Method and system for dynamically predicting disease risk of teenagers

Abstract

The invention provides a teenager disease risk dynamic prediction method and system, and the method comprises the steps: obtaining an electronic health record and subcutaneous metabolite dynamic concentration data of a teenager, and the electronic health record comprises chronic disease related information, a genetic risk value and historical diagnosis data; performing multi-modal fusion on the metabolite concentration data, the genetic risk value and the historical diagnosis data according to a time axis to form a mixed time sequence information set; adaptive optimization is carried out on the mixed time sequence information set, and the collaborative change trend of metabolite fluctuation characteristics and genetic characteristics is identified from the optimized mixed time sequence information set; generating a dynamic graph by adopting a graph neural network based on the collaborative change trend; and tracking an abnormal conduction chain of a metabolic pathway in the dynamic map in real time, and calculating a risk probability matched with the growth stage of the teenagers according to the abnormal conduction chain and chronic disease related information. The dynamic prediction precision and timeliness of the chronic disease risk of teenagers are improved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent medical health monitoring and dynamic risk assessment, and in particular to a method and system for dynamically predicting adolescent disease risks. Background Art

[0002] Early risk prediction of chronic diseases in adolescents (such as diabetes and obesity) requires solving the technical difficulties of linking dynamic biomarker monitoring with genetic risks. Since adolescents are in a critical period of growth and development, their metabolic characteristics are highly volatile and have individual differences. Traditional static assessment methods are difficult to capture real-time risk changes. There is an urgent need to establish a time series prediction model that integrates dynamic metabolic data and genetic characteristics.

[0003] Currently, a time series prediction model based on a recurrent neural network is used. Time series data of metabolites such as blood glucose and cortisol are collected through wearable devices. Combined with the genetic risk scores in electronic medical records, the long-short-term memory network is used to explore the long-term dependence of metabolite fluctuations and output risk prediction values ​​at fixed time intervals.

[0004] This scheme can only process the metabolite time series characteristics of a single dimension, and the genetic risk score is calculated in the form of static parameters, which simplifies the nonlinear interactions between metabolic pathways; the discrete risk assessment results output by the model cannot reflect the risk transmission pathway during the continuous changes in the metabolic status of adolescents, and the prediction results are not coupled with the characteristics of the growth stage. Summary of the Invention

[0005] The present application provides a method and system for dynamic prediction of adolescent disease risk, which is used to solve the problems of low accuracy and poor timeliness of dynamic prediction of adolescent chronic disease risk in the existing technology.

[0006] In a first aspect, the present application provides a method for dynamically predicting disease risks in adolescents, comprising:

[0007] Obtaining electronic health records of adolescents and dynamic concentration change data of individual subcutaneous metabolites of adolescents, wherein the electronic health records include information related to chronic diseases, genetic risk values, and historical diagnosis data;

[0008] Performing multimodal fusion on the dynamic concentration change data, the genetic risk value, and the historical diagnosis data according to a time axis to generate a mixed time series information set;

[0009] Adaptively optimizing the mixed time series information set, and identifying the coordinated change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set;

[0010] Based on the synergistic change trend, a graph neural network is used to generate a dynamic map reflecting the interaction intensity of metabolic pathways;

[0011] The abnormal conduction chain of the metabolic pathway in the dynamic map is tracked in real time, and the risk probability matching the adolescent growth stage is calculated based on the abnormal conduction chain and chronic disease related information.

[0012] Optionally, based on the synergistic change trend, a graph neural network is used to generate a dynamic graph reflecting the interaction intensity of metabolic pathways, including:

[0013] The concentrations of various types of metabolites in the synergistic change trend are used as corresponding nodes, and the node attributes include fluctuation amplitude, change rate and period stability index;

[0014] Assigning weight values ​​between the nodes according to the change direction of the metabolite concentration and the genetic expression strength of each type;

[0015] During a preset physiological cycle, a graph neural network is used to iteratively propagate the state of the node to adjacent nodes to update the weight values ​​between the nodes;

[0016] When the updated weight value between the nodes exceeds a preset metabolic steady-state threshold, an abnormal interaction edge is generated, and the metabolic pathway containing the abnormal interaction edge is encoded into a dynamic graph.

[0017] Optionally, within a preset physiological cycle, using a graph neural network to iteratively propagate the state of the node to adjacent nodes to update the weight values ​​between the nodes includes:

[0018] Determining a state propagation direction of the node to an adjacent node according to a change direction of a metabolite concentration corresponding to the node;

[0019] According to the state propagation direction, the current metabolite concentration corresponding to the node is compared with the historical concentration average of the adjacent nodes to generate a concentration difference adjustment coefficient;

[0020] According to the concentration difference adjustment coefficient, the weight value between the node and the adjacent nodes is adjusted through a dynamic propagation mechanism.

[0021] Optionally, adjusting the weight value between the node and the adjacent nodes through a dynamic propagation mechanism according to the concentration difference adjustment coefficient includes:

[0022] Uniquely dividing the adjacent nodes into a primary association layer or a secondary association layer according to the metabolic path distance between the adjacent nodes and the starting node;

[0023] In the primary association layer, a primary weight correction amount is generated according to the product of the concentration difference adjustment coefficient and the weight value, and the primary weight correction amount is added to the weight value as a genetic risk compensation value;

[0024] In the secondary association layer, the ratio of the superimposed weight value to the metabolite concentration change rate is used as a conduction attenuation factor, and the concentration difference adjustment coefficient is proportionally reduced according to the conduction attenuation factor to generate a secondary weight correction amount;

[0025] The primary weight correction amount and the secondary weight correction amount are normalized.

[0026] Optionally, performing multimodal fusion on the dynamic concentration change data, the genetic risk value, and the historical diagnosis data according to a time axis to generate a mixed time series information set includes:

[0027] Matching the genetic risk value with the time window of the historical diagnosis data according to the time stamp of the dynamic concentration change data;

[0028] Within the same time window, characteristically correlating the metabolite fluctuation characteristics with the gene locus expression intensity in the genetic risk value;

[0029] Binding the disease development stage label in the historical diagnosis data with the association result to form a time series data unit;

[0030] The time series data units are divided according to physiological cycles, and multiple time series data units divided into the same physiological cycle are superimposed to generate a mixed time series information set.

[0031] Optionally, the adaptively optimizing the mixed time series information set and identifying the coordinated change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set includes:

[0032] Dynamically adjusting the weight ratio between metabolite concentration and genetic expression intensity in the mixed time series information set;

[0033] Based on the adjusted weight ratio, the overlap intervals between the peak and valley values ​​of metabolite concentrations and the threshold of genetic expression intensity were analyzed;

[0034] Within the overlap interval, determining a synchronous fluctuation pattern in which both the rate of change of metabolite concentration and the direction of change of genetic expression intensity satisfy a preset isotropic condition;

[0035] When the duration of the synchronous fluctuation pattern exceeds a preset physiological period and the fluctuation amplitude of the synchronous fluctuation pattern exceeds a preset baseline, it is marked as a coordinated change trend.

[0036] Optionally, the real-time tracking of the abnormal conduction chain of the metabolic pathway in the dynamic map and calculating the risk probability matching the adolescent growth stage based on the abnormal conduction chain and chronic disease related information include:

[0037] In the dynamic graph, detecting conductive paths in which the weight changes of the connecting edges between N consecutive nodes are consistent, where N is greater than or equal to three;

[0038] The conduction pathways whose weight changes exceed the steady-state range of the corresponding metabolic pathways are marked as abnormal conduction chains;

[0039] Generate a risk weight coefficient by matching preset metabolic pathway sensitivity parameters based on the chronic disease-related information and combining the starting node type and conduction direction of the abnormal conduction chain;

[0040] Based on the risk weight coefficient and combined with the metabolic rate correction factor of the adolescent growth stage, the risk probability is calculated.

