Early prediction method for novel coronavirus myocardial injury based on mitochondrial function

Abstract

The application discloses a novel coronavirus myocardial injury early prediction method based on mitochondrial function, and the method comprises the following steps: determining a sampling object and completing sample collection; collecting node index data including mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level and mPTP opening degree; completing feature importance evaluation on all node index data, outputting importance scores of each node and interaction relationship, and calibrating pathological stage weight factors in time sequence logistic regression; reserving node progressive correlation penalty items in the time sequence logistic regression layer, calculating myocardial injury risk probability of a single sub-model by integrating node trigger state, time decay weight and penalty items; and integrating output results of multiple time sequence logistic regression sub-models to obtain a final risk prediction value, so that the final prediction has robustness and pathological fitting degree.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of early prediction of myocardial injury, and more particularly relates to a novel coronavirus myocardial injury early prediction method based on mitochondrial function. BACKGROUND

[0002] In the field of myocardial injury risk prediction, accurately grasping the dynamic characteristics of the pathological process is crucial for early intervention. Especially in the context of novel coronavirus infection, the novel coronavirus can induce myocardial injury through multiple pathways such as direct invasion of myocardial cells and induction of systemic inflammatory storm, and its pathological process is often more insidious and complex. The development of myocardial injury follows a clear pathological chain, starting from the decline of mitochondrial membrane potential, and gradually triggering a series of progressive changes, the timing of each node directly determines the severity and speed of injury. However, current prediction techniques face multiple real-world conflicts, severely restricting the improvement of prediction effectiveness.

[0003] Traditional prediction models mostly use static feature weight settings, which are fundamentally contradictory to the dynamic development characteristics of myocardial injury. Such models treat pathological nodes triggered at different times as equally important, and cannot distinguish the differential impact of early and recent triggered nodes on risk, leading to distorted characterization of the injury process. In the context of coronavirus-related myocardial injury, the immune imbalance triggered by the virus may accelerate or disrupt the progression order of pathological nodes, and static models are difficult to cope with such complex timing changes. At the same time, model training relies too much on data statistics, which is easily disturbed by sample noise and individual differences, often resulting in prediction results that violate the logic of pathological progression, making clinical decision-making face the dilemma of "data conclusions contradicting biological common sense".

[0004] In terms of multi-dimensional information fusion, existing methods have not effectively balanced the relationship between pathological priori and data features. Myocardial injury in patients with novel coronavirus infection is often accompanied by respiratory dysfunction, coagulation abnormalities and other multi-system lesions, increasing the complexity of pathological node interaction. On the one hand, existing models do not adequately capture the interaction between nodes, and the independent contribution of a single node and the synergistic effect of multiple nodes are not reasonably quantified, resulting in the underestimation of the warning value of key pathological links; on the other hand, the model integration strategy is simple and rough, and does not fully utilize the prediction advantages of different sub-models, making it difficult to cope with individual diversity of the pathological process, resulting in insufficient robustness of the prediction results.

[0005] These conflicts directly lead to the problems of "prediction lag" and "high false positive rate" in clinical practice, which may cause some mild patients to progress to severe myocardial injury, missing the best window of early intervention. Therefore, building a prediction method that can accurately capture the pathological timing correlation, dynamically integrate multi-source information, and deeply integrate pathological priori and data rules, has become the key to breaking through the existing technical bottleneck. This not only can improve the accuracy and reliability of myocardial injury risk prediction caused by new crown and other reasons, but also can provide a scientific basis for the development of early intervention strategies, which has important practical significance for reducing myocardial injury mortality and improving patient prognosis. SUMMARY

[0006] In view of the difficulty of myocardial injury risk prediction, especially the complex scenario of new crown induced myocardial injury, the present application dynamically captures the timing correlation of mitochondrial pathological nodes, calibrates the node weight to adapt to the pathological logic, integrates the output of multiple sub-models, accurately calculates the risk probability, provides a reliable basis for early intervention, and improves the prediction accuracy and clinical adaptability.

