Intelligent diagnosis system and method for clinical microbial infection based on multi-omics data fusion

Abstract

The present application relates to the technical field of intelligent medical diagnosis, in particular to a clinical microorganism infection intelligent diagnosis system and method based on multi-omics data fusion, comprising: collecting and processing time-series multi-omics data to generate a snapshot vector of instantaneous infection state; constructing a time-series state transition model, inputting the snapshot vector of instantaneous infection state into the time-series state transition model, predicting and generating a future state vector; based on the future state vector, respectively calculating a pathogen threat index and a host disorder index; based on the pathogen threat index and the host disorder index, generating a comprehensive risk score, and determining an early warning level according to a preset early warning grading threshold; based on the early warning level, and combining the pathogen threat index and the host disorder index, generating a graded intervention strategy; outputting the graded intervention strategy as specific diagnosis and treatment suggestions; the present application realizes prospective early warning, effectively avoiding the delay of the best treatment opportunity due to waiting for index changes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical diagnosis, in particular to a clinical microorganism infection intelligent diagnosis system and method based on multi-omics data fusion. BACKGROUND

[0002] In clinical medical practice, accurate diagnosis of microorganism infection is crucial. Current clinical infection diagnosis methods mainly rely on single-dimensional biomarkers. These data often have significant lag, and this static evaluation method cannot fully reflect the complex dynamic interaction process between the host and the pathogen.

[0003] This dependence on lagging and one-sided data often results in delayed or inaccurate clinical interventions, which may miss the best treatment opportunity. Therefore, how to dynamically and prospectively evaluate the risk of the infection process to achieve early warning and precise intervention is a technical problem that needs to be solved in the field of clinical microorganism infection diagnosis. SUMMARY

[0004] To solve the above technical problems, the present application provides a clinical microorganism infection intelligent diagnosis system and method based on multi-omics data fusion. Specifically, the technical solution of the present application is as follows:

[0005] The clinical microorganism infection intelligent diagnosis method based on multi-omics data fusion comprises the following steps:

[0006] S1, collecting and processing time-series multi-omics data to generate a snapshot vector of instantaneous infection state;

[0007] S2, constructing a time-series state transition model, inputting the snapshot vector of instantaneous infection state into the time-series state transition model, and predicting and generating a future state vector;

[0008] S3, based on the future state vector, calculating a pathogen threat index and a host disorder index, respectively;

[0009] S4, based on the pathogen threat index and the host disorder index, generating a comprehensive risk score, and determining an early warning level according to a pre-set early warning grading threshold;

[0010] S5, based on the early warning level, and combining the pathogen threat index and the host disorder index, generating a graded intervention strategy;

[0011] S6, outputting the graded intervention strategy as specific diagnosis and treatment suggestions.

[0012] Preferably, S1 collecting and processing time-series multi-omics data specifically comprises:

[0013] S11, continuously collecting pathogen omics data, host response data and clinical phenotype data of the patient at multiple key time points to form time-series multi-omics data.

[0014] Preferably, the S1 generates the instantaneous infection state snapshot vector specifically includes:

[0015] S12, convert the time series multi-omics data into numerical feature vectors;

[0016] S13, at any key time point, splice the corresponding numerical feature vector to generate the instantaneous infection state snapshot vector corresponding to the key time point.

[0017] Preferably, S2 specifically includes:

[0018] S21, taking the initial two consecutive time point instantaneous infection state snapshot vectors as the starting point of the time series state transition model calculation;

[0019] S22, according to the instantaneous infection state snapshot vectors of the current time and the historical time, the future state vector is inferred and generated through the time series state transition model.

[0020] Preferably, S3 calculates the pathogen threat index specifically includes:

[0021] S31, extract the predicted expression level of drug resistance gene and the predicted expression level of virulence factor from the future state vector;

[0022] S32, based on the preset risk weight, the predicted expression level of drug resistance gene and the predicted expression level of virulence factor are weighted and summed to obtain the pathogen threat index.

[0023] Preferably, S3 calculates the host disorder index specifically includes:

[0024] S33, extract the predicted activation level of pro-inflammatory pathway and the predicted activation level of anti-inflammatory pathway from the future state vector;

[0025] S34, based on the baseline steady state level obtained from the healthy population database, the relative change amount of the predicted activation level of pro-inflammatory pathway and the predicted activation level of anti-inflammatory pathway is calculated and weighted and summed to obtain the host disorder index.

[0026] Preferably, S4 specifically includes:

[0027] S41, using the modified logistic function, the pathogen threat index and the host disorder index are normalized respectively;

[0028] S42, according to the preset weight coefficient, the normalized pathogen threat index and the host disorder index are weighted and fused to generate the comprehensive risk score;

[0029] S43, compare the comprehensive risk score with a preset early warning grading threshold, and determine the early warning level as one of a stable level, a first-level early warning and a second-level early warning.