[0041] In a second aspect, the present application provides a system for dynamically predicting disease risks in adolescents, comprising:

[0042] An acquisition module is used to obtain the adolescent's electronic health record and the dynamic concentration change data of the adolescent's individual subcutaneous metabolites, wherein the electronic health record includes information related to chronic diseases, genetic risk values ​​and historical diagnosis data;

[0043] A fusion module, configured to perform multimodal fusion of the dynamic concentration change data, the genetic risk value, and the historical diagnosis data according to a time axis to generate a hybrid time series information set;

[0044] an optimization module, configured to adaptively optimize the mixed time series information set and identify the coordinated change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set;

[0045] A generation module, configured to generate a dynamic graph reflecting the interaction intensity of metabolic pathways using a graph neural network based on the synergistic change trend;

[0046] The calculation module is used to track the abnormal conduction chain of the metabolic pathway in the dynamic map in real time, and calculate the risk probability matching the adolescent growth stage based on the abnormal conduction chain and chronic disease related information.

[0047] In a third aspect, the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a method for dynamically predicting disease risks in adolescents as described in any one of the first aspects.

[0048] In a fourth aspect, the present application provides a computer storage medium having computer program instructions stored thereon, which, when executed by a processor, implement a method for dynamically predicting disease risks in adolescents as described in any one of the first aspects.

[0049] In the present application, a method for dynamic prediction of disease risk in adolescents is provided, which includes: obtaining electronic health records of adolescents and dynamic concentration change data of individual subcutaneous metabolites of adolescents, wherein the electronic health records include information related to chronic diseases, genetic risk values ​​and historical diagnosis data; multimodally fusing the dynamic concentration change data, the genetic risk values ​​and the historical diagnosis data according to the time axis to generate a mixed time series information set; adaptively optimizing the mixed time series information set, and identifying the synergistic change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set; based on the synergistic change trend, using a graph neural network to generate a dynamic map reflecting the interaction intensity of metabolic pathways; tracking the abnormal conduction chain of the metabolic pathway in the dynamic map in real time, and calculating the risk probability matching the growth stage of the adolescent based on the abnormal conduction chain and information related to chronic diseases.

[0050] The technical solution provided by this application has the following beneficial effects:

[0051] This application integrates electronic health records and real-time biosensor data to establish a multi-dimensional data set covering genetic background and dynamic metabolic characteristics, providing a data basis for accurate prediction. It achieves spatiotemporal alignment of metabolite temporal fluctuations, genetic risks and medical history data, and solves the problem of collaborative analysis of multi-source heterogeneous data. It suppresses physiological noise interference, extracts the correlation fluctuation pattern between metabolites and genetic characteristics, and enhances the reliability of feature expression. It converts linear time series data into a metabolic pathway interaction network to intuitively present the cascade influence relationship of key biomarkers. Through the spatiotemporal evolution analysis of abnormal conduction chains, it outputs dynamic risk values ​​adapted to the individual's growth and development stage to achieve personalized early warning.

[0052] Furthermore, based on the synergistic trend between metabolites and genetic characteristics, the present application maps the concentrations of various metabolites and their dynamic properties into graph nodes, initializes edge weights based on the correlation between the direction of concentration change and the intensity of genetic expression, updates the weights in real time through iterative state propagation of the graph neural network, and generates a dynamic graph marking abnormal interaction edges when the weights exceed the metabolic steady-state threshold.

[0053] Moreover, it breaks through the linear analysis limitations of traditional time series models, reveals the nonlinear interaction network between metabolites through graph structure modeling, uses the weight iteration mechanism to capture the dynamic imbalance characteristics of metabolic pathways unique to adolescents, and generates abnormal interaction edges to achieve visual positioning of disease risk transmission pathways.

[0054] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0056] Figure 1 A flowchart of a method for dynamically predicting disease risks in adolescents provided in an embodiment of the present application;

[0057] Figure 2 A schematic diagram of the structure of a dynamic prediction system for adolescent disease risks provided in an embodiment of the present application;

[0058] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0060] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0061] Existing adolescent disease risk prediction schemes rely on time-series modeling using recurrent neural networks. While these approaches can process dynamic metabolite data, they suffer from two key flaws: First, using genetic risk scores as static parameter inputs simplifies the real-time interaction between metabolite fluctuations and genetic background; second, discrete risk assessments fail to reflect the continuous dynamics of metabolic pathway abnormalities, and model outputs are disconnected from the metabolic characteristics of adolescents during their growth stages, resulting in a high rate of missed detection of early, weak risk signals. This limitation stems from the inadequate ability of existing technologies to collaboratively analyze multimodal data and the lack of dynamic metabolic network modeling tools.

[0062] In response to the above-mentioned defects, the present application proposes a method for dynamic prediction of adolescent disease risk, the innovation of which lies in: constructing a mixed time series information set containing coordinated changes of metabolites and genetic characteristics through multimodal fusion, and using graph neural networks to map the time series characteristics into a dynamic map of the interaction intensity of metabolic pathways. Specifically, by real-time tracking of the spatiotemporal expansion characteristics of abnormal conduction chains in the map, combined with the growth stage metabolic correction factor, the dynamic risk probability is calculated. This method breaks through the limitations of static parameterization processing in the prior art, and not only realizes real-time coupling analysis of genetic risks and metabolic fluctuations, but also intuitively presents risk conduction paths through dynamic maps. At the same time, it introduces a growth stage adaptation mechanism, so that weak abnormal signals can be accurately captured in the early stages. It solves the three core problems of multimodal data fragmentation analysis, risk conduction path ambiguity, and neglect of growth characteristics in the prior art as a whole, and improves warning sensitivity and clinical applicability.

[0063] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0064] Figure 1 A flowchart of a method for dynamically predicting disease risks in adolescents provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0065] Step 101: Obtain the electronic health record of the adolescent and the dynamic concentration change data of the adolescent's individual subcutaneous metabolites, wherein the electronic health record includes information related to chronic diseases, genetic risk values, and historical diagnosis data.

[0066] In step 101, the electronic health record contains a structured data set of the adolescent's previous medical records. The dynamic concentration change data of subcutaneous metabolites represents the sequence data of the concentration changes of biomarkers such as blood glucose, cortisol, and inflammatory factors over time, which are continuously collected by wearable biosensors, reflecting the real-time metabolic status. Chronic disease-related information refers to the types of metabolic diseases that the adolescent has been diagnosed with or is at high risk for, including specific diagnostic information such as diabetes and obesity. The genetic risk value is a disease susceptibility score calculated based on the results of genetic testing, which is obtained by analyzing the association between variants at specific gene sites and diseases. Historical diagnostic data includes previous examination results and treatment plans recorded in outpatient records, inpatient medical records, etc.

[0067] In an embodiment of the present application, the electronic health record of the target adolescent is retrieved through the hospital information system interface, the wearable sensor data transmitted by Bluetooth is received synchronously, the genetic risk values ​​in the record are normalized, and the historical diagnostic data are encoded according to the International Classification of Diseases standards, and finally a structured data set is formed, in which the sensor data is updated at a frequency of minutes, and the electronic health record data is loaded all at once.

[0068] For example, taking a 12-year-old adolescent at high risk of obesity as an example, their electronic health records were obtained from the database of a tertiary hospital, including a family history genetic risk value of 0.72 for obesity and a BMI excess mark in previous physical examination records; at the same time, a wrist sensor was used to collect subcutaneous interstitial fluid glucose concentration data for 7 consecutive days, with a sampling interval of 5 minutes, forming a dynamic data set containing 10,080 time points.

[0069] Step 102: Perform multimodal fusion on the dynamic concentration change data, the genetic risk value, and the historical diagnosis data according to the time axis to generate a mixed time series information set.

[0070] In step 102, the hybrid time series information set represents a structured data set formed by fusing data from different sources and different sampling frequencies under a unified time reference, retaining the spatiotemporal correlation characteristics of the original data.

[0071] In an embodiment of the present application, the genetic risk value is linearly interpolated based on the acquisition timestamp of the dynamic concentration data to make it have the same time resolution, the disease stage label in the historical diagnostic data is expanded to a duration label, and the three types of data are aligned in the same time interval through a sliding time window. The window width is set to the typical metabolic cycle of adolescents, and finally a fusion data set containing the three-dimensional characteristics of time, metabolism and genetics is generated.

[0072] For example, continuing with the previous example, the minute-by-minute blood glucose data is integrated with the daily updated genetic risk score: the genetic risk value is linearly interpolated within 24 hours, the obesity diagnostic label is expanded to a marker valid throughout the day, and the window is divided according to the 4-hour metabolic cycle. In each window, a data unit containing 240 blood glucose values, 24 interpolated genetic risk values ​​and 1 disease label is generated.

