a protein chip comprising specific binding agents of the biomarker proteins; the kit comprises a gene detection kit comprising reagents or chips for detecting the transcription level of the biomarker genes, or a protein detection kit comprising reagents or chips for detecting the expression level of the biomarker proteins.

[0075] In some embodiments of the present application, the kit is a qPCR kit, an immunoblotting detection kit, an immunochromatographic detection kit, a flow cytometry analysis kit, an immunohistochemical detection kit, an ELISA kit, or an electrochemiluminescence detection kit.

[0076] The present application also provides the use of biomarkers in constructing a sarcopenia risk prediction model, the biomarkers being IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1 and NEFL.

[0077] Embodiments

[0078] A method for constructing an AI sarcopenia long-term risk prediction model based on proteomics data, comprising the following steps:

[0079] 1) Data collection: collect samples (n=502132) from the UK Biobank (UKBB), exclude samples with missing follow-up data, samples with missing baseline sarcopenia data, and samples with missing plasma proteomics data, finally a total of 4918 samples are included, including 86 patients with sarcopenia within 10 years and 4832 normal controls without sarcopenia within 10 years. Collect plasma proteomics data (Olink Proximity Extension Assay detection), demographic and sociological data, lifestyle, diet, physical activity, health status, and other clinical data.

[0080] 2) Sarcopenia diagnosis: according to the European Working Group on Sarcopenia in Older People (EWGSOP2) standard, combined with grip strength (male <27kg, female <16kg) and skeletal muscle mass index (SMI, male <7.0kg / m 2 , female <5.5kg / m 2 ) for determination.

[0081] 3) Proteomics data processing: Plasma samples were detected by antibody chip using Olink platform, and the raw data was Normalized Protein Expression (NPX, Log2 scale). Samples that did not pass the built-in quality control of Olink were excluded, and data with a protein detection completeness rate of ≥85% was retained. Proteins with missing values >40% were excluded.

[0082] 4) Feature selection: First, a preliminary classifier was trained using LightGBM (gradient boosting decision tree), and the protein importance was sorted according to the information gain. The sequential forward selection method (SFS) was used to gradually introduce proteins until the AUC stopped rising (no improvement for three consecutive DeLong tests), as shown in Figure 1 .

[0083] 5) Model establishment: A total of 7 proteins, including IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1, and NEFL, were finally determined as key risk prediction factors. Based on the above 7 proteins, a SarcoPro model was established using the LightGBM model. The SarcoPro+ model was established by combining the 7 protein factors with clinical variables (age, gender, education, race, baseline grip strength, BMI).

[0084] 6) Model training and verification: The AI model was trained using historical medical records, and the five-fold nested cross-validation (inner training and outer validation) was used to optimize the parameters and evaluate the model. In each inner loop, the training set and the validation set were divided according to the ratio of 1:4, the training set was used for model training and hyperparameter optimization, and the validation set was used for model validation, as shown in Figure 2 .

[0085] 7) Evaluation of model performance in the validation set: The discriminant ability of the model was measured by AUC, AU-PRC, sensitivity, and specificity. The AUC of the SarcoPro model was 0.751, the AU-PRC was 0.993, the sensitivity was 0.719, and the specificity was 0.673; the AUC of the SarcoPro+ model was 0.899, the AU-PRC was 0.998, the sensitivity was 0.822, and the specificity was 0.802, all of which were better than the traditional logistic regression, as shown in Figures 3-4 . The calibration performance of the model was evaluated by Brier score and calibration slope, and the Brier score of the SarcoPro model was 0.174, and the calibration slope was 1.434; the Brier score of the SarcoPro+ model was 0.121, and the calibration slope was 1.117, both of which were well calibrated, as shown in Figure 5The clinical net benefit of the model at different risk thresholds is shown by the decision curve (DCA) method, which is superior to traditional models such as Figure 6 as shown.

[0086] 8) Explainability and individualized risk interpretation: Global interpretation: SHAP method is used to analyze the marginal contribution of each protein to the model output, and IGFBP2 is confirmed as the strongest predictor, as shown. Figure 7 Individual interpretation: Break-down atlas is generated for single cases to show the positive and negative effects of each protein on the risk score, which is convenient for clinicians to understand and apply.