[0030] Preferably, S5 specifically comprises:

[0031] S51, if the early warning level is the stable level, a suggestion of maintaining the current diagnosis and treatment scheme is generated;

[0032] S52, if the early warning level is the first-level early warning, and it is determined that the early warning is driven by the pathogen threat index, a suggestion of adjusting the antibiotic scheme is generated;

[0033] S53, if the early warning level is the first-level early warning, and it is determined that the early warning is driven by the host imbalance index, a suggestion of closely monitoring the immune markers is generated;

[0034] S54, if the early warning level is the second-level early warning, an alarm of immediately taking comprehensive intervention measures is generated.

[0035] The clinical microorganism infection intelligent diagnosis system based on multi-omics data fusion comprises the following modules:

[0036] A data preprocessing module is configured to collect and process time-series multi-omics data to generate a snapshot vector of instantaneous infection state;

[0037] A trajectory prediction module is configured to input the snapshot vector of instantaneous infection state to predict and generate a future state vector;

[0038] A risk quantification module is configured to calculate a pathogen threat index and a host imbalance index based on the future state vector;

[0039] An early warning grading module is configured to generate a comprehensive risk score and determine an early warning level based on the pathogen threat index and the host imbalance index;

[0040] A decision generation module is configured to generate a graded intervention strategy as a specific diagnosis and treatment suggestion based on the early warning level and in combination with the pathogen threat index and the host imbalance index.

[0041] Compared with the prior art, the present application has the following beneficial effects:

[0042] 1. The present application realizes forward-looking early warning, wins a time window for clinical intervention, and can dynamically predict the future infection state of the patient based on historical and current multi-dimensional data by constructing a time-series state transition model.

[0043] 2. The present application provides comprehensive evaluation, deep insight into the infection process, by integrating the omics data of the pathogen, the host response data and the clinical phenotype data, a high-dimensional infection state snapshot is constructed, this multi-dimensional data fusion overcomes the limitations of traditional methods relying on isolated and one-sided biomarkers, provides a comprehensive and solid data foundation for deep understanding of the complex host-pathogen interaction, and improves the accuracy of diagnosis;

[0044] 3. The present application realizes the precise quantification of risk, and traces the core driving factors. It not only provides a general risk score, but also calculates the pathogen threat index and the host disorder index based on the predicted future state. This two-dimensional risk decomposition can accurately distinguish whether the risk is dominated by the virulence and drug resistance of the pathogen or by the immune response disorder of the host, thereby tracing the core driving factors of the risk and providing deep pathophysiological insights for individualized treatment;

[0045] 4. The present application generates individualized strategies with strong decision-making assistance capabilities. It can generate differentiated intervention strategies based on the risk level and the risk source of the pathogen and host in two dimensions, for example, the same medium risk may prompt adjustment of antibiotics if it is driven by the pathogen, and monitoring of immune status if it is driven by host disorder. This same-part different strategy decision-making capability directly translates complex model output into clear and executable clinical recommendations, greatly improving the precision and individualization level of treatment. BRIEF DESCRIPTION OF DRAWINGS

[0046] The present application will be further explained in conjunction with the accompanying drawings and examples:

[0047] Figure 1 is a flowchart of the method of the present application;

[0048] Figure 2 is a structural diagram of the system of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail in conjunction with specific examples.

[0050] Example 1:

[0051] Please refer to Figure 1 , the multi-omics data fusion clinical microbial infection intelligent diagnosis method comprises the following steps:

[0052] S1, collecting and processing time-series multi-omics data to generate an instantaneous infection state snapshot vector;

[0053] S2, constructing a time-series state transition model, inputting the instantaneous infection state snapshot vector into the time-series state transition model, predicting and generating a future state vector;

[0054] S3, respectively calculate the pathogen threat index and the host disorder index based on the future state vector;

[0055] S4, generate a comprehensive risk score based on the pathogen threat index and the host disorder index, and determine an early warning level according to a preset early warning grading threshold;

[0056] S5, generate a graded intervention strategy based on the early warning level, in combination with the pathogen threat index and the host disorder index;

[0057] S6, output the graded intervention strategy as specific diagnosis and treatment suggestions;

[0058] Please refer to Figure 2 , the intelligent diagnosis system for clinical microbial infection based on multi-omics data fusion includes the following modules:

[0059] The data preprocessing module is used to collect and process time series multi-omics data to generate an instantaneous infection state snapshot vector;

[0060] The trajectory prediction module is used to input the instantaneous infection state snapshot vector to predict and generate a future state vector;

[0061] The risk quantification module is used to calculate the pathogen threat index and the host disorder index based on the future state vector;

[0062] The early warning grading module is used to generate a comprehensive risk score based on the pathogen threat index and the host disorder index, and determine an early warning level;

[0063] The decision generation module is used to generate a graded intervention strategy as specific diagnosis and treatment suggestions based on the early warning level, in combination with the pathogen threat index and the host disorder index;