[0073] Step 103: Adaptively optimize the mixed time series information set, and identify the coordinated change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set.

[0074] In step 103, metabolite fluctuation characteristics refer to the regular patterns of temporal changes in blood glucose, cortisol, and other concentrations, obtained through sliding window statistics of extreme points and rates of change. Genetic characteristics reflect the regulatory effects of gene expression on metabolites and are derived from an analysis of the gradient of genetic risk values ​​over corresponding time periods. Co-variation trends represent the coupling between metabolite concentration fluctuations and changes in genetic expression intensity in the time and frequency domains, manifesting as synchronized fluctuations with specific phase differences.

[0075] In an embodiment of the present application, wavelet denoising is performed on the mixed data to eliminate motion artifacts, the local extreme points of the metabolite concentration are extracted as feature anchor points, and the gradient of the change of the genetic risk value in the corresponding anchor point interval is calculated. When the overlap between the blood glucose peak time and the rising interval of the genetic risk gradient exceeds a threshold, it is marked as an effective synergistic period, and the final output is a trend set including the fluctuation amplitude and duration.

[0076] For example, it was detected that during the adolescent's blood sugar peak 2 hours after dinner, the genetic risk interpolation data increased by 1.8 times the baseline value, and the overlap of the two change time accounted for 85%, which was determined to be an effective synergistic period. The blood sugar fluctuation amplitude of 2.8mmol / L and the genetic risk increment of 0.15 during this period were recorded as trend characteristics.

[0077] Step 104: Based on the synergistic change trend, a graph neural network is used to generate a dynamic graph reflecting the interaction intensity of the metabolic pathway.

[0078] In step 104, a graph neural network (GNN) is a machine learning model that processes graph-structured data. In this application, it is used to model the dynamic relationships between metabolite nodes. Metabolic pathway interaction strength refers to the edge weights between nodes, which quantitatively represent the degree of metabolite interaction. A dynamic graph represents a time-varying network with metabolites as nodes and interactions between nodes as edges. Edge weights represent the regulatory strength of metabolic pathways.

[0079] In an embodiment of the present application, the concentrations of blood glucose, cortisol, and inflammatory factors are set as three types of nodes, and the node attributes include the current concentration value and the rate of change; the genetic risk value is used as the initial value of the edge weight, and messages are transmitted through the graph convolutional network within a 6-hour physiological cycle. The node status update formula is h_v^(l+1)=σ(∑_(u∈N(v))W^lh_u^l / |N(v)|+b^l). When the change in edge weight exceeds 2 times the standard deviation of the historical mean, the abnormal edge mark is triggered.

[0080] For example, a graph containing a blood glucose main node and a cortisol subnode was constructed with an initial edge weight of 0.72. After three rounds of iterative updates, the weight increased to 1.35 (exceeding the steady-state threshold of 1.2), generating a red abnormal interaction edge from blood glucose to cortisol.

[0081] Step 105: Track the abnormal conduction chain of the metabolic pathway in the dynamic map in real time, and calculate the risk probability matching the adolescent growth stage based on the abnormal conduction chain and chronic disease related information.

[0082] In step 105, an abnormal conduction chain represents a path consisting of three or more consecutive abnormal edges in the dynamic spectrum, reflecting the cascade effect of metabolic imbalance. Adolescent growth stages are divided into pre-puberty, mid-puberty, and late-puberty, based on bone age test results. Risk probability is the numerical probability of disease occurrence adjusted by weight calculation and correction factors.

[0083] In an embodiment of the present application, the connection relationship of the abnormal edges in the graph is detected. When there is a continuous abnormal path from blood glucose to cortisol and then to inflammatory factors, the total weight change of the path is extracted, multiplied by the metabolic pathway sensitivity coefficient of 0.7 corresponding to obesity, and then the 14-year-old age correction factor 1.2 is introduced. The final risk probability = 1.35×0.7×1.2=1.134, exceeding the threshold of 1.0 triggers an orange warning.

[0084] For example, the total weight of the adolescent's abnormal conduction chain is 1.35, and the risk value after correction for sensitivity coefficient and age is 1.134. The system pushes dietary intervention recommendations to the guardian's mobile phone.

[0085] This method constructs a metabolic-genetic dynamic interaction network through multi-source data fusion, realizing end-to-end calculation from raw biosensor data to disease risk level, breaking through the limitations of traditional methods on static parameter and isolated indicator analysis, improving the accuracy and timeliness of identifying early risks of chronic diseases in adolescents, and providing a quantifiable decision-making basis for personalized health management.

[0086] To address the problem of insufficient dynamic interaction modeling of metabolic pathways in existing adolescent disease risk prediction, in some embodiments, step 104: generating a dynamic graph reflecting the interaction intensity of metabolic pathways using a graph neural network based on the synergistic change trend, includes:

[0087] Step 201: The concentrations of various types of metabolites in the synergistic change trend are used as corresponding nodes, and the node attributes include fluctuation amplitude, change rate and period stability index.

[0088] In step 201, a node refers to an abstract unit of metabolite concentration data. The time series concentration values ​​of three types of metabolites, namely blood glucose, cortisol, and inflammatory factors, are mapped to independent nodes respectively. The fluctuation amplitude in the node attributes reflects the concentration range, the change rate represents the concentration change per unit time, and the periodic stability index measures the regularity of the fluctuation.

[0089] In an embodiment of the present application, concentration sequences of three types of metabolites are extracted from a mixed time series information set, the difference between the maximum and minimum values ​​in each sampling window is calculated as the fluctuation amplitude, the rate of change per minute is calculated using the differential method, and the amplitude ratio of the main frequency components is analyzed by Fourier transform to obtain the periodic stability index, and finally a node set containing three types of attribute values ​​is generated.

[0090] Step 202: assign weight values ​​between the nodes according to the change direction of the metabolite concentration and the genetic expression strength of each type.

[0091] In step 202, genetic expression strength is a key parameter directly related to metabolite concentration. It is defined as a quantified value of the ability of a specific gene locus to regulate metabolite concentration. Its physical meaning is the intensity of the impact of genetic risk factors on metabolic fluctuations. It is obtained by analyzing the correlation between genetic testing data in electronic health records and real-time metabolite concentrations. The weight value represents the strength of the metabolic association between nodes. The initial weight is set based on the matching degree between the direction of metabolite concentration change and the genetic expression strength. A high weight is assigned when the two directions increase, and a low weight is assigned when the two directions change in opposite directions.

[0092] In the embodiment of the present application, the signs of the slopes of the concentration changes of metabolites corresponding to adjacent nodes and the directions of changes in the genetic expression intensity are compared. When both blood glucose and cortisol concentrations increase and the genetic risk score increases synchronously, the weight between the two nodes is initialized to a higher value; if the concentration of inflammatory factors decreases and the genetic score increases, the corresponding weight is initialized to a lower value.

[0093] Step 203: Within a preset physiological cycle, a graph neural network is used to iteratively propagate the state of the node to adjacent nodes to update the weight values ​​between the nodes.

[0094] In step 203, state propagation refers to the process of transmitting node attribute information along the connection edge. The neighborhood aggregation mechanism of the graph neural network is adopted. The node's fluctuation feature vector and the adjacent node features are weighted and summed in each physiological cycle and then updated.

[0095] In the embodiment of the present application, 24 hours is set as the physiological cycle, and iterative calculation is started every morning: each blood glucose node merges its fluctuation amplitude and change rate attributes with the corresponding attributes of the connected cortisol node according to the weight value ratio, and the updated cortisol node continues to propagate the status to the inflammatory factor node, and the cycle continues until all nodes complete the status update.

[0096] Step 204: When the updated weight value between the nodes exceeds a preset metabolic steady-state threshold, an abnormal interaction edge is generated, and the metabolic pathway containing the abnormal interaction edge is encoded into a dynamic graph.