[0007] In view of the above defects or improvement needs of the prior art, as a first aspect of the present application, the present application provides a new coronavirus myocardial injury early prediction method based on mitochondrial function, comprising:

[0008] S1. Determine the sampling object and complete sample collection;

[0009] S2. Complete the node index data collection including mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level and mPTP opening degree;

[0010] S3. Complete feature importance evaluation on all node index data, output importance score of each node and its interaction relationship, and calibrate pathological stage weight factor in time series logistic regression ; Reserve node progressive correlation penalty term in time series logistic regression layer, calculate myocardial injury risk probability of single sub-model by integrating node trigger state, time decay weight and penalty term;

[0011] S4. Integrate the output results of multiple time series logistic regression sub-models to obtain the final risk prediction value.

[0012] Further, the sampling object in S1 is a new coronavirus nucleic acid positive person, including asymptomatic, mild and ordinary patients, and excluding those who have already had severe myocardial injury; the new coronavirus nucleic acid positive person is included within 24 hours after onset / confirmation.

[0013] Further, the specific process of sample collection in S1 is:

[0014] The collected nodes are divided into baseline period, dynamic monitoring period and endpoint period;

[0015] The baseline period is the day of inclusion, when the first peripheral blood sample is collected and the initial mitochondrial status is recorded; the dynamic monitoring period is the 3rd and 7th day, when at least one sample is collected to track the time series of mitochondrial function impairment, respectively; the endpoint period is the 14th day, when the last sample is collected;

[0016] At the time of sample collection, at least 2 ml of peripheral blood is collected each time, and PBMC cells are separated by density gradient centrifugation within 1 hour and stored at below -80℃.

[0017] Further, the calibration method of the pathological stage weight factor in S3 is:

[0018] For the mitochondrial damage nodes , 1 to 5 correspond to membrane potential drop, ROS burst, ATP depletion, mtDNA release, and mPTP opening, respectively; two types of importance scores are calculated: single node importance and interaction importance . ;

[0019] For each node , the integrated importance and associated interaction importance are integrated to obtain a comprehensive score:

[0020] ,

[0021] wherein, is a weight coefficient; is the normalized single node importance score of node ; is the normalized interaction importance score of node and node ; represents the strongest interaction contribution of node to other nodes;

[0022] Based on the progressive relationship of “membrane potential drop → ROS burst → ATP depletion → mtDNA release → mPTP opening”, the initial weight is set;

[0023] The comprehensive score and the initial pathological weight are fused to obtain the calibrated weight :

[0024] ,

[0025] wherein, is an adjustment coefficient; ​The average comprehensive score of all nodes;

[0026] The calibrated weights are normalized to a sum of 1 to facilitate model parameter convergence:

[0027] ,

[0028] In the formula, is the normalized pathological stage weight; is the weight sum of all nodes, is the node index.

[0029] Further, the calculation method of the two types of importance scores is:

[0030] Calculate the importance of a single node:

[0031] ,

[0032] Where, is the number of decision trees, is the node The impurity is reduced in the first tree;

[0033] Evaluate the improvement in the prediction result of the synergistic effect of each pair of nodes, and the node pair , calculate the synergistic effect:

[0034] ,

[0035] Where, is the and single node importance when used as a feature; a positive value indicates the existence of a synergistic early warning value;

[0036] At the same time, further normalize and :

[0037] ,

[0038] ,

[0039] Where, are the maximum and minimum values of all single node original scores, respectively; are the maximum and minimum values of all interaction pair original scores, respectively; is the node The normalized single node importance score; is the node and node ​The normalized interaction importance score.