[0064] The embodiment of the present application provides a multi-omics data fusion intelligent diagnosis method and system for clinical microbial infection; the method aims to solve the technical problems that the existing clinical infection diagnosis relies on lagging and single-dimensional data, resulting in untimely or inaccurate intervention measures; the embodiment realizes dynamic and forward-looking risk assessment and intelligent auxiliary decision-making of the infection process by constructing a complete technical closed loop from time series data collection to graded intervention decision-making; the implementation subject of the method can be an intelligent diagnosis system deployed on a hospital server or a cloud platform;

[0065] The intelligent diagnosis system, which corresponds to the method steps one by one, includes a data preprocessing module, a trajectory prediction module, a risk quantification module, an early warning grading module, and a decision generation module; the method process will be described in detail in combination with the system modules;

[0066] S1, collecting and processing time-series multi-omics data to generate a snapshot vector of instantaneous infection state;

[0067] This step is performed by a data preprocessing module, which is a collection of software functions responsible for collecting raw data from the clinical environment, cleaning, standardizing, and converting it into structured data that can be calculated by subsequent models. Its purpose is to provide high-quality, high-dimensional time-series input for the entire diagnostic process.

[0068] In this embodiment, the data preprocessing module collects data within a specific time window after the patient enters the intensive care unit (ICU) at multiple key time points, such as: Continuous collection of multi-source heterogeneous data of patients to form time-series multi-omics data; the specific data dimensions include at least:

[0069] Pathogenomics data: through metagenomic sequencing technology, obtain gene sequence fragments that can reflect the species, abundance, drug resistance genes, and virulence factor expression of pathogenic microorganisms;

[0070] Host response data: through transcriptome or proteomics technology, obtain the expression levels of genes or proteins that can reflect the state of the host immune system, such as cytokines, chemokines, etc.

[0071] Clinical phenotype data: including vital signs such as body temperature, heart rate, and inflammation markers such as C-reactive protein, procalcitonin, and other standardized indicators widely used in clinical practice;

[0072] In addition, the data preprocessing module also has built-in data quality control and outlier processing mechanisms. For some missing data at key time points, the module can use time series interpolation or regression prediction based on multi-modal data to fill in the missing data to ensure the integrity of the snapshot vector of instantaneous infection state The module will mark or smooth the extreme values that are beyond the reasonable range of clinical practice, thereby enhancing the robustness of the entire diagnostic system to imperfect data in the real world;

[0073] The module converts these multi-modal raw data into a unified numerical feature vector; this process is a conventional data processing procedure known to those skilled in the art, such as log transformation of gene expression data and normalization of clinical indicators;

[0074] At any key time point t, the data preprocessing module concatenates the pathogenomics, host response, and clinical phenotype numerical feature vectors corresponding to that time point to generate a snapshot vector of instantaneous infection state The instantaneous infection state snapshot vector St refers to a high-dimensional numerical vector. Its special meaning in the specific technical environment of this invention is that it is not an isolated set of data points, but a complete and quantitative description of the host-pathogen interaction system at a specific moment. As a whole, this vector provides a comprehensive state basis for subsequent dynamic prediction.

[0075] S2. Construct a temporal state transition model. Input the instantaneous infection state snapshot vector into the temporal state transition model to predict and generate future state vectors.

[0076] This step is performed by the trajectory prediction module; the trajectory prediction module is a core algorithm module whose purpose is to infer the state of the system at a future point in time based on the historical and current states, realizing a paradigm shift from passively waiting for data to actively predicting the future.

[0077] In this embodiment, the trajectory prediction module constructs a temporal state transition model; the temporal state transition model refers to a mathematical model used to describe how the system state evolves over time; its design concept draws on the state-space model in control theory, especially the core idea of ​​state prediction in Kalman filtering, and has been adaptively improved for complex biomedical scenarios;

[0078] The starting point for the model's calculations is the initial two consecutive time points. and Instantaneous infection state snapshot vector and As input; The initial position of the system is defined, and and The difference provides the initial dynamic trend of the system;

[0079] Based on the instantaneous infection state snapshot vectors at the current time t and the historical time t-1, the future state vector is inferred and generated through a temporal state transition model. This prediction process does not rely on actual data collection at time t+1, but is purely based on mathematical inference of historical evolution trends. Its core state transition prediction equation is as follows:

[0080]

[0081] in, The predicted future state vector at time 1, a high-dimensional numerical vector, is calculated using this formula;

[0082] Current moment A snapshot vector of the instantaneous infection state;

[0083] : historical moment snapshot vector of instantaneous infection state;

[0084] : system intrinsic state transition matrix, which refers to a weight matrix describing the next moment state evolution trend caused by the interaction between the characteristics of the system; its role is to capture the inherent and stable change rule of the infection state;

[0085] : dynamic change influence matrix, which refers to a weight matrix for quantifying the influence of the current momentum of the system, i.e. the speed of state change, on the future state; its role is to capture the short-term and accelerated change trend of the infection state;