[0097] In step 204, the term "exceeding" specifically refers to triggering the generation of abnormal interaction edges when the updated weight of any node pair in the dynamic graph exceeds a preset metabolic homeostasis threshold, rather than requiring that all node weights exceed this threshold. This design addresses the clinical characteristic that local abnormalities in adolescent metabolic pathways precede global imbalances, ensuring early warning sensitivity. The metabolic homeostasis threshold is the upper limit of fluctuation in the interaction strength of metabolic pathways in adolescents under normal physiological conditions. It is statistically determined by analyzing the correlation between changes in metabolite concentrations and genetic expression in historical data from healthy adolescent populations. When the weights between nodes in the dynamic graph exceed this threshold, it indicates a disruption in metabolic balance and a predisposition to disease risk transmission. This threshold is set for different metabolic pathway types (e.g., glucose metabolism pathway, lipid metabolism pathway) and different growth stages (e.g., pre-pubertal and mid-adolescent), reflecting the critical point of metabolic regulation capacity during specific developmental periods in adolescents. Abnormal interaction edges are graphical elements that mark potential risk transmission pathways and are triggered when the incremental weight of a connecting edge exceeds the upper limit of the historical fluctuation of that metabolic pathway.

[0098] In an embodiment of the present application, the weight changes between the blood glucose and cortisol nodes are monitored. When the cumulative weight increase in three consecutive physiological cycles exceeds 1.5 times the previous maximum fluctuation range of the path, a red abnormal edge is generated and the path is marked as a high-risk conduction pathway.

[0099] Here's a specific example:

[0100] Taking a 12-year-old adolescent at high risk for obesity as an example, to generate a dynamic graph, the coordinated trend data for blood glucose and cortisol were first converted into graph nodes. The blood glucose node attributes included a fluctuation amplitude of 2.8 mmol / L (derived from the difference between the peak and baseline values ​​2 hours after dinner), a rate of change of 0.023 mmol / L / min (calculated from the minute-by-minute concentration change), and a cyclic stability index of 0.85 (calculated from the consistency of fluctuations over the same period in the 7-day data). The cortisol node attributes were derived from simultaneously collected saliva data. Based on the upward trend in both blood glucose and cortisol concentrations and the simultaneous increase in genetic expression, the weight between the two nodes was initialized to 0.72, which is consistent with the genetic risk value in the electronic health record. Over a 24-hour physiological cycle, three rounds of state propagation were performed via a graph neural network. The first round propagated the fluctuation characteristics of the blood glucose node to the cortisol node according to the weight ratio, increasing the edge weight to 0.93. The second round adjusted the blood glucose node output based on the updated cortisol node state, increasing the edge weight to 1.14. The third round, incorporating new data collected in real time, finally updated the edge weight to 1.35. This value exceeds the preset metabolic homeostasis threshold of 1.2. This threshold is derived from historical data statistics of the same metabolic pathway in healthy adolescents of the same age group. The system then generates a red abnormal interaction edge from blood glucose to cortisol and marks the pathway as a high-risk conduction pathway.

[0101] In the embodiments of the present application, by constructing a dynamic graph of metabolite interactions, a visual presentation of the risk transmission pathway from discrete data to a network is achieved, making the metabolic imbalance process in a specific growth stage of adolescents traceable and quantifiable, providing an intuitive basis for early and precise intervention.

[0102] To address the problem of lack of directionality and dynamic regulation in the metabolite state propagation process in the prior art, in some embodiments, step 203: iteratively propagating the node state to adjacent nodes using a graph neural network within a preset physiological cycle to update the weight values ​​between the nodes, includes:

[0103] Step 301: Determine the state propagation direction of the node toward the adjacent node based on the change direction of the metabolite concentration corresponding to the node.

[0104] In step 301, the state propagation direction refers to the transmission path of the metabolite concentration change trend in the node network. When the blood glucose concentration rises, the direction pointing to the cortisol node is defined as positive propagation, and vice versa as negative propagation.

[0105] In an embodiment of the present application, the concentration change slope of each metabolite node in the current time window is analyzed. When the slope is positive, the propagation direction is set to the downstream metabolic node. For example, when blood sugar increases, it positively affects the cortisol node. When the slope is negative, the reverse propagation path is set.

[0106] Step 302: According to the state propagation direction, the current metabolite concentration corresponding to the node is compared with the historical concentration average of the adjacent nodes to generate a concentration difference adjustment coefficient.

[0107] In step 302, the current metabolite concentration refers to the instantaneous detection value of blood glucose, cortisol, or inflammatory factors within the most recent time window (e.g., the last 5 minutes) collected in real time by the biosensor, reflecting the most recent changes in the adolescent's metabolic state. The historical concentration mean refers to the average concentration of the same metabolite node over several complete physiological cycles (e.g., the same period over the past three days). This is calculated by querying historical data stored in the mixed time series information set and is used to establish an individualized metabolic baseline level. This mean is the arithmetic mean after excluding abnormal fluctuations, with one mean data point corresponding to each physiological cycle. The concentration difference adjustment coefficient reflects the degree of deviation of the current metabolic state from the historical baseline and is calculated by comparing the current concentration value with the mean of the past three physiological cycles of the adjacent node.

[0108] In an embodiment of the present application, the detected concentration value of the cortisol node at the current moment is obtained, and the difference is made from the average concentration value of the node in the same period of the previous three days. The difference is divided by the previous fluctuation range to obtain a standardized adjustment coefficient, which is used to quantify the degree of current metabolic abnormality.

[0109] Step 303: According to the concentration difference adjustment coefficient, the weight value between the node and the adjacent nodes is adjusted through a dynamic propagation mechanism.

[0110] In step 303, the dynamic propagation mechanism refers to a method of adaptively adjusting the weight according to the real-time metabolic state, using the product of the adjustment coefficient and the current weight as an increment, and adding it to the original weight to achieve dynamic update.

[0111] In an embodiment of the present application, when the adjustment coefficient of the blood glucose to cortisol pathway shows the current concentration difference, the coefficient is multiplied by the existing weight to obtain a weight increment, so that important metabolic abnormality signals can quickly affect network parameters.

[0112] Here's a specific example:

[0113] For example, during the state propagation process, a 12-year-old adolescent at high risk for obesity is first determined based on the direction of change in the current blood glucose concentration at the node. A positive slope is detected, indicating a post-breakfast blood glucose concentration increase of 0.018 mmol / L per minute. Therefore, the propagation is set to proceed in a positive direction toward the cortisol node. The difference between the current blood glucose value of 8.2 mmol / L and the average of the three previous post-breakfast concentrations of 7.5 mmol / L is calculated. This average is obtained by taking the arithmetic mean of the blood glucose data from the 7:00-8:00 AM period over the past three days. A concentration difference adjustment factor of 0.93 is generated, calculated by subtracting the average from the current value and dividing it by the average fluctuation range of 1.2 mmol / L. Based on this coefficient, a dynamic propagation mechanism adjusts the weight. The current weight of 0.72 is multiplied by the adjustment factor of 0.93, resulting in an increment of 0.67. This increment is then added to the original weight, bringing the updated weight to 1.39. This value exceeds the preset metabolic homeostasis threshold of 1.2, and the system immediately flags the propagation path as abnormal. After lunch, an abnormal increase in blood glucose concentration was detected again, and the above process was repeated for a second round of weight adjustment, ultimately raising the weight of the pathway to 1.52, strengthening the risk assessment of the abnormal conduction chain.

[0114] In the embodiment of the present application, through the directed state propagation and dynamic weight adjustment mechanism, the metabolic abnormality signal can be accurately diffused along the physiological association path, avoiding irrelevant interference and ensuring that important risk characteristics are captured in a timely manner, thereby improving the efficiency of identifying the metabolic imbalance risk of adolescents.

[0115] To address the issue of unclear hierarchical associations during dynamic adjustment of metabolic pathway interaction strength, in some embodiments, step 303: adjusting the weights between the node and adjacent nodes through a dynamic propagation mechanism based on the concentration difference adjustment coefficient, includes:

[0116] Step 401: Uniquely divide the adjacent nodes into a primary association layer or a secondary association layer according to the metabolic path distance between the adjacent nodes and the starting node.

[0117] In step 401, the metabolic pathway distance refers to the number of shortest connecting edges between metabolite nodes. The primary association layer contains nodes directly connected to the starting node. The secondary association layer contains nodes that are connected to the primary association layer nodes but not directly connected.