[0040] Further, the calculation method of the myocardial injury risk probability of the single sub-model in S3 is:

[0041] Let the mitochondrial injury node be wherein , the trigger state is defined as : the node has triggered at time ; if not, then ; the trigger time is ; if not triggered, then ; ;

[0042] Combined with the pathological stage weight factor and the time decay factor, the dynamic weight of the node at time is obtained:

[0043] ,

[0044] wherein is the time decay coefficient, ensuring that the weight of the untriggered node is 0; is the current monitoring time; is the trigger time of the node ;

[0045] A penalty for the node combination violating the pathological order is imposed, denoted as the progressive correlation penalty term ; and then the linear combination result is mapped to a probability through a logic function:

[0046] ,

[0047] ,

[0048] wherein is the linear prediction value; is the intercept term; is the regression coefficient of the node ; , i.e., the myocardial injury risk probability at time .

[0049] Further, the progressive correlation penalty term in S3 is specifically

[0050] The progressive correlation penalty term offsets the interference of abnormal data by imposing a penalty on "feature combinations that violate the pathological order", so that the model output conforms to the biological progressive relationship of "membrane potential decline → ROS burst → ATP depletion → mtDNA release → mPTP opening";

[0051] Let the node trigger time be , the penalty term can be defined as , and its calculation method is as follows:

[0052] ,

[0053] wherein, is the penalty coefficient; is the maximum allowed time interval; is the trigger time of node ; when the actual trigger time satisfies or the interval is too large , is increased, forcing the model to avoid such "wrong correlation" of samples during parameter optimization.

[0054] Further, the calculation method of the final risk prediction value in the S4 is:

[0055] According to the prediction accuracy of each sub-model on the validation set, the weight is assigned, the higher the accuracy of the sub-model, the greater the weight, satisfying ;

[0056] Calculate the weighted sum of the prediction values of the sub-models:

[0057] ,

[0058] wherein, is the basic integrated prediction probability; is the total number of sub-models; is the sub-model index; is the current monitoring time; is the prediction probability of the sub-model at time ;

[0059] Introduce the comprehensive importance score of the node in the early stage output , and correct the basic result:

[0060] ,

[0061] wherein, is the correction coefficient; is the proportion of triggered nodes; is the maximum value of the comprehensive importance of the node.

[0062] As a second aspect of the present application, the present application provides a novel coronavirus myocardial injury early prediction system based on mitochondrial function, comprising:

[0063] A sample collection unit is configured to determine a sampling object and complete sample collection.

[0064] A node data collection unit is configured to complete collection of node index data including mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level, and mPTP opening degree.

[0065] A sub-model construction unit is configured to complete feature importance evaluation of all node index data, output importance scores of each node and its interaction relationship, and calibrate pathological stage weight factors in time sequence logistic regression. A time sequence logistic regression layer retains a node progressive correlation penalty term, and calculates myocardial injury risk probability of a single sub-model by integrating node trigger state, time decay weight, and the penalty term.

[0066] A prediction result output unit is configured to integrate output results of multiple time sequence logistic regression sub-models to obtain a final risk prediction value.

[0067] As a third aspect of the present application, a computer readable storage medium having a computer program stored thereon is also provided, and the computer program is executed by a processor to perform any step of the novel coronavirus myocardial injury early prediction method based on mitochondrial function.

[0068] Overall, compared with the prior art, the above technical solutions conceived by the present application can achieve the following beneficial effects:

[0069] 1. The novel coronavirus myocardial injury early prediction method based on mitochondrial function can extract single feature importance and interaction importance of nodes such as mitochondrial membrane potential and ROS generation level, calibrate pathological stage weight factors of each node in combination with pathological prior constraints. This technical feature enables the model to capture the actual early warning value of each node in clinical data and follow the biological progressive logic of “membrane potential decline ROS explosion ATP depletion mtDNA release mPTP opening”, avoids deviation of the weight from the pathological law due to data noise, and thus improves the adaptability of the weight parameter to the novel coronavirus myocardial injury pathological mechanism, and lays a reliable foundation for subsequent risk calculation.