[0086] : system random disturbance term, which refers to a vector representing random biological noise or measurement error that is not captured by the model;

[0087] matrix and as adjustable parameters of the model, the internal weight values are not preset fixed values, but are obtained by deep learning training on a time series multi-omics database containing a large number of historical sepsis patients; specifically, a neural network, such as a multilayer perceptron containing several hidden layers, can be constructed with the spliced vector of the historical state vector and as input, with the goal of minimizing the mean square error loss function between the model predicted state and the real subsequent state in the calibration data set; the number of output layer neurons of the model is the same as the total number of all elements in the matrix and , so as to directly generate the weight values of the two matrices; for clear distinction, the historical database here constitutes the calibration data set, and the internal variables are independent of the variables collected during model operation; optimization algorithms such as back propagation and gradient descent are used to iteratively optimize the weights with the goal of minimizing the mean square error loss function between the model predicted state and the real subsequent state in the calibration data set, until the model converges;

[0088] It should be noted that the linear state transition equation in the present embodiment is a simplification adopted to achieve robustness and timeliness of calculation; although the and matrices obtained by deep learning training can capture a large number of nonlinear relationships, the prediction form of the model itself is linear, and in some extremely complex nonlinear pathological processes, its prediction accuracy may be limited. In future improvements, nonlinear terms or more complex model structures such as recurrent neural networks (RNN) can be introduced to further improve the fidelity of the prediction; ​

[0089] S3, calculating the pathogen threat index and the host dysregulation index respectively based on the future state vector;

[0090] This step is performed by the risk quantification module; the risk quantification module refers to an analytical module aiming to reduce the dimensionality of the high-dimensional, complex future state vector generated by the trajectory prediction module, and transform it into risk indicators that are clinically interpretable and have clear biological significance;

[0091] In this embodiment, the risk quantification module calculates the pathogen threat index and the host dysregulation index based on the predicted future state vector For the sake of clarity, hereinafter the predicted state , output by the model, is referred to as the future state

[0092] Pathogen threat index The pathogen threat index refers to a comprehensive score for quantifying the clinical risk directly caused by the pathogen itself; its technical consideration lies in integrating various pathogenic factors of the pathogen, including drug resistance and virulence, into a single indicator to evaluate its potential harm; the calculation method is as follows:

[0093]

[0094] wherein, is the predicted expression level of the th drug resistance gene, a dimensionless value, directly extracted from the corresponding component of the future state vector

[0095] is the risk weight of the th drug resistance gene, a dimensionless scalar, which is derived from the Cox proportional hazards model regression analysis of an independent, large-scale clinical historical database calibration dataset; the hazard ratio statistical value of the genes significantly associated with negative prognosis of patients in the calibration dataset is used as the basis for setting the weight;

[0096] is the predicted expression level of the th virulence factor, a dimensionless value, directly extracted from the corresponding component of the future state vector

[0097] is the risk weight of the th virulence factor, a dimensionless scalar, which is derived in a similar manner as

[0098] is the total number of drug resistance genes;

[0099] ​​​​​​​the total number of virulence factors;

[0100] Host Dysregulation Index : The Host Dysregulation Index refers to a composite score used to assess the intensity and abnormality of the host immune system response; the technical consideration is that the harm of infection not only comes from the pathogen, but also from the inappropriate immune response of the host, this index aims to quantify this endogenous risk; the calculation method is as follows:

[0101]

[0102] wherein, : the predicted activation level of the pro-inflammatory pathway, a dimensionless value, is derived from the predicted expression levels of a pre-defined set of biological features associated with the pro-inflammatory response in the predicted state vector , the position index of these features in the state vector is fixed according to prior biological knowledge; this composite value is specifically obtained by weighted summation of the pre-set weights on the set of features , that is, , the weights are not set empirically, but are determined according to their statistical importance in the historical calibration data set, for example, the first principal component load value corresponding to each feature can be normalized to reflect the relative contribution of each feature in the core pro-inflammatory pattern after principal component analysis PCA is performed on the set of pro-inflammatory features;

[0103] : the baseline steady-state level of the pro-inflammatory pathway, a dimensionless value, is derived from the reference average value statistically obtained from an independent large-scale healthy population or patient stable phase database;

[0104] : the predicted activation level of the anti-inflammatory pathway, a dimensionless value, is derived from the composite value of a set of features representing the anti-inflammatory response in the predicted state vector ;

[0105] : the baseline steady-state level of the anti-inflammatory pathway, a dimensionless value, is derived in a similar manner as ;

[0106] : the dysregulation weight of the two types of pathways, a dimensionless scalar, is derived by performing machine learning classification tasks such as predicting whether multiple organ failure occurs on the historical sepsis patient database calibration data set, determining the contribution of each type of immune pathway to the final clinical outcome through feature importance analysis, and assigning weights accordingly;

[0107] S4. Based on the pathogen threat index and host dysregulation index, generate a comprehensive risk score and determine the warning level according to the preset warning level threshold.