[0087] Application deployment: The model has been integrated into an online web application, and users can obtain 10-year sarcopenia risk and visual interpretation results after inputting key variables, which is convenient for clinical promotion and patient self-test, as shown. Figure 8 The system and method are not only suitable for sarcopenia prediction, but also can be extended to other proteomic related chronic disease risk fields. As a preferred embodiment, the machine learning online calculation webpage can also directly interface with the hospital information management system to automatically call the 7 protein contents (protein content can be detected by antibody chip on Olink platform for plasma samples, and the NPX value of 7 proteins is called from the proteomic data) or the 7 proteins and the related clinical variables (age, gender, education, race, baseline grip strength, BMI) of the corresponding patient from the information management system, and then predict the corresponding sarcopenia risk probability. Herein, no limitation is made.

[0088] The above is only a preferred embodiment of the present application, and is not a limitation on the form and substance of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the present application, a number of improvements and supplements can also be made, which should be considered as the protection scope of the present application.

Claims

1. A method for constructing an AI-based long-term risk prediction model for sarcopenia based on proteomics data, characterized in that, Includes the following steps: Step 1) Obtain proteomics data and screen characteristic proteins as key factors for risk prediction; Step 2) Obtain clinical variables; Step 3) Establish a risk prediction model based on the characteristic protein or by combining the characteristic protein with clinical variable information; The characteristic proteins include seven proteins: IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1, and NEFL. The clinical variables included age, sex, education, race, baseline grip strength, and BMI.

2. The construction method according to claim 1, characterized in that, The method for screening feature proteins in step 1) includes: first, training a preliminary classifier using a gradient boosting decision tree, sorting protein importance by information gain, and then gradually introducing proteins using a sequential forward selection method until AUC improvement stagnates.

3. The construction method according to claim 1, characterized in that, In step 3), the SarcoPro model is constructed using the LightGBM algorithm based on the feature protein, or the SarcoPro+ model is constructed using the LightGBM algorithm in combination with the feature protein and clinical variable information.

4. A sarcopenia risk prediction model, characterized in that, The risk prediction model is constructed according to any one of claims 1-3.

5. Any of the following computer-readable storage media: 1) It stores a program for performing the method according to any one of claims 1-3; 2) It stores a program for executing the risk prediction model of claim 4.

6. Any of the following electronic devices: 1) The electronic device includes a processor and a memory, the processor being configured to execute an executable program stored in the memory, the executable program being configured to run the method according to any one of claims 1-3; 2) The electronic device includes a processor and a memory, the processor being used to execute an executable program stored in the memory, the executable program being used to run the risk prediction model of claim 4.

7. The electronic device according to claim 6, characterized in that, The electronic device also includes an input device connected to the processor for accepting user input information.

8. The electronic device according to claim 6, characterized in that, The electronic device also includes a display device that displays the risk prediction results according to instructions from the input device and the processor.

9. A sarcopenia prediction device, characterized in that, The risk prediction device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor executing the risk prediction model of claim 4.

10. The sarcopenia prediction device according to claim 9, characterized in that, The prediction device includes a data acquisition module and a risk prediction module. The data acquisition module is used to acquire the content detection indicators of the seven characteristic proteins of the subject or the content detection indicators and clinical variable indicators. The risk prediction module is connected to the data acquisition module and is used to input each of the indicators into a pre-trained risk prediction model to predict the probability of sarcopenia in the subject, so as to provide doctors with reference for clinical diagnosis.

11. The sarcopenia prediction device according to claim 9, characterized in that, The prediction device also includes a feature extraction and selection module, an AI model construction module, and a result output and display module.

12. The application of reagents for detecting biomarkers in samples in the preparation of products for predicting sarcopenia, characterized in that, The biomarkers are IGFBP2, ACP5, LEP, FOLR3, OXT, PTPRZ1, and NEFL; The reagent is a probe for recognizing the biomarker gene, a primer for amplifying the biomarker gene, or a binding agent for binding the biomarker protein.

13. The application according to claim 12, characterized in that, The samples were selected from peripheral blood samples. And / or, the product is a nucleic acid membrane strip, chip, or reagent kit.

14. The application of biomarkers in constructing a sarcopenia risk prediction model, characterized in that, The biomarkers are IGFBP2, ACP5, LEP, FORR3, OXT, PTPRZ1, and NEFL.

Images