[0118] In the embodiment of the present application, the blood glucose node is used as the starting node, the directly connected cortisol node is classified into the primary association layer, and the inflammatory factor node connected downstream of the cortisol node is classified into the secondary association layer, ensuring that each adjacent node belongs to only a single level.

[0119] Step 402: In the primary association layer, a primary weight correction amount is generated according to the product of the concentration difference adjustment coefficient and the weight value, and the primary weight correction amount is added to the weight value as a genetic risk compensation value.

[0120] In step 402, the primary weight modifier reflects the magnitude of the weight adjustment of the directly associated node. The genetic risk compensation value is a supplementary adjustment based on the impact of genetic background on metabolism.

[0121] In an embodiment of the present application, the concentration difference adjustment coefficient between the blood glucose node and the cortisol node is multiplied by the current weight value to obtain a primary correction amount, and then the compensation value obtained based on the genetic risk database is superimposed to complete the weight update of the primary association layer.

[0122] Step 403: In the secondary association layer, the ratio of the superimposed weight value to the metabolite concentration change rate is used as the conduction attenuation factor, and the concentration difference adjustment coefficient is proportionally reduced according to the conduction attenuation factor to generate a secondary weight correction amount.

[0123] In step 403, the conduction attenuation factor is used to quantify the degree to which the metabolic influence is attenuated in the secondary pathway. This is achieved through the dynamic ratio of the weight value to the metabolite change rate. The secondary weight correction refers to the weight adjustment value for the indirectly connected nodes (secondary connection layer). It is calculated by combining the updated weight value of the primary connection layer with the metabolite concentration change rate. It reflects the adjustment magnitude of the metabolic influence after attenuation during the conduction process. This correction is generated by multiplying the concentration difference adjustment coefficient by the conduction attenuation factor. It quantifies the intensity loss of the metabolic signal when it is transmitted from the intermediate node to the secondary node, ensuring that the weight update conforms to the physiological conduction characteristics of the metabolic pathway.

[0124] In an embodiment of the present application, the updated weight value of the primary association layer is divided by the current metabolite concentration change rate of the secondary node plus one to prevent division by zero, and the adjustment coefficient is proportionally reduced after obtaining the attenuation factor to generate a secondary correction amount that takes conduction loss into account.

[0125] Step 404: normalize the primary weight correction amount and the secondary weight correction amount.

[0126] In step 404 , the normalization process ensures that the total weight on the same metabolic pathway does not exceed the physiological load limit of the pathway.

[0127] In the embodiment of the present application, after the primary and secondary correction values ​​are summed, the sum is compared with the preset metabolic load upper limit, and the smaller one is taken as the final weighted output value.

[0128] Here's a specific example:

[0129] Taking a 12-year-old adolescent at high risk for obesity as an example, during the dynamic weight adjustment process, adjacent nodes are first divided according to metabolic pathway distance: the cortisol node directly connected to the blood glucose node is placed in the primary association layer, and the inflammatory factor node connected to the cortisol node is placed in the secondary association layer. In the primary association layer, the current blood glucose concentration difference adjustment coefficient of 0.93 is taken. This value is calculated by subtracting the historical mean of 7.5 mmol / L from the current blood glucose value of 8.2 mmol / L and dividing it by the fluctuation range of 1.2 mmol / L. This value is multiplied by the current weight of 0.72 to obtain a primary weight correction of 0.67. The genetic risk compensation value of 0.15, which is derived from the obesity genetic risk score in the electronic health record, is added to update the primary weight to 0.82. In the secondary association layer, the updated primary layer weight of 0.82 is divided by the current change rate of the inflammatory factor node, 0.005 mmol / L / min, plus 1 to prevent division by zero. This yields a conduction attenuation factor of 0.82. The original adjustment coefficient of 0.93 is proportionally reduced to 0.77 by this factor, generating a secondary weight correction of 0.60. Finally, the sum of the primary correction of 0.67 and the secondary correction of 0.60 yields 1.27. After normalization, this value does not exceed the preset metabolic load upper limit of 1.5, ultimately assigning the conduction pathway a weight of 1.27.

[0130] In the embodiment of the present application, through the hierarchical dynamic adjustment mechanism, the rapid response capability of directly related metabolites is retained, and the conduction attenuation characteristics of indirect associations are taken into account, so that the weight update process is more in line with the actual metabolic conduction laws of adolescents and the physiological rationality of risk prediction is improved.

[0131] To address the issue of inaccurate spatiotemporal alignment during fusion of multi-source heterogeneous data, in some embodiments, step 102: performing multimodal fusion of the dynamic concentration change data, the genetic risk value, and the historical diagnostic data along a time axis to generate a hybrid time series information set includes:

[0132] Step 501: Match the genetic risk value with the time window of the historical diagnosis data according to the time stamp of the dynamic concentration change data.

[0133] In step 501, the time stamp refers to the timestamp when the biosensor collects metabolite data. Gene locus expression intensity is a specific representation of genetic expression intensity. Both refer to the quantified risk value of a specific gene locus in the electronic health record, with the former specifying that the intensity data originates from the gene locus test results. The time window is the data alignment interval divided according to physiological characteristics.

[0134] In an embodiment of the present application, based on the time point collected by the blood glucose sensor every minute, the daily genetic risk score data is linearly interpolated hourly, and the historical diagnosis data is expanded into a continuous period according to the date of consultation, ensuring that the three types of data are aligned within the same physiological activity period.

[0135] Step 502: Within the same time window, feature association is performed between the metabolite fluctuation characteristics and the gene locus expression intensity in the genetic risk value.

[0136] In step 502, the gene locus expression intensity is a quantitative value obtained by typing the target adolescent's specific disease-related gene locus (such as the rs9939609 locus of the FTO gene) through genetic testing technology, and matching the variation type of the locus with a database of known associations with disease risks. A larger value indicates a higher genetic risk.

[0137] In an embodiment of the present application, the peak and trough values ​​of the blood glucose data in each time window are extracted, and a Pearson correlation analysis is performed with the change gradient of the genetic risk score in the corresponding time period to generate a feature pair reflecting the strength of the metabolic-genetic association.

[0138] Step 503: Bind the disease development stage label in the historical diagnosis data with the association result to form a time series data unit.

[0139] In step 503, the disease stage label is a standardized code for the disease progression status. The label is assigned by clinicians based on a comprehensive assessment of the patient's physical examination indicators (such as BMI percentile), laboratory test results (such as fasting blood glucose level), and clinical symptoms (such as the severity of acanthosis nigricans). The label is encoded using internationally recognized disease staging standards (such as the WHO obesity classification). The time series data unit is the smallest analytical unit after fusion, and contains metabolite concentrations, genetic expression strengths, and disease labels.

[0140] In the embodiment of the present application, the obesity diagnostic marker is coded into levels 1-3 according to severity and is bound to the aforementioned metabolic-genetic feature pair to form a complete data unit including a timestamp, metabolic features, genetic features and disease labels.

[0141] Step 504: Divide the time series data units according to the physiological cycle, and superimpose multiple time series data units divided into the same physiological cycle to generate a mixed time series information set.

[0142] In step 504, the physiological cycle refers to the metabolic rhythm cycle unique to adolescents. This division results in the classification of time series data units according to adolescent circadian rhythms (e.g., sleep period, activity period, eating period) or developmental stages (e.g., pre-puberty, mid-puberty, post-puberty). This is used in subsequent processes to determine the data aggregation range when overlaying cross-period features and serves as the time reference for dynamic graph generation. The overlay operation enhances the extraction of cyclical features.

[0143] In an embodiment of the present application, the data units are divided into sleep period, activity period and diet period according to the circadian rhythm, and the metabolic feature mean, genetic feature maximum and disease label of the units in the same time period for multiple consecutive days are logically ORed to generate fusion data with periodic representativeness. The specific implementation process is as follows: the time series data units are divided into three physiological stages of sleep period, activity period and diet period according to the adolescent circadian rhythm, and the fluctuation range of blood glucose concentration, cortisol secretion peak and inflammatory factor baseline level are statistically analyzed in each stage; the data of the same physiological stage for 7 consecutive days are vertically superimposed, the synchronous fluctuation pattern of metabolite concentration and genetic risk value is retained, and the isolated fluctuation noise is eliminated to form the mixed time series information set in weeks. Specific embodiment: Taking adolescents aged 12-14 years old at high risk of obesity as an example: in the sleep period, the blood glucose fluctuation data from 23:00 to 6:00 every night are superimposed. When the blood glucose trough value monitored for 3 consecutive weeks exceeds the predicted range of the genetic risk value, the abnormal pattern is associated with the obesity family history label in the electronic health record to generate a mixed time series information set containing time-metabolism-genetic three-dimensional features.