[0108] This step is performed by the early warning classification module; the early warning classification module is a functional module that integrates risks and determines levels. Its purpose is to combine the risk indices of the two dimensions into a single, intuitive comprehensive risk score, and give a clear early warning level accordingly.

[0109] To eliminate and To address the impact of differences in the numerical ranges between exponents, the module employs a modified logistic function to normalize them, mapping them to a unified, probabilistic representation. Within the interval, the normalization operator is as follows:

[0110]

[0111] in, The original exponent value to be normalized or The dimensionless value is calculated by the risk quantification module.

[0112] : Gain coefficient that controls the steepness of the curve, a dimensionless scalar, whose value is determined based on the statistical distribution characteristics of each index in the historical database calibration dataset;

[0113] : The reference center value when the index is in a critical risk state. It is a dimensionless scalar and its value can be set as the statistical boundary point of the index that can distinguish between mild and severe patients in the historical database calibration dataset.

[0114] Normalized exponent and The two indices, calculated separately by the function, are then weighted and fused by the module to obtain the final comprehensive risk score. :

[0115]

[0116] in, : The weighting coefficients of pathogen and host risk, a dimensionless scalar, with the following constraints. Its value is determined based on clinical expert knowledge or through data-driven methods on a calibrated dataset in a historical database, reflecting the relative importance of the two types of risks in a specific type of infection;

[0117] Comprehensive risk score determining the early warning level by comparing with preset early warning grading thresholds; the early warning grading thresholds refer to critical values for dividing different risk levels; the setting logic is based on the receiver operating characteristic (ROC) curve analysis of the score distribution of all patients in the historical database calibration data set, and the statistical cut-off point that can best distinguish different clinical outcomes such as survival and organ failure rate is selected;

[0118] In this embodiment, the early warning level is determined to be one of the following three:

[0119] stable level: For example, ;

[0120] primary early warning: For example, ;

[0121] secondary early warning: ;

[0122] S5, based on the early warning level, and combined with the pathogen threat index and the host imbalance index, a graded intervention strategy is generated;

[0123] S6, output the graded intervention strategy as specific diagnosis and treatment suggestions;

[0124] These two steps are performed by the decision generation module; the decision generation module refers to the logic judgment and information generation module that converts the numerical output of the model into clinically executable and personalized suggestions; the purpose is to complete the closed loop from risk prediction to intervention decision, and to convert the prediction ability of the model into actual clinical action;

[0125] Based on the early warning level, and further analyzing the relative contribution of the core risk sources driving the early warning, i.e. and The following graded intervention strategy is generated:

[0126] If the early warning level is stable, the system generates a suggestion to maintain the current diagnosis and treatment plan to avoid excessive medical treatment;

[0127] If the early warning level is primary early warning, and it is determined that the early warning is mainly driven by the pathogen threat index For example, If the proportion in exceeds the preset threshold, the system generates a suggestion to adjust the antibiotic regimen, such as prompting that the current medication may have a drug resistance risk;

[0128] If the early warning level is primary early warning, and it is determined that the early warning is mainly driven by the host imbalance index , the system generates a suggestion to closely monitor immune markers, prompting the doctor to pay attention to whether the host has excessive inflammatory response;​

[0129] If the early warning level is level two, the system will issue an alarm of the highest priority regardless of the driving factor, generate an alarm for immediate comprehensive intervention, such as suggesting hemodynamic support, immune regulation treatment, etc.

[0130] The generated hierarchical intervention strategies will be pushed to medical staff in the form of specific diagnosis and treatment suggestions through a clinical information system interface;

[0131] Predictive and timeliness: by constructing a time series state transition model, the invention changes the diagnosis paradigm from static assessment based on current lag indicators to dynamic early warning based on future state prediction, which wins a valuable time window for clinical intervention;

[0132] Comprehensiveness and multidimensionality: by integrating pathogenomics, host response and clinical phenotype data, a comprehensive snapshot of the instantaneous infection state is constructed, and risk quantification is performed from both pathogen and host dimensions; this overcomes the limitations of traditional methods relying on a single biomarker, and provides a deeper understanding of the complex infection process;

[0133] Precision and individualization: the invention not only gives a comprehensive risk level, but also traces the main driving factor of risk, pathogen threat or host disorder, to generate differentiated hierarchical intervention strategies; this ability of isomerization makes the diagnosis and treatment suggestions more precise and individualized, and improves the treatment effect;

[0134] Closed loop and operability: the invention constructs a complete intelligent diagnosis closed loop from data input, trajectory prediction, risk quantification, early warning grading to decision suggestion, directly converts complex model output into clear and executable clinical suggestions, greatly improving the clinical application value and operability of the technology.

[0135] Embodiment 2:

[0136] S1 collects and processes time series multi-omics data, specifically including:

[0137] S11, continuously collect pathogenomics data, host response data and clinical phenotype data of the patient at multiple key time points to form time series multi-omics data;

[0138] S1 generates an instantaneous infection state snapshot vector, specifically including:

[0139] S12, convert the time series multi-omics data into a numerical feature vector;

[0140] S13, at any key time point, splice the corresponding numerical feature vector to generate an instantaneous infection state snapshot vector corresponding to the key time point.