[0070] 2. The early prediction method of novel coronavirus myocardial injury based on mitochondrial function according to the present application, by integrating node trigger state, time decay weight and progressive correlation penalty term through time series logistic regression, the risk probability of a single sub-model is calculated. Among them, the time decay mechanism dynamically adjusts the weight according to the interval between the node trigger time and the current monitoring time, and the penalty term imposes constraints on the situation that violates the pathological order. This technical feature allows the model to accurately depict the dynamic evolution of pathological nodes in novel coronavirus infection, reflecting the strong influence of recently triggered nodes, and ensuring that the prediction conforms to the pathological process rules, effectively improving the ability to capture the temporal development of the disease.

[0071] 3. The early prediction method of novel coronavirus myocardial injury based on mitochondrial function according to the present application, by integrating the outputs of multiple time series logistic regression sub-models, and combining node comprehensive importance to modify the basic integrated result to obtain the final risk prediction value. The sub-models are diversified through sample sampling and feature selection, and the weights are allocated according to the accuracy during integration, and the modification part strengthens the influence of high importance nodes. This technical feature combines the prediction advantages of multiple models, reduces the bias and variance of a single model, and at the same time highlights the role of key pathological nodes in novel coronavirus myocardial injury, so that the final prediction has both robustness and pathological fitness. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 The flow chart of the early prediction method of novel coronavirus myocardial injury based on mitochondrial function according to the present application is shown in the figure.

[0073] Figure 2 The time series progressive diagram of mitochondrial function injury according to the present application is shown in the figure.

[0074] Figure 3 The mitochondrial membrane potential detection diagram according to the present application is shown in the figure.

[0075] Figure 4 The system unit diagram according to the present application is shown in the figure. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0077] Example 1

[0078] Please refer to Figure 1 The present application 1 provides an early prediction method of novel coronavirus myocardial injury based on mitochondrial function, which comprises:

[0079] S1. Identify the sampling target and complete sample collection;

[0080] S2. Complete the collection of node index data, including mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level, and mPTP opening degree;

[0081] S3. Perform feature importance assessment on all node indicator data, output importance scores for each node and its interactions, and use these scores to calibrate the pathological stage weighting factors in time-series logistic regression. In the temporal logistic regression layer, the progressive association penalty term of nodes is retained. By integrating the node trigger state, time decay weight and penalty term, the probability of myocardial injury risk of a single sub-model is calculated.

[0082] S4. Integrate the outputs of multiple time-series logistic regression sub-models to obtain the final risk prediction value.

[0083] This embodiment 1 further elaborates on the above steps.

[0084] (1) Sample collection

[0085] The sampling subjects in this embodiment 1 were individuals who tested positive for COVID-19 nucleic acid, including asymptomatic, mild, and moderate cases, excluding those with severe myocardial damage; these individuals were included within 24 hours of symptom onset / diagnosis. This setting can eliminate the interference of severe myocardial damage on the early mitochondrial functional state, while ensuring the capture of the initial stage before myocardial damage occurs, providing clean baseline data for subsequent observation of damage development.

[0086] The sample collection process consisted of three phases: the baseline phase, the dynamic monitoring phase, and the endpoint phase. The baseline phase occurred on the day of enrollment, during which the first peripheral blood sample was collected, and the initial state of mitochondria was recorded. This phase established a baseline for individual mitochondrial function, providing a comparative reference for subsequent observation of damage changes. The dynamic monitoring phase took place on days 3 and 7, with at least one sample collected at each time point. Continuous sampling allowed for real-time tracking of a series of temporal chain reactions triggered by mitochondrial functional impairment, enabling timely capture of key changes in the pathological process. The endpoint phase occurred on day 14, during which the final sample collection was completed. This phase fully presented the dynamic evolution of mitochondrial function over two weeks, forming a complete temporal monitoring data chain covering the early to middle stages of the disease.