[0144] Here's a specific example:

[0145] Taking a 12-year-old adolescent at high risk for obesity as an example, during the multimodal fusion process, the daily genetic risk value of 0.72 was first decomposed into 1440 minute-level data points using the timestamp of minute-by-minute blood glucose data as the baseline through linear interpolation. The calculation formula was: the minute-by-minute risk value equals the daily risk value divided by 1440. Simultaneously, the excessive BMI label was expanded to a full-day label. Within the 4-hour eating cycle, 240 blood glucose values ​​were extracted within the window from 6:00 PM to 10:00 PM after dinner. The calculated average concentration was 7.8 mmol / L, with a peak of 8.9 mmol / L. The interpolated mean genetic risk for this period was 0.75, a 4.2% increase from the daily baseline of 0.72. The Pearson correlation coefficient (covariance divided by the product of the respective standard deviations) yielded a correlation of 0.82. The blood glucose fluctuation range of 2.1 mmol / L within this window was calculated as the peak minus trough value, a genetic risk change of 0.03, and a disease label with two levels of binding to form a time series data unit. The data of the same period for 7 consecutive days were analyzed, and the average blood glucose fluctuation value was 2.3mmol / L, the genetic risk peak value was 0.78, and the disease label was maintained at level 2 to generate a mixed time series information unit representing the dinner period.

[0146] In the examples of the present application, through precise time series alignment and multi-dimensional feature fusion, a comprehensive data set that can simultaneously reflect metabolic dynamics, genetic background and disease progression is constructed, providing a comprehensive and temporally and spatially consistent data foundation for subsequent risk prediction.

[0147] In order to solve the problems of signal interference and feature mismatch in the coordinated analysis of metabolic and genetic features, in some embodiments, step 103: adaptively optimizing the mixed time series information set and identifying the coordinated change trend of metabolite fluctuation features and genetic features from the optimized mixed time series information set includes:

[0148] Step 601: Dynamically adjust the weight ratio between the metabolite concentration and the genetic expression intensity in the mixed time series information set.

[0149] In step 601, the weight ratio refers to the relative importance distribution of metabolite concentration data and genetic expression intensity data in the analysis, and is dynamically adjusted according to the metabolite type and genetic risk level.

[0150] In the embodiments of the present application, the weight ratio of concentration data is increased for key metabolites such as blood glucose and high genetic risk scores; for minor metabolites and low risk scores, the weight of genetic data is increased, and the contribution of the two types of data is adjusted in real time through a sliding window mechanism.

[0151] Step 602: Based on the adjusted weight ratio, the overlap interval between the metabolite concentration peak and valley values ​​and the genetic expression intensity threshold is analyzed.

[0152] In step 602, metabolite concentration peaks and valleys refer to the highest (peak) and lowest (valley) concentrations of blood glucose, cortisol, and inflammatory factors within a preset time period. These extreme values ​​are directly extracted from the raw concentration data continuously collected by the wearable biosensor through time series analysis. The genetic expression intensity threshold refers to the critical value of the expression level of a gene locus, corresponding to the preset risk level of the gene locus in the electronic health record. When the actual detected gene expression level exceeds this critical value, a genetic risk warning is triggered. This threshold is derived from a benchmark database established by large-scale adolescent population genetics studies. The overlap interval refers to the period when abnormalities occur simultaneously between the extreme metabolite concentration and the genetic expression intensity threshold, reflecting the temporal correlation between the two.

[0153] In the embodiment of the present application, the peak-valley value period exceeding the physiological reference range in the metabolite concentration curve is detected, and the intersection with the period when the genetic expression intensity exceeds the preset risk threshold is calculated to obtain the time interval with metabolic-genetic dual abnormality characteristics.

[0154] Step 603: Within the overlap interval, determine a synchronous fluctuation pattern in which both the rate of change of the metabolite concentration and the direction of change of the genetic expression intensity satisfy a preset isotropic condition.

[0155] In step 603, the metabolite concentration change rate refers to the magnitude of change in blood glucose, cortisol, or inflammatory factor concentration per unit time. This is calculated by dividing the concentration difference between two consecutive sampling time points by the time interval. Data is derived from the real-time monitoring data stream of the wearable biosensor, and the minute-by-minute rate of change is calculated using a sliding time window method. The direction of change in genetic expression intensity refers to the trend of change (increase / decrease / stabilization) in the expression level of a specific gene locus relative to the baseline level. This is determined by comparing the mean gene expression level of the current monitoring cycle with the mean gene expression level of historical cycles. Data is derived from regularly updated genetic testing reports in the electronic health record, and a three-period moving average method is used to eliminate measurement fluctuations. The homotropy condition requires that the direction of change in metabolite concentration and the direction of change in genetic expression intensity are consistent, including both increasing and decreasing. Specifically, homotropy is considered when the metabolite concentration change rate is positive (increasing concentration) and the direction of change in genetic expression intensity is increasing, or when the metabolite concentration change rate is negative (decreasing concentration) and the direction of change in genetic expression intensity is decreasing. Conversely, if the rate and direction are inversely related (e.g., increasing concentration but decreasing expression), the results are considered inconsistent. The synchronous fluctuation pattern refers to the state of correlation in which the rate of change of metabolite concentration and the direction of change of genetic expression intensity remain consistent within the same period of time, including fluctuations in which both are simultaneously enhanced or weakened, reflecting the real-time synergistic relationship between metabolic activity and genetic regulation.

[0156] In the examples of the present application, the slope of the metabolite concentration change and the gradient of the genetic expression intensity change are calculated. When the signs of the two are the same, it is determined that the isotropic condition is met, and the fluctuation characteristics of the period are retained for subsequent analysis.

[0157] Step 604: When the duration of the synchronous fluctuation pattern exceeds a preset physiological period and the fluctuation amplitude of the synchronous fluctuation pattern exceeds a preset baseline, it is marked as a coordinated change trend.

[0158] In step 604, duration refers to the length of the continuous period that satisfies the isotropic condition. It is calculated from the start and end time points when the changes in the marker metabolite and genetic expression intensities are aligned. To be valid, this duration must exceed the typical physiological cycle of an adolescent (e.g., a complete sleep cycle). Fluctuation amplitude refers to the range of metabolite concentration within the period of synchronous fluctuation. It is calculated by calculating the difference between the peak and trough metabolite concentrations within this period. This value is compared and evaluated with the baseline fluctuation range of healthy peers in the same state. The preset baseline is derived from the statistical value of metabolite fluctuations in a healthy peer group.

[0159] In the embodiment of the present application, the synchronous fluctuation pattern is required to last for at least one complete sleep-wake cycle, and the fluctuation amplitude exceeds two standard deviations of the average level of healthy adolescents of the same age group, in order to be marked as a valid synergistic trend.

[0160] Here's a specific example:

[0161] For example, in the identification of synergistic trends in a 12-year-old adolescent at high risk for obesity, a weight of 0.7 was assigned to blood glucose data and 0.3 to identify genetic risk data based on their importance. During the post-dinner period from 6:00 PM to 8:00 PM, a peak blood glucose level of 8.9 mmol / L was detected, exceeding the upper limit of normal (7.8 mmol / L). During this period, the interpolated genetic risk value rose from 0.72 to 0.78, exceeding the threshold of 0.75. This overlap was determined to be 108 minutes. Within this interval, the rate of change in blood glucose was calculated to be 0.015 mmol / L per minute, and the genetic risk increased by 0.03 per hour, both in a positive direction. This synchronized fluctuation pattern persisted for 108 minutes, exceeding the predefined 90-minute circadian cycle. The blood glucose fluctuation amplitude of 2.3 mmol / L, calculated by subtracting the trough value of 6.6 mmol / L from the peak value of 8.9 mmol / L, exceeded the baseline post-prandial fluctuation amplitude of 1.5 mmol / L for healthy peers, based on the 95th percentile of the healthy adolescent population. The system ultimately marked this period as a synergistic change trend, where the genetic risk increment of 0.06 was calculated by subtracting the initial value of 0.72 from the end value of the period of 0.78.

[0162] In the embodiments of the present application, dynamic weight adjustment and multi-dimensional feature matching are used to effectively capture the true correlation pattern between metabolic abnormalities and genetic risks, eliminate the interference of accidental fluctuations, and provide a reliable basis for synergistic change characteristics for disease risk warning.

[0163] To address the issue of inaccurate identification of risk pathways in dynamic maps, in some embodiments, step 105: real-time tracking of abnormal conduction chains in metabolic pathways in the dynamic map, and calculating risk probabilities matching adolescent growth stages based on the abnormal conduction chains and chronic disease-related information, includes:

[0164] Step 701: In the dynamic graph, detect conductive paths in which the weight changes of the connecting edges between N consecutive nodes are consistent, where N is greater than or equal to three.

[0165] In step 701, the conduction path specifically refers to a specific weight change trajectory consisting of N consecutive nodes and connecting edges in the dynamic graph, while the metabolic pathway is a physiologically established biochemical reaction pathway (such as the insulin signaling pathway), and the conduction path is the dynamic instantiation representation of the metabolic pathway in the graph.

[0166] In an embodiment of the present application, all possible node connection combinations in the dynamic graph are traversed, and paths with three or more nodes in which the direction of change of the connection edge weights is either increasing or decreasing are screened out as a candidate set of potential abnormal conduction paths.

[0167] Step 702: Mark the conduction pathway whose weight change amplitude exceeds the steady-state range of the corresponding metabolic pathway as an abnormal conduction chain.

[0168] In step 702, the steady-state range refers to the normal fluctuation range of weight changes in the corresponding metabolic pathway of healthy adolescents, which is obtained through long-term monitoring data statistics.

[0169] In the embodiment of the present application, the total weight change of the candidate conduction path is compared with the historical health data of the same path. When it exceeds the mean of the healthy group plus two times the standard deviation, it is determined to be an abnormal conduction chain.

[0170] Step 703: Matching preset metabolic pathway sensitivity parameters according to the chronic disease related information, combining the starting node type and conduction direction of the abnormal conduction chain, and generating a risk weight coefficient.

[0171] In step 703, the metabolic pathway sensitivity parameter reflects the sensitivity of a specific disease to abnormalities in that pathway and is preset based on the disease type and pathway category. The starting node type refers to the metabolite category of the first abnormal node in the conduction chain (e.g., the blood glucose node), determined by analyzing the initial starting point of the weighted abnormal edge in the dynamic graph. The conduction direction represents the trend of weight change (positive for an increase, negative for a decrease), derived from the incremental sign calculation of the connecting edge weights. The risk weight coefficient quantifies the risk of an abnormal conduction chain and is composed of the product of the pathway sensitivity coefficient, the starting node type weight, and the conduction direction coefficient. A larger value indicates a higher risk.

[0172] In an embodiment of the present application, the disease-pathway association database is queried to obtain the sensitivity coefficient of obesity to the blood glucose-cortisol pathway, and the risk weight coefficient is calculated based on the characteristics that the starting node of the conduction chain is a blood glucose node and the direction is positive conduction. The specific process is as follows: a mapping relationship table of chronic disease-related information and metabolic pathway sensitivity parameters is pre-established, wherein the sensitivity parameter of the insulin resistance pathway corresponding to diabetes is 0.7, and the sensitivity parameter of the lipid metabolism pathway corresponding to obesity is 0.5; when the starting node type of the abnormal conduction chain is a blood glucose node and the conduction direction is positive, the weight change amplitude of each node on the conduction path is summed and multiplied by the sensitivity parameter of the corresponding disease, wherein the weight change amplitude is calculated by subtracting the absolute value of the difference between the current node weight and the previous node weight; finally, the calculated result is multiplied by the metabolic rate correction factor corresponding to the current growth stage of the adolescent, which is determined according to the bone age test results and is uniformly taken as 1.2 for adolescents aged 12-14 years old, to generate the final risk weight coefficient. Specific embodiment: Taking a 15-year-old adolescent at high risk of diabetes as an example, when an abnormal conduction chain starting from the blood glucose node, passing through the cortisol node and conducting to the inflammatory factor node appears in the dynamic graph, the weight changes of the three nodes are measured to be 0.3, 0.2, and 0.4, respectively; first, the sum is obtained to obtain 0.9, which is multiplied by the diabetes sensitivity parameter 0.7 to obtain 0.63, and then multiplied by the metabolic rate correction factor 1.2 for this age group, and finally a risk weight coefficient of 0.756 is generated; this coefficient and the duration of the abnormal conduction chain are jointly input into the risk probability calculation formula, where the duration exceeds 8 hours, a secondary warning is triggered.

[0173] Step 704: Calculate the risk probability based on the risk weight coefficient and the metabolic rate correction factor of the adolescent growth stage.

[0174] In step 704, the metabolic rate correction factor is a growth stage adjustment parameter determined based on the bone age test results. This metabolic rate correction factor represents a baseline model established by collecting basal metabolic rate data from healthy adolescents of different age groups. A baseline value of 1.0 is used for those aged 12-14 years, 1.2 for those aged 15-17 years due to accelerated metabolism during puberty, and 0.9 for those aged 18 and above due to metabolic stabilization. The specific values ​​are derived from the resting energy expenditure percentile curves for each age group published in the Youth Health Database, normalized by the 50th percentile value.

[0175] In an embodiment of the present application, the risk weight coefficient is multiplied by the metabolic activity factor corresponding to the current growth stage, and the risk probability value taking into account the individual development characteristics is finally output. For example, the risk weight coefficient is first multiplied by the metabolic rate correction factor to obtain the basic risk value. When the basic risk value exceeds 0.5, the time decay compensation calculation is started, and the formula risk probability = basic risk value × [1 + 0.1 × duration of abnormal conduction chain (hours) / 24] is used; taking the 14-year-old obesity monitoring as an example, the risk weight coefficient is measured to be 0.6, and the age correction factor 1.0 is matched to obtain a basic risk value of 0.6. When the abnormal conduction chain lasts for 12 hours, the final risk probability = 0.6 × 1 + 0.1 × 12 / 24 = 0.65. When the preset threshold value of 0.6 is exceeded, a first-level warning is triggered.

[0176] Here's a specific example:

[0177] Taking a 12-year-old adolescent at high risk for obesity as an example, during the risk probability calculation process, the system first detected a pathway with three consecutive nodes in the dynamic graph: a blood glucose node, a cortisol node, and then a inflammatory factor node. The weight changes of all three connecting edges were positive. The total weight change of this pathway was calculated to be 0.63, obtained by adding the weight increments of 0.21, 0.18, and 0.24 for each edge. This value exceeded the upper limit of the steady-state range of 0.5 for this metabolic pathway, which was determined by the 95th percentile of the weight changes for the same pathway in a healthy adolescent population. The system labeled this pathway as an abnormal pathway, with the starting node being a blood glucose node and the transmission direction being positive. A query of the disease-pathway association database revealed that the sensitivity coefficient of obesity to the blood glucose-cortisol-inflammatory factor pathway was 0.7. Combined with the blood glucose node type weight of 1.0 and the positive transmission coefficient of 1.2, the risk weight coefficient was calculated as 0.84 using the formula: 0.7 multiplied by 1.0 multiplied by 1.2. Combined with the metabolic rate correction factor of 1.0 for 12-year-old adolescents, the final risk probability is equal to 0.84 multiplied by 1.0, which is 0.84. It exceeds the warning threshold of 0.8 and triggers a second-level warning.

[0178] In the embodiments of the present application, a dynamic quantitative assessment of the risk of chronic diseases in adolescents is achieved through a growth stage adaptation algorithm with multi-dimensional feature fusion. This algorithm not only takes into account the conduction pathway characteristics of metabolic abnormalities, but also combines the physiological characteristics of the individual development stage, thereby improving the accuracy and clinical applicability of the early warning signal.

[0179] Figure 2 This is a schematic diagram of the structure of a dynamic prediction system for adolescent disease risk provided by an embodiment of the present application, such as Figure 2 As shown, the system includes:

[0180] The acquisition module 21 is used to obtain the electronic health records of adolescents and the dynamic concentration change data of individual subcutaneous metabolites of adolescents. The electronic health records include information related to chronic diseases, genetic risk values ​​and historical diagnosis data.

[0181] The fusion module 22 is used to perform multimodal fusion on the dynamic concentration change data, the genetic risk value and the historical diagnosis data according to the time axis to generate a mixed time series information set.

[0182] The optimization module 23 is used to perform adaptive optimization on the mixed time series information set, and identify the coordinated change trend of the metabolite fluctuation characteristics and the genetic characteristics from the optimized mixed time series information set.

[0183] The generation module 24 is used to generate a dynamic graph reflecting the interaction intensity of the metabolic pathway based on the collaborative change trend using a graph neural network.

[0184] The calculation module 25 is used to track the abnormal conduction chain of the metabolic pathway in the dynamic map in real time, and calculate the risk probability matching the adolescent growth stage based on the abnormal conduction chain and chronic disease related information.

[0185] Figure 2 The dynamic prediction system for adolescent disease risk can be implemented Figure 1 The implementation principle and technical effects of the method for dynamically predicting disease risk in adolescents described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the system for dynamically predicting disease risk in adolescents in the above embodiment has been described in detail in the embodiments of the method and will not be elaborated on here.

[0186] In one possible design, Figure 2 A dynamic prediction system for adolescent disease risk in the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0187] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0188] The processing component 32 is used to perform the above Figure 1 The embodiment provides a method for dynamically predicting disease risks in adolescents.

[0189] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0190] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0191] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0192] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0193] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0194] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0195] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A method for dynamically predicting disease risks in adolescents according to the illustrated embodiment.

[0196] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0197] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0198] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0199] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for dynamic prediction of adolescent disease risk, characterized by: include: Obtaining electronic health records of adolescents and dynamic concentration change data of individual subcutaneous metabolites of adolescents, wherein the electronic health records include information related to chronic diseases, genetic risk values, and historical diagnosis data; Performing multimodal fusion on the dynamic concentration change data, the genetic risk value, and the historical diagnosis data according to a time axis to generate a mixed time series information set; Adaptively optimizing the mixed time series information set, and identifying the coordinated change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set; Based on the synergistic change trend, a graph neural network is used to generate a dynamic map reflecting the interaction intensity of metabolic pathways; The abnormal conduction chain of the metabolic pathway in the dynamic map is tracked in real time, and the risk probability matching the adolescent growth stage is calculated based on the abnormal conduction chain and chronic disease related information.

2. The method according to claim 1, characterized in that Based on the synergistic change trend, a graph neural network is used to generate a dynamic graph reflecting the interaction intensity of the metabolic pathway, including: The concentrations of various types of metabolites in the synergistic change trend are used as corresponding nodes, and the node attributes include fluctuation amplitude, change rate and period stability index; Assigning weight values ​​between the nodes according to the change direction of the metabolite concentration and the genetic expression strength of each type; During a preset physiological cycle, a graph neural network is used to iteratively propagate the state of the node to adjacent nodes to update the weight values ​​between the nodes; When the updated weight value between the nodes exceeds a preset metabolic steady-state threshold, an abnormal interaction edge is generated, and the metabolic pathway containing the abnormal interaction edge is encoded into a dynamic graph.

3. The method according to claim 2, characterized in that The method of iteratively propagating the state of the node to adjacent nodes using a graph neural network within a preset physiological cycle to update the weight values ​​between the nodes includes: Determining a state propagation direction of the node to an adjacent node according to a change direction of a metabolite concentration corresponding to the node; According to the state propagation direction, the current metabolite concentration corresponding to the node is compared with the historical concentration average of the adjacent nodes to generate a concentration difference adjustment coefficient; According to the concentration difference adjustment coefficient, the weight value between the node and the adjacent nodes is adjusted through a dynamic propagation mechanism.

4. The method according to claim 3, characterized in that The step of adjusting the weight value between the node and the adjacent nodes through a dynamic propagation mechanism according to the concentration difference adjustment coefficient includes: Uniquely dividing the adjacent nodes into a primary association layer or a secondary association layer according to the metabolic path distance between the adjacent nodes and the starting node; In the primary association layer, a primary weight correction amount is generated according to the product of the concentration difference adjustment coefficient and the weight value, and the primary weight correction amount is added to the weight value as a genetic risk compensation value; In the secondary association layer, the ratio of the superimposed weight value to the metabolite concentration change rate is used as a conduction attenuation factor, and the concentration difference adjustment coefficient is proportionally reduced according to the conduction attenuation factor to generate a secondary weight correction amount; The primary weight correction amount and the secondary weight correction amount are normalized.

5. The method according to claim 1, characterized in that The multimodal fusion of the dynamic concentration change data, the genetic risk value, and the historical diagnosis data according to the time axis to generate a mixed time series information set includes: Matching the genetic risk value with the time window of the historical diagnosis data according to the time stamp of the dynamic concentration change data; Within the same time window, characteristically correlating the metabolite fluctuation characteristics with the gene locus expression intensity in the genetic risk value; Binding the disease development stage label in the historical diagnosis data with the association result to form a time series data unit; The time series data units are divided according to physiological cycles, and multiple time series data units divided into the same physiological cycle are superimposed to generate a mixed time series information set.

6. The method according to claim 1, characterized in that Adaptively optimizing the mixed time series information set and identifying the coordinated change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set includes: Dynamically adjusting the weight ratio between metabolite concentration and genetic expression intensity in the mixed time series information set; Based on the adjusted weight ratio, the overlap intervals between the peak and valley values ​​of metabolite concentrations and the threshold of genetic expression intensity were analyzed; Within the overlap interval, determining a synchronous fluctuation pattern in which both the rate of change of metabolite concentration and the direction of change of genetic expression intensity satisfy a preset isotropic condition; When the duration of the synchronous fluctuation pattern exceeds a preset physiological period and the fluctuation amplitude of the synchronous fluctuation pattern exceeds a preset baseline, it is marked as a coordinated change trend.

7. The method according to claim 1, characterized in that The real-time tracking of the abnormal conduction chain of the metabolic pathway in the dynamic map and the calculation of the risk probability matching the adolescent growth stage based on the abnormal conduction chain and chronic disease related information include: In the dynamic graph, detecting conductive paths in which the weight changes of the connecting edges between N consecutive nodes are consistent, where N is greater than or equal to three; The conduction pathways whose weight changes exceed the steady-state range of the corresponding metabolic pathways are marked as abnormal conduction chains; Generate a risk weight coefficient by matching preset metabolic pathway sensitivity parameters based on the chronic disease-related information and combining the starting node type and conduction direction of the abnormal conduction chain; Based on the risk weight coefficient and combined with the metabolic rate correction factor of the adolescent growth stage, the risk probability is calculated.

8. A dynamic prediction system for adolescent disease risk, characterized by: include: An acquisition module is used to obtain the adolescent's electronic health record and the dynamic concentration change data of the adolescent's individual subcutaneous metabolites, wherein the electronic health record includes information related to chronic diseases, genetic risk values ​​and historical diagnosis data; A fusion module, configured to perform multimodal fusion of the dynamic concentration change data, the genetic risk value, and the historical diagnosis data according to a time axis to generate a hybrid time series information set; an optimization module, configured to adaptively optimize the mixed time series information set and identify the coordinated change trend of metabolite fluctuation characteristics and genetic characteristics from the optimized mixed time series information set; A generation module, configured to generate a dynamic graph reflecting the interaction intensity of metabolic pathways using a graph neural network based on the synergistic change trend; The calculation module is used to track the abnormal conduction chain of the metabolic pathway in the dynamic map in real time, and calculate the risk probability matching the adolescent growth stage based on the abnormal conduction chain and chronic disease related information.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a dynamic prediction method for adolescent disease risk as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, a method for dynamically predicting disease risks in adolescents as described in any one of claims 1 to 7 is implemented.

Images