[0141] This embodiment is the embodiment and optimization of step S1 performed by the data preprocessing module; based on the previous embodiment, this embodiment further clarifies the collection of time series multi-omics data and the generation process of the instantaneous infection state snapshot vector;

[0142] Specifically, S11, continuously collect the pathogenomics data, host response data and clinical phenotype data of the patient at multiple key time points to form time series multi-omics data; this step emphasizes two key attributes of data collection: time series and multi-source; time series refers to continuous sampling on the key time axis of infection evolution to capture the dynamic change process of the disease; multi-source refers to simultaneously obtaining information reflecting three different dimensions of pathogen, host and clinical manifestation, ensuring the comprehensiveness of subsequent analysis;

[0143] S12, convert the time series multi-omics data into a numerical feature vector; this step is the process of standardizing and quantifying the heterogeneous raw data collected, ensuring that data of different sources and types can be calculated in the same mathematical model;

[0144] S13, at any key time point, splice the corresponding numerical feature vector to generate an instantaneous infection state snapshot vector corresponding to the key time point; this step is the key to building a high-dimensional state space; by splicing all dimensional features at the same time into a long vector, a mathematical object that can completely describe the panorama of host-pathogen interaction at that moment is constructed;

[0145] Compared with only generally proposing to collect and process data, this embodiment defines the time sequence attribute of data collection, the diversity of data sources, and the specific technical means of constructing the instantaneous state vector through splicing, ensuring the high quality and high information density of the model input; such structured and comprehensive state vector provides a solid data foundation for the accurate prediction of the subsequent time series state transition model, and improves the accuracy and reliability of the entire diagnostic system from the source.

[0146] Embodiment 3:

[0147] S2 specifically includes:

[0148] S21, take the instantaneous infection state snapshot vectors of the initial two consecutive time points as the starting point for calculation of the time series state transition model;

[0149] S22, according to the instantaneous infection state snapshot vectors of the current time and the historical time, generate the future state vector by the time series state transition model.

[0150] This embodiment is the embodiment of step S2 performed by the trajectory prediction module; it further clarifies the specific calculation logic of the time series state transition model for prediction;

[0151] As described above, the core of the model is the state transition prediction equation; this embodiment further defines the calculation starting point and iteration mode of the model:

[0152] S21, taking the initial two consecutive time point instantaneous infection state snapshot vectors and as the calculation starting point of the time series state transition model; the technical consideration of this design is that the state of a single time point is isolated, while the state of two consecutive time points contains the initial change rate or trend information, i.e. ; taking the initial trend as the input of the model makes the model better capture the dynamic characteristics of the system from the beginning;

[0153] S22, according to the instantaneous infection state snapshot vectors of the current time and the historical time , the future state vector is inferred and generated through the time series state transition model; this clearly defines the iteration prediction mechanism of the model; the model not only considers the inherent evolution of the current state through the term, but also explicitly takes the current change rate as an important basis for predicting the future state through the term;

[0154] This embodiment makes the trajectory prediction module not only a simple state extrapolator, but also a prediction engine that can understand and utilize the system momentum; by explicitly including the state change amount in the model, the model's ability to capture nonlinear change trends such as infection acceleration and deterioration or improvement is greatly enhanced; this makes the prediction result more accurate, especially in the critical period when the disease changes dramatically, thereby improving the sensitivity and accuracy of the warning.

[0155] Embodiment 4:

[0156] S3, calculating the pathogen threat index specifically includes:

[0157] S31, extracting the predicted expression level of drug-resistant genes and the predicted expression level of virulence factors from the future state vector;

[0158] S32, based on the preset risk weight, the predicted expression level of drug-resistant genes and the predicted expression level of virulence factors are weighted and summed to obtain the pathogen threat index;

[0159] S3, calculating the host disorder index specifically includes:

[0160] S33, extracting the predicted activation level of pro-inflammatory pathways and the predicted activation level of anti-inflammatory pathways from the future state vector;

[0161] S34, based on the baseline steady-state level obtained from the healthy population database, the relative change amount of the predicted activation level of the pro-inflammatory pathway and the predicted activation level of the anti-inflammatory pathway is calculated, and weighted sum is performed to obtain the host imbalance index.

[0162] This embodiment is a further refinement of step S3 performed by the risk quantification module, which respectively describes the accurate calculation process of the pathogen threat index and the host imbalance index.

[0163] The calculation of the pathogen threat index:

[0164] S31, the predicted expression level of drug-resistant genes and the predicted expression level of virulence factors are extracted from the future state vector ; this step clearly shows that the data source for index calculation is the future state predicted by the model, which embodies the foresight;

[0165] S32, based on the preset risk weight and , the extracted predicted expression levels are weighted and summed to obtain the pathogen threat index ; in this way, the risks of different genes are quantitatively integrated;

[0166] The calculation of the host imbalance index:

[0167] S33, the predicted activation level of the pro-inflammatory pathway and the predicted activation level of the anti-inflammatory pathway are extracted from the future state vector ;

[0168] S34, based on the baseline steady-state level obtained from the healthy population database , the relative change amount of the predicted activation level is calculated, and weighted sum is performed to obtain the host imbalance index ; the core of this calculation method is the quantification of the degree of deviation, that is, the degree of deviation of the host immune response from the normal steady state;

[0169] This embodiment produces a synergistic gain effect by defining the risk indexes of the pathogen and the host respectively; it enables the diagnostic system to distinguish between two completely different clinical deterioration scenarios: one is driven by high virulence and strong drug-resistant pathogens; the other is caused by excessive immune system response of the host itself; this differentiation ability is not possessed by traditional single indicator diagnosis methods, which directly leads to the precision of subsequent intervention strategies, and provides profound pathophysiological insights for individualized treatment;​​

[0170] Embodiment 5:

[0171] S4 specifically comprises:

[0172] S41, using a modified logistic function, respectively, the pathogen threat index and host disorder index are normalized;

[0173] S42, according to the preset weight coefficient, the normalized pathogen threat index and host disorder index are weighted and fused to generate a comprehensive risk score;

[0174] S43, the comprehensive risk score is compared with the preset warning classification threshold, and the warning level is determined as one of stable level, first level warning and second level warning.

[0175] This embodiment is a specific implementation of step S4 performed by the warning classification module, which details the entire process from the two-dimensional risk index to the single, standardized warning level;

[0176] S41, using a modified logistic function, respectively, the pathogen threat index and host disorder index are normalized; the purpose of this step is to solve the problem of different dimensions and large numerical range of the two original indexes, and to map them into the probability space, enhancing the stability and interpretability of subsequent fusion calculation;

[0177] S42, according to the preset weight coefficient , the normalized pathogen threat index and host disorder index are weighted and fused to generate a comprehensive risk score ; this step integrates multi-dimensional risk information into a single, easy-to-understand score, facilitating rapid clinical decision-making;

[0178] S43, the comprehensive risk score is compared with the preset warning classification threshold , and the warning level is determined as one of stable level, first level warning and second level warning; this step finally converts a continuous risk score into a discrete, clear action-oriented warning level.

[0179] This embodiment realizes the standardization and interpretability of risk assessment by introducing a series of processes such as normalization, weighted fusion and threshold classification; it converts complex multi-dimensional original risk information into a comprehensive score with probability significance and a clear warning level that clinicians can intuitively understand and trust; this greatly reduces the complexity and uncertainty of clinical decision-making, making the output of the model seamlessly integrated into the existing clinical workflow.

[0180] Embodiment 6:

[0181] S5 specifically includes:

[0182] S51, if the early warning level is stable level, generating a suggestion of maintaining the current diagnosis and treatment scheme;

[0183] S52, if the early warning level is level one early warning, and it is determined that the early warning is driven by pathogen threat index, generating a suggestion of adjusting the antibiotic scheme;

[0184] S53, if the early warning level is level one early warning, and it is determined that the early warning is driven by host disorder index, generating a suggestion of closely monitoring immune markers;

[0185] S54, if the early warning level is level two early warning, generating an alarm of immediately taking comprehensive intervention measures.

[0186] This embodiment is a specific implementation of step S5 performed by the decision generation module, which directly associates the early warning level with specific and differentiated intervention strategies, forming a decision-making loop;

[0187] S51, if the early warning level is stable level, generating a suggestion of maintaining the current diagnosis and treatment scheme; this strategy aims to avoid unnecessary medical intervention when the condition is stable;

[0188] S52, if the early warning level is level one early warning, and it is determined that the early warning is driven by pathogen threat index, generating a suggestion of adjusting the antibiotic scheme; this traces the early warning to the pathogen factor and gives a targeted treatment suggestion;

[0189] S53, if the early warning level is level one early warning, and it is determined that the early warning is driven by host disorder index, generating a suggestion of closely monitoring immune markers; similarly, this strategy traces the early warning to the host factor and suggests that the doctor pay attention to the immune status rather than blindly upgrading the antibiotic;

[0190] S54, if the early warning level is level two early warning, generating an alarm of immediately taking comprehensive intervention measures; this strategy corresponds to the highest risk level and requires immediate implementation of the highest level of clinical response;

[0191] This embodiment effectively converts the diagnostic ability of the model into decision-making ability; it not only issues a general high-risk alarm, but also provides specific, differentiated and executable clinical action suggestions according to the level and internal driving factors of the risk; this deepening from alarm to suggestion is the key difference between this invention and general prediction models, making it a truly intelligent auxiliary decision-making system that can empower clinical diagnosis and treatment processes, significantly improving the timeliness and accuracy of clinical intervention.