[0087] Each time the peripheral blood is collected, the amount of collection is not less than 2ml, so as to ensure that there is sufficient sample amount for subsequent cell separation and detection. After the sample is collected, the density gradient centrifugation method is used within 1 hour to separate the PBMC cells, which can quickly obtain high-purity target cells and reduce the influence on mitochondrial function in the sample processing process. The separated cells are stored and preserved, and the storage environment temperature is lower than -80℃, which can maximize the inhibition of cell metabolism, maintain the stability of the mitochondrial related indicators in the sample, ensure the accuracy and reliability of the subsequent detection results, and meet the needs of different detection items for sample quality.

[0088] (2) Node data collection

[0089] Please refer to Figure 2 , Figure 2 is a curve graph of the change of mitochondrial indicators over time when myocardial injury occurs, the horizontal axis is time, and the vertical axis is the level of mitochondrial function. The blue line (ATP content) decreases over time, reflecting energy generation disorder; the red line (mtDNA release) and the green line (mPTP opening degree) increase over time, reflecting mitochondrial injury, inflammation triggering and injury amplification. The curve intersection verifies the pathological node progression sequence of "membrane potential→ATP→mtDNA→mPTP" in the application. Based on this, the node collection data of the application are as follows:

[0090] The five node indicators of mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level and mPTP opening degree are detected; first, the mitochondrial function state is comprehensively captured; the five indicators correspond to the key dimensions of mitochondrial energy metabolism, oxidative stress, structural integrity and injury cascade reaction, etc., which together constitute a complete system reflecting the function of mitochondria, and can systematically present the influence of new coronavirus infection in the early stage on mitochondria. Second, it provides a basis for early warning of myocardial injury; mitochondrial dysfunction is an important pathological basis for myocardial injury, and the changes of these indicators can early indicate the occurrence and progression of mitochondrial injury, which helps to capture the early signal of injury before the heart shows obvious structural or functional abnormalities. Third, it supports the analysis of pathological process; the five indicators have a time sequence correlation, and their dynamic changes can reflect the pathological chain of "membrane potential decline→ROS burst→energy depletion→mtDNA release→mPTP opening", which provides data support for analyzing the molecular mechanism of new coronavirus induced myocardial injury, and helps to build and verify the subsequent risk prediction model.

[0091] The mitochondrial membrane potential is detected by using a fluorescence probe method, which allows the probe to enter the cell and bind to the mitochondria, and the fluorescence intensity change is detected to reflect the membrane potential. When the mitochondrial function is damaged, the membrane potential usually decreases, and this detection can directly reflect the basic functional state of mitochondria.

[0092] ROS production level detection is achieved by means of a fluorescent probe. The probe produces fluorescence after reacting with ROS in the cell, and the ROS production level is quantitatively analyzed by fluorescence intensity. Excessive ROS can cause oxidative stress, which in turn damages mitochondria. This detection can reflect the degree of oxidative damage to mitochondria.

[0093] ATP content detection uses bioluminescence method to measure the content by using the light signal intensity generated by the reaction of ATP and luciferin-luciferase system. ATP, as the energy currency of cells, its content change can directly reflect the energy metabolism function state of mitochondria.

[0094] mtDNA release level detection is carried out by real-time fluorescent quantitative PCR technology to detect the content of mtDNA released by exosomes or PBMC cells in blood samples. Increased mtDNA release is often associated with mitochondrial damage and inflammatory response. This indicator can reflect the chain reaction triggered by mitochondrial damage.

[0095] mPTP opening degree detection uses a specific fluorescent probe to monitor the opening. When mPTP opens, the probe enters the mitochondria and emits a specific fluorescence. The opening degree is evaluated by fluorescence change. Excessive opening of mPTP will lead to mitochondrial dysfunction. This detection can measure the damage degree of mitochondrial structural integrity.

[0096] Through the detection of the above five indicators, the node data related to mitochondrial function can be comprehensively collected, which provides key evidence for subsequent analysis of the early pathological process of myocardial damage caused by novel coronavirus.

[0097] Please refer to Figure 3 The scatter plot takes "mitochondrial membrane potential" as the horizontal axis and "fluorescent indicator" as the vertical axis, presenting the results of mitochondrial membrane potential detection under the background of new crown infection. Green scatter points represent membrane potential collapse (JC-1 monomer emits green light), and red scatter points represent normal membrane potential (JC-1 aggregate emits red light). This directly shows the distribution of sample populations with normal and collapsed membrane potential, which is the result of the index detection link, provides key data support for model input and verification of the pathological logic that "membrane potential decline is the starting node of damage", and subsequent risk prediction.

[0098] (3) Sub-model construction

[0099] This process mainly calibrates the pathological stage weight factor by evaluating the importance of mitochondrial damage nodes and their interaction, and integrates the node trigger state, time decay weight and progressive correlation penalty term by combining time series logistic regression to calculate the myocardial injury risk probability of a single sub-model.

[0100] Specifically, when calibrating the pathological stage weight factor, the obtained node comprehensive importance is combined with the initial weight based on the pathological progression relationship, which not only reflects the actual importance difference reflected by the data, but also follows the biological logic, so that the weight is consistent with the data law and the pathological mechanism, providing a reliable basis for subsequent risk calculation. Calculating the importance score and normalizing it can effectively quantify the early warning value of individual nodes and the synergistic effect between nodes, eliminate the dimensional differences of different dimension indicators, and make the importance evaluation more scientific and facilitate comprehensive analysis.

[0101] In a preferred embodiment, the pathological stage weight factor is calibrated by the following method:

[0102] For the mitochondrial damage node , wherein 1 to 5 correspond to membrane potential decline, ROS burst, ATP depletion, mtDNA release, and mPTP opening, respectively; two types of importance scores are calculated: individual node importance and interaction importance .

[0103] For each node , the individual importance and the associated interaction importance are integrated to obtain a comprehensive score:

[0104] ,

[0105] wherein, is a weight coefficient; is the normalized individual node importance score of node ; is the normalized interaction importance score of node and node ; represents the strongest interaction contribution of node to other nodes;

[0106] Based on the progressive relationship of "membrane potential decline → ROS burst → ATP depletion → mtDNA release → mPTP opening", the initial weight is set;

[0107] The comprehensive score is combined with the initial pathological weight to obtain the calibrated weight :

[0108] ,

[0109] wherein, is an adjustment coefficient; is the average comprehensive score of all nodes (to ensure that the calibrated weight still satisfies the pathological progression, );

[0110] The calibrated weights are normalized to a sum of 1 to facilitate model parameter convergence.

[0111] ,

[0112] In the formula, The normalized weights for pathological stages; The sum of the weights of all nodes. For node indexing.

[0113] In a preferred embodiment, the two types of importance scores are calculated as follows:

[0114] Calculate the importance of a single node:

[0115] ,

[0116] in, For the number of decision trees, For nodes In the The reduction in the number of trees Impurity;

[0117] To assess the extent to which the synergistic effect between pairs of nodes improves the prediction results, the node pairs... Calculate the synergistic effect:

[0118] ,

[0119] in, for and Simultaneously, the importance of a single node when used as a feature; a positive value indicates the existence of collaborative early warning value (such as...). (Reflecting the interaction between membrane potential decrease and ROS burst).

[0120] At the same time, further as well as Normalization is performed:

[0121] ,

[0122] ,

[0123] in, These are the maximum and minimum values ​​of the original scores for all individual nodes, respectively. These represent the maximum and minimum values ​​of the original ratings for all interactions; For nodes Normalized importance score of a single node; For nodes With nodes Normalized interaction importance score.

[0124] In calculating the risk probability of individual sub-models, a time decay mechanism is introduced to give higher weight to nodes with earlier trigger time, which conforms to the characteristics of damage progression over time and can accurately capture the impact of different time nodes on risk. Setting progressive association penalty terms can avoid the model being disturbed by abnