A health life prediction method, a prediction model construction method and a prediction system
By obtaining mediating effect values to correct protein expression data and constructing a comprehensive prediction model, this solves the problem that existing technologies have not deeply explored the influence pathways of environmental and disease states, and achieves accurate prediction of individual healthy lifespan.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INSTITUTE OF BASIC MEDICAL SCIENCES CHINESE ACADEMY OF MEDICAL SCIENCES
- Filing Date
- 2026-04-01
- Publication Date
- 2026-06-09
AI Technical Summary
Existing healthy life expectancy prediction models fail to delve into the pathways through which environmental factors and disease states influence life through plasma proteins, resulting in inaccurate predictions.
By obtaining mediating effect values to correct protein expression data, eliminating indirect influencing factors such as environment and lifestyle, a comprehensive prediction model is constructed that integrates protein expression data and disease risk scores.
It improves the accuracy and clinical interpretability of healthy life expectancy prediction, enabling precise assessment of an individual's future health risks.
Smart Images
Figure CN122177462A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical fields of computer technology and biomedical information, and in particular to a method for predicting healthy life expectancy, a method for constructing a prediction model, and a prediction system. Background Technology
[0002] With the accelerating aging of the population, predicting healthy life expectancy has become an important research direction in the field of precision medicine. Currently, health risk prediction models based on proteomics data are gradually becoming a research hotspot. Researchers analyze the statistical association between individual protein expression levels and health outcomes, and use machine learning algorithms to construct predictive models to output an individual's health risk score.
[0003] However, due to the high stability and irreversibility of individual genes, intervention and regulation at the gene level are quite difficult. Environmental factors (including the external environment, lifestyle, and psychological factors) play a crucial role in the regulation of healthy aging by plasma proteins. Studies have shown that environmental factors such as PM2.5 can affect the levels of inflammatory factors such as interleukin and tumor necrosis factor, increasing the risk of COPD; persistent organic pollutants (PFAS) further exacerbate health risks by affecting blood lipid levels. Some lifestyle factors, such as lifelong smoking, are significantly associated with the risk of various mental illnesses (such as anxiety disorders, bipolar disorder, and major depressive disorder) through plasma proteins; dietary habits play a key role in regulating weight and metabolic balance by affecting leptin levels. Long-term psychological stress may also lead to elevated levels of inflammation-related proteins, thereby increasing the risk of chronic diseases. Furthermore, various disease states themselves also have a significant impact on healthy lifespan.
[0004] Existing models are mostly trained based on the statistical association between protein expression levels and health outcomes, failing to delve into the complex pathways through which environmental factors and disease states influence healthy lifespan via plasma proteins. Ignoring these mediating pathways can lead to biased estimations of protein effects, affecting the model's predictive accuracy. Summary of the Invention
[0005] The purpose of this application is to provide a method for predicting healthy lifespan, a method for constructing a prediction model, and a prediction system. By obtaining the mediating effect value, the protein expression data is corrected to eliminate indirect influencing factors such as environment and lifestyle. The corrected data is then integrated with the disease risk score to construct a comprehensive prediction model. This solves the technical problem of inaccurate prediction caused by ignoring the mediating path in the prior art, and achieves accurate prediction of individual healthy lifespan.
[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for constructing a healthy lifespan prediction model, comprising: acquiring training data and pre-determined mediating effect values of multiple longevity causal proteins affecting healthy lifespan through several mediating links; the training data including protein expression data, age data, and gender data of multiple samples; based on the training data and using the multiple longevity causal proteins as input features, training a baseline risk model for outputting a preliminary health risk score; calculating the corrected protein risk contribution of each sample based on the protein expression data of each sample and the mediating effect values corresponding to the multiple longevity causal proteins, using a preset correction rule; wherein, the correction rule is used to weaken the part of the protein expression data that indirectly affects healthy lifespan through the mediating links based on the mediating effect values, so as to highlight the direct effect of the corrected protein expression data on healthy lifespan; using the corrected protein risk contribution and the disease risk score obtained by evaluating the samples based on multiple preset disease risk models as new input features to retrain the baseline risk model, thereby obtaining the final comprehensive health lifespan risk prediction model.
[0007] Secondly, this application provides a method for predicting healthy lifespan, comprising: acquiring protein expression data, age data, and gender data of an individual to be tested; inputting the protein expression data, age data, and gender data of the individual to be tested into a comprehensive health lifespan risk prediction model constructed by the above-mentioned method for constructing a health lifespan prediction model; and acquiring and outputting a health risk score output by the comprehensive health lifespan risk prediction model, which characterizes the degree of health risk of the individual to be tested in a future preset time period.
[0008] Thirdly, this application provides a healthy lifespan prediction system, comprising: a data acquisition module configured to acquire protein expression data, age data, and gender data of an individual to be tested; a model processing module configured to input the protein expression data, age data, and gender data of the individual to be tested into a comprehensive health lifespan risk prediction model constructed by the above-described healthy lifespan prediction model construction method; and an output module configured to output a health risk score output by the comprehensive health lifespan risk prediction model, which characterizes the degree of health risk of the individual to be tested within a preset future time period.
[0009] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method for predicting healthy lifespan, a method for constructing a prediction model, and a prediction system. By acquiring training data and mediation effect values, it solves the problems of missing values and inaccurate age information in raw proteomics data, achieving complete processing of training data and accurate age information. Through training, a baseline risk model is obtained to output a preliminary health risk score, solving the problem of directly quantifying health risk from protein expression levels and enabling preliminary proteomics-based health risk assessment for individuals. By calculating the corrected protein risk contribution for each sample, and particularly by using mediation effect values to weaken the part that indirectly affects healthy lifespan through mediating pathways, it addresses the problem of existing technologies neglecting environmental and lifestyle factors. The problem of protein effect estimation bias caused by the mediating pathway is highlighted by this study, which emphasizes the direct effect of proteins on healthy lifespan, thereby improving the accuracy and biological interpretability of risk contribution estimation. By evaluating samples based on multiple pre-defined disease risk models, a disease risk score is obtained, addressing the issue that traditional prediction models do not fully consider the impact of disease status on healthy lifespan, and enabling independent quantitative assessment of the risks of multiple major disease categories. The final comprehensive healthy lifespan risk prediction model is obtained through retraining, solving the problem of incomplete prediction dimensions in single proteomics, achieving deep integration of direct protein effects and disease risk information, and significantly improving the accuracy and clinical interpretability of the model in predicting individual future healthy lifespan. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a diagram illustrating the application environment of the healthy lifespan prediction method and prediction model construction method in the embodiments of this application; Figure 2 A flowchart illustrating a method for constructing a healthy life expectancy prediction model provided in Embodiment 1 of this application; Figure 3 Table of effect coefficients β (partial examples) of longevity causal proteins provided in Embodiment 1 of this application; Figure 4 This is a flowchart illustrating the modification rules in Embodiment 1 of this application; Figure 5 This is a flowchart illustrating the process of constructing multiple disease risk models in Embodiment 1 of this application; Figure 6 This is a flowchart illustrating a method for predicting healthy life expectancy provided in Embodiment 2 of this application; Figure 7 This is a schematic diagram of the functional modules of a healthy lifespan prediction system provided in Embodiment 3 of this application. Detailed Implementation
[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0013] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0014] The healthy lifespan prediction method and prediction model construction method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up separately, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send the protein expression data, age data, and gender data of the individual to be tested to server 102. After receiving the data, server 102 inputs it into a comprehensive health lifespan risk prediction model pre-constructed using the health lifespan prediction model construction method of this application, obtaining a health risk score characterizing the degree of health risk of the individual to be tested within a preset future time period. Server 102 can then feed back the obtained health risk score to terminal 101. Furthermore, in some embodiments, the above-mentioned health lifespan prediction method can also be implemented separately by server 102 or terminal 101, such as terminal 101 directly processing the protein expression data, age data, and gender data to be processed.
[0015] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.
[0016] Example 1: A method for constructing a healthy life expectancy prediction model, such as... Figure 2 As shown, the method is executed by a computer device. In this embodiment of the application, the method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps: S11: Acquire training data and pre-determined mediating effect values of multiple longevity causal proteins affecting healthy lifespan through several mediating pathways; training data includes protein expression data, age data, and gender data of multiple samples.
[0017] S12, based on training data and using multiple longevity causal proteins as input features, trains a baseline risk model for outputting a preliminary health risk score.
[0018] S13. Based on the protein expression data of each sample and the mediating effect values corresponding to each of the multiple longevity causal proteins, the corrected protein risk contribution of each sample is calculated using a preset correction rule.
[0019] S14. The corrected protein risk contribution and the disease risk score obtained by evaluating the sample based on multiple preset disease risk models are used together as new input features to retrain the baseline risk model and obtain the final comprehensive risk prediction model for healthy lifespan.
[0020] By implementing steps S11 to S14 above, and obtaining training data and mediation effect values in step S1, the problems of missing values and inaccurate age information in the original proteomics data are solved, and the training data is fully processed and the age is made more accurate. Step S2 trains a baseline risk model for outputting a preliminary health risk score, addressing the issue of directly quantifying health risk from protein expression levels and enabling preliminary proteomics-based health risk assessment for individuals. Step S3 calculates the corrected protein risk contribution for each sample, particularly by using mediating effect values to weaken the portion indirectly affecting healthy lifespan through mediating pathways. This addresses the problem of existing technologies neglecting mediating pathways such as environment and lifestyle, leading to biased protein effect estimation and highlighting the direct effect of proteins on healthy lifespan, thereby improving the accuracy and biological interpretability of risk contribution estimation. Step S4 evaluates samples based on multiple pre-set disease risk models to obtain disease risk scores, addressing the issue that traditional prediction models do not fully consider the impact of disease states on healthy lifespan and enabling independent quantitative assessment of risks for multiple major disease categories. Step S5 retrains the model to obtain the final comprehensive healthy lifespan risk prediction model, addressing the incompleteness of single-proteomics prediction dimensions and achieving deep integration of direct protein effects and disease risk information. This significantly improves the accuracy and clinical interpretability of the model in predicting an individual's future healthy lifespan.
[0021] In practice, the training data in step S11 includes protein expression data, age data, and gender data from multiple samples. In this embodiment, the training data comes from 16,320 participants from the UK Biobank, including complete proteomics data, age, and gender data of the participants.
[0022] Protein expression data are standardized protein expression levels (NPX). Missing values in the proteomics data were imputed using the Multiple Imputation (MICE) method, implemented using the miceforest package (v3.0) in Python. Only samples with a missing proteomics data ratio of less than 30% were retained. Since UK Biobank only provides integer ages, the precise age of all samples was estimated using birth date and enrollment time information to obtain a decimal age.
[0023] In this embodiment, there are 60 longevity causal proteins. The effect size of a single longevity causal protein on healthy lifespan through a specific mediating link (i.e., mediating factor) is used as the effect coefficient β. The effect coefficient β value is predetermined by Mendelian randomization analysis combined with the delta method. There are a total of 1686 mediating links of health-related factors, and each longevity causal protein will be involved through several mediating links. In addition, the mediating effect value of each longevity causal protein is the algebraic sum of the corresponding β values of all mediating factors.
[0024] like Figure 3 The table showing the effect coefficients β of longevity causal proteins (partial examples) illustrates this. Each effect coefficient β represents the magnitude of the effect of that plasma protein on healthy lifespan through a specific mediating factor. Positive values indicate that the plasma protein increases health risk through that mediating pathway, while negative values indicate a decrease in health risk. The same plasma protein may correspond to multiple mediating pathways, such as... Figure 3 As shown, the ADA2 plasma protein corresponds to 33 mediating factors. The mediating effect value of the protein is obtained by summing the β values of the 33 mediating factors (i.e., -0.0487).
[0025] In this embodiment, the prediction task is defined as survival status classification over a 16-year timeframe. This time window is selected based on the following considerations: the average follow-up period of the UK Biobank training data used is 16 years, and this time window maximizes sample size balance and minimizes systematic bias caused by class imbalance. Those skilled in the art will understand that this time window can be adjusted according to the follow-up period of different datasets or actual prediction needs, for example, set to 10 years, 15 years, or 20 years.
[0026] In practical implementation, during the baseline risk model training process, the training data obtained in step S11 can be randomly divided into a training set (80%) and a validation set (20%) with comparable baseline features. Specifically, the aforementioned 60 longevity causal proteins can be used as input features. The classification learner module of MATLAB R2021b was used to systematically evaluate 30 classification algorithms across 8 main categories, and five-fold cross-validation was performed. Based on the cross-validation performance, prioritizing AUC and accuracy metrics, the Light GradientBoosting Machine (LightGBM), a sub-model under the Gradient Boosted Trees (GBT) algorithm category, was selected as the baseline model. LightGBM is an efficient ensemble learning method based on gradient boosting decision trees. Its model architecture consists of a series of additive decision trees, as follows: Equation (1); In equation (1), For the sample The predicted output, where K is the total number of trees. For the set of all regression tree functions, each tree The input features are mapped to leaf node scores.
[0027] To ensure clinical interpretability, the preliminary health risk score output by the baseline risk model is mapped using a 40-100 scoring system.
[0028] In specific implementation, such as Figure 4 As shown, the correction rule in step S13 is implemented through the following sub-steps: S131, Obtain the effect coefficient of each longevity causal protein. and effect coefficient Convert to ratio .
[0029] Specifically, effect coefficient Predetermined through Mendelian randomization analysis.
[0030] S132, Obtain the expression level of the i-th longevity causal protein in the sample. And the average expression level of the i-th longevity causal protein in all samples within the age group to which the sample belongs. and standard deviation .
[0031] S133, based on the effect coefficient Expression level Average expression level and standard deviation Calculate the original risk contribution of the i-th longevity causal protein.
[0032] Specifically, the formula for calculating the original risk contribution is as follows: ; S134, Obtain the total effect value of all longevity causal proteins, and calculate the ratio of the mediating effect value to the total effect value of all longevity causal proteins. and the total ratio Perform pre-defined standard processing to obtain the standardized sum ratio.
[0033] Specifically, the total effect value of each longevity causal protein was independently predetermined through Mendelian randomization analysis, and the ratio of the sum of the mediating effect values of all longevity causal proteins to the sum of the total effect values of all longevity causal proteins was used as the sum ratio. .
[0034] Specifically, the preset standard treatment is standard deviation standardization, i.e., Z-score standardization. Therefore, the sum ratio after Z-score standardization is expressed as: .
[0035] S135, the correction factor is calculated based on the standardized sum ratio.
[0036] Specifically, the formula for calculating the correction factor is as follows: .
[0037] S136, sum the original risk contributions of all longevity causal proteins to obtain the cumulative original risk contribution of the sample.
[0038] Specifically, the formula for calculating the cumulative original risk contribution is as follows: , where n is the total number of longevity causal proteins.
[0039] S137 multiplies the cumulative original risk contribution by the correction factor to obtain the corrected protein risk contribution of the sample.
[0040] Specifically, the revised formula for calculating protein risk contribution is as follows: Equation (2); In equation (2), This is represented as the corrected protein risk contribution; This is represented as the original risk contribution; This represents the cumulative original risk contribution; Represented as a correction factor; This completes the correction of the risk contribution based on protein expression data for each sample. The corrected protein risk contribution highlights the direct effect of proteins on healthy lifespan and weakens the indirect effects mediated by factors such as lifestyle.
[0041] In specific implementation, such as Figure 5 As shown, the multiple pre-defined disease risk models in step S14 are pre-built in the following ways: S141, Obtain disease diagnosis data from multiple samples in the training data.
[0042] Specifically, the disease diagnosis data covers multiple preset disease categories; the preset disease categories are divided into 34 major disease categories based on ICD categories.
[0043] S142, for each preset disease category, after excluding samples of the disease category already contracted, the protein expression data of the samples are used as input, and the gradient boosting tree algorithm is used to train a disease risk model for outputting the disease risk score of the disease category.
[0044] Specifically, all models were developed and validated using a rigorous 5x cross-validation framework.
[0045] S143, the trained disease risk models are screened according to the preset performance threshold, and the disease risk models that meet the performance threshold are selected as multiple preset disease risk models.
[0046] Specifically, the preset performance threshold is an area under the average cross-validation curve (AUC) greater than 0.7. In this embodiment, using this threshold, 17 disease risk models with robust predictive performance were selected from 34 major disease categories.
[0047] S144 standardizes the output of the disease risk model to ensure consistency with the scoring scale of the baseline risk model.
[0048] Specifically, the standardized mapping process maps the output of the disease risk model to a standardized risk score of 40-100.
[0049] In specific implementation, the construction of the comprehensive risk prediction model in step S14 involves using the corrected protein risk contribution calculated in step S13 and the disease risk score obtained by evaluating the sample based on multiple preset disease risk models in step S14 as new input features to retrain the baseline risk model and obtain the final comprehensive risk prediction model for healthy lifespan.
[0050] In this implementation, to evaluate the biological and clinical effectiveness of the comprehensive risk model, the model performance was validated. Participants (16,320 participants) were divided into three risk groups based on the risk scores output by the model: low risk (85-100), intermediate risk (60-85), and high risk (40-60). Survival analyses were performed using the R packages survival (v3.8.3) and survminer (v0.5.0), generating Kaplan-Meier curves for all-cause mortality and longevity-related disease incidence for these stratifications. The results showed significant differences in the survival curves among the three risk groups, with the high-risk group exhibiting significantly higher mortality and morbidity rates, validating the model's ability to differentiate long-term health outcomes.
[0051] Furthermore, the association between the model's risk score and the established biological age clock was examined. Biological age (BA) was calculated using the Klemera-Doubal method (KDM) and PhenoAge (PA) algorithm via the BioAge software package. Age acceleration (AA) was quantified as the regression residual of actual age (CA) against biological age (BA). The results showed a significant correlation between the model's risk score and these independent biomarkers of physiological aging, providing validation evidence for the model's characterization of potential aging processes.
[0052] Example 2: A method for predicting healthy life expectancy, such as Figure 6 As shown, it includes the following steps: S21, Obtain protein expression data, age data, and gender data of the individual to be tested.
[0053] Specifically, blood samples were obtained from the individuals to be tested, and their protein expression data were acquired using proteomics detection technology. Simultaneously, the individuals' actual age and sex were recorded. The protein expression data underwent the same preprocessing as in Example 1, including standardization and imputation of missing values.
[0054] S22, input the protein expression data, age data, and gender data of the individual to be tested into the comprehensive risk prediction model for healthy lifespan.
[0055] Specifically, the protein expression data, age data, and gender data of the individual to be tested are input into the comprehensive risk prediction model for healthy lifespan, which is constructed using the method described in Example 1 for constructing a healthy lifespan prediction model.
[0056] Specifically, the model first calculates the corrected protein risk contribution of the individual under test based on their protein expression data and age data, combined with pre-determined longevity causal proteins and their effect coefficients, mediating effect values, and other parameters, following the method described in step S13 of Example 1. Simultaneously, the individual's protein expression data is input into multiple pre-defined disease risk models to obtain disease risk scores for each disease category. Finally, the corrected protein risk contribution is merged with the multiple disease risk scores and input into the core algorithm of the comprehensive risk prediction model.
[0057] S23, acquire and output the health risk score, which is used to characterize the degree of health risk of the individual under test in the future within a preset time period, as output by the comprehensive risk prediction model for healthy lifespan.
[0058] Specifically, in this embodiment, the preset future time period is 16 years, and the health risk score ranges from 40 to 100 points. A higher score indicates a lower future health risk and a longer expected healthy lifespan; a lower score indicates a higher future health risk and a shorter expected healthy lifespan.
[0059] Depending on the specific application scenario, the output health risk score can be used for various purposes such as health management recommendations, disease prevention and intervention, and medical resource allocation.
[0060] Example 3: A healthy lifespan prediction system, used to implement the healthy lifespan prediction method described in Example 2. For example... Figure 7 As shown, the system includes: The data acquisition module 201 is configured to acquire protein expression data, age data, and sex data of the individual to be tested. This module may include a data interface with a proteomics detection device to support direct reading of detection results; it may also include a human-computer interaction interface to support manual input of basic information of the individual to be tested.
[0061] The model processing module 202, connected to the data acquisition module 201, is configured to input the protein expression data, age data, and gender data of the individual to be tested into the comprehensive health lifespan risk prediction model constructed using the method described in Example 1. This module includes a storage unit and a computation unit. The storage unit stores the pre-trained comprehensive risk prediction model and its related parameters (including a list of longevity causal proteins, effect coefficients, mediating effect values, disease risk models, etc.). The computation unit performs forward computation of the model and outputs a health risk score for the individual to be tested.
[0062] The output module 203, connected to the model processing module 202, is configured to output a health risk score from the comprehensive health lifespan risk prediction model, which characterizes the degree of health risk of the individual under test within a preset future time period.
[0063] It is understood that the healthy lifespan prediction system of this embodiment can be deployed on a local server, cloud server or mobile terminal device, and provide healthy lifespan prediction services to medical institutions, physical examination centers or individual users through Web services, applications or API interfaces.
[0064] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the healthy lifespan prediction method described in Embodiment 2.
[0065] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0066] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0067] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0068] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Furthermore, any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory.
[0069] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0070] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for constructing a healthy life expectancy prediction model, characterized in that, include: Acquire training data and pre-determined mediating effect values of multiple longevity causal proteins influencing healthy lifespan through several mediating pathways; The training data includes protein expression data, age data, and gender data from multiple samples; Based on the training data and using multiple longevity causal proteins as input features, a baseline risk model is trained to output a preliminary health risk score. Based on the protein expression data of each sample and the mediating effect values corresponding to the various longevity causal proteins, the corrected protein risk contribution of each sample is calculated using a preset correction rule. The correction rule is used to weaken the portion of the protein expression data that indirectly affects healthy lifespan through the mediating pathway, based on the mediating effect value, so as to highlight the direct effect of the corrected protein expression data on healthy lifespan. The modified protein risk contribution and the disease risk score obtained by evaluating the sample based on multiple preset disease risk models are used together as new input features to retrain the baseline risk model, resulting in the final comprehensive risk prediction model for healthy lifespan.
2. The method for constructing a healthy life expectancy prediction model according to claim 1, characterized in that, The calculation of the corrected protein risk contribution for each sample using preset correction rules includes: Obtain the effect coefficient of each longevity causal protein and the effect coefficient Convert to ratio The effect coefficient Predetermined through Mendelian randomization analysis; Obtain the expression level of the i-th longevity causal protein in the sample. And the average expression level of the i-th longevity causal protein in all samples within the age group to which the sample belongs. and standard deviation ; According to the effect coefficient The expression level The average expression level and the standard deviation Calculate the original risk contribution of the i-th longevity causal protein, wherein the original risk contribution is calculated using the following formula: ; Obtain the total effect value of all longevity causal proteins, and calculate the ratio of the mediating effect value to the total effect value of all longevity causal proteins. and the sum ratio Perform pre-defined standard processing to obtain the standardized sum ratio; Based on the standardized sum ratio, the correction factor is calculated, and the formula for calculating the correction factor is as follows: ; The original risk contributions of all longevity causal proteins are summed to obtain the cumulative original risk contribution of the sample. The formula for calculating the cumulative original risk contribution is as follows: , where n is the total number of longevity causal proteins; Multiplying the cumulative original risk contribution by the correction factor yields the corrected protein risk contribution of the sample. The formula for calculating the corrected protein risk contribution is as follows: .
3. The method for constructing a healthy life expectancy prediction model according to claim 2, characterized in that, The preset standard processing is standard deviation standardization processing.
4. The method for constructing a healthy life expectancy prediction model according to claim 1, characterized in that, The baseline risk model is a machine learning model based on gradient boosting decision trees.
5. The method for constructing a healthy life expectancy prediction model according to claim 1, characterized in that, The pre-built multiple disease risk models are constructed in advance in the following ways: Obtain disease diagnosis data from multiple samples in the training data; the disease diagnosis data covers multiple preset disease categories; For each preset disease category, after excluding samples that have already contracted the disease category, the protein expression data of the samples are used as input, and the gradient boosting tree algorithm is used to train a disease risk model for outputting the disease risk score of the disease category. The trained disease risk models are screened according to a preset performance threshold, and the disease risk models that meet the performance threshold are selected as the preset multiple disease risk models. The output of the disease risk model is standardized and mapped to be consistent with the output scoring scale of the baseline risk model.
6. The method for constructing a healthy life expectancy prediction model according to claim 5, characterized in that, The preset performance threshold is that the area under the average cross-validation curve is greater than 0.7; The standardized mapping process maps the output of the disease risk model to a standardized risk score of 40-100.
7. The method for constructing a healthy life expectancy prediction model according to claim 1, characterized in that, The protein expression data in the training data are standardized protein expression levels. Missing values in the protein expression data of the training data were imputed using a multiple imputation method, and only samples with a missing proteomics data ratio of less than 30% were retained.
8. The method for constructing a healthy life expectancy prediction model according to claim 1, characterized in that, The age data in the training data is a decimal form calculated from the birth information and enrollment time information of the samples.
9. A method for predicting healthy life expectancy, characterized in that, include: Obtain protein expression data, age data, and gender data of the individuals to be tested; The protein expression data, age data, and gender data of the individual to be tested are input into the comprehensive risk prediction model for healthy lifespan constructed using the method for constructing a healthy lifespan prediction model according to any one of claims 1-8; Obtain and output the health risk score, which is used to characterize the degree of health risk of the individual under test in the future within a preset time period, as output by the comprehensive health life risk prediction model.
10. A healthy lifespan prediction system, characterized in that, include: The data acquisition module is configured to acquire protein expression data, age data, and gender data of the individual to be tested; The model processing module is configured to input the protein expression data, age data, and gender data of the individual to be tested into the comprehensive risk prediction model for healthy lifespan constructed using the healthy lifespan prediction model construction method according to any one of claims 1-8; The output module is configured to output a health risk score, which is a characteristic of the degree of health risk of the individual under test within a preset time period, as output by the comprehensive health lifespan risk prediction model.