[0192] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A smart diagnostic method for clinical microbial infections based on multi-omics data fusion, characterized in that, Includes the following steps: S1. Collect and process time-series multi-omics data to generate instantaneous infection status snapshot vectors; the time-series multi-omics data should include at least: Pathogenomics data: Gene sequence fragments that can reflect the species, abundance, drug resistance genes, and virulence factor expression of pathogenic microorganisms are obtained through metagenomic sequencing technology; Host response data: Genes or proteins that reflect the state of the host's immune system are obtained through transcriptomics or proteomics technologies; Clinical phenotypic data: including vital signs; S2. Construct a temporal state transition model. Input the instantaneous infection state snapshot vector into the temporal state transition model to predict and generate future state vectors. S3. Calculate the pathogen threat index and host dysregulation index based on the future state vector; S4. Based on the pathogen threat index and host dysregulation index, generate a comprehensive risk score and determine the warning level according to the preset warning level threshold. S5. Based on the early warning level and combined with the pathogen threat index and host dysregulation index, generate a graded intervention strategy; S6. Output graded intervention strategies as specific diagnostic and treatment recommendations; The S3 calculation of the pathogen threat index specifically includes: S31. Extract the predicted expression levels of drug resistance genes and virulence factors from the future state vector. S32. Based on the preset risk weights, the predicted expression levels of extracted drug resistance genes and the predicted expression levels of virulence factors are weighted and summed to obtain the pathogen threat index. The S3 calculation of the host disorder index specifically includes: S33. Extract the predicted activation levels of pro-inflammatory pathways and anti-inflammatory pathways from the future state vector. S34. Based on the baseline homeostasis level obtained from the healthy population database, calculate the relative changes in the predicted activation levels of pro-inflammatory pathways and the predicted activation levels of anti-inflammatory pathways, and perform a weighted summation to obtain the host dysregulation index.

2. The intelligent diagnostic method for clinical microbial infections based on multi-omics data fusion according to claim 1, characterized in that, S1 collects and processes time-series multi-omics data, specifically including: S11. Collect pathogen omics data, host response data and clinical phenotype data of patients continuously at multiple key time points to form time-series multi-omics data.

3. The intelligent diagnostic method for clinical microbial infections based on multi-omics data fusion according to claim 2, characterized in that, S1 generates a snapshot vector of the instantaneous infection state, specifically including: S12. Convert time-series multi-omics data into numerical feature vectors; S13. At any key time point, concatenate the corresponding numerical feature vectors to generate a snapshot vector of the instantaneous infection status corresponding to that key time point.

4. The intelligent diagnostic method for clinical microbial infections based on multi-omics data fusion according to claim 1, characterized in that, S2 specifically includes: S21. Use the instantaneous infection state snapshot vectors of the two initial consecutive time points as the starting point for the calculation of the temporal state transition model; S22. Based on the instantaneous infection state snapshot vectors of the current time and historical time, infer and generate future state vectors through the temporal state transition model.

5. The intelligent diagnostic method for clinical microbial infections based on multi-omics data fusion according to claim 1, characterized in that, S4 specifically includes: S41. The modified logistic function is used to normalize the pathogen threat index and the host dysregulation index respectively. S42. Based on the preset weighting coefficients, the normalized pathogen threat index and the host dysregulation index are weighted and fused to generate a comprehensive risk score. S43. Compare the comprehensive risk score with the preset early warning classification threshold, and determine the early warning level as one of the following: stable level, level one early warning, and level two early warning.

6. The intelligent diagnostic method for clinical microbial infections based on multi-omics data fusion according to claim 1, characterized in that, S5 specifically includes: S51. If the warning level is stable, a recommendation to maintain the current treatment plan will be generated. S52. If the warning level is Level 1 and it is determined that the warning is driven by the pathogen threat index, then a recommendation to adjust the antibiotic regimen is generated. S53. If the warning level is Level 1 and it is determined that the warning is driven by the host dysregulation index, then a recommendation to closely monitor immune biomarkers is generated. S54. If the warning level is Level II, an alarm will be generated to immediately take comprehensive intervention measures.

7. A multi-omics data fusion-based intelligent diagnostic system for clinical microbial infections, based on the multi-omics data fusion-based intelligent diagnostic method for clinical microbial infections as described in any one of claims 1-6, characterized in that, Includes the following modules: The data preprocessing module is used to collect and process time-series multi-omics data to generate instantaneous infection status snapshot vectors; The trajectory prediction module is used to input a snapshot vector of the instantaneous infection state and predict and generate a future state vector. The risk quantification module is used to calculate the pathogen threat index and the host dysregulation index based on the future state vector, respectively. The early warning grading module is used to generate a comprehensive risk score and determine the early warning level based on the pathogen threat index and the host dysregulation index. The decision generation module is used to generate graded intervention strategies as specific diagnosis and treatment recommendations based on the warning level and in combination with the pathogen threat index and host dysregulation index.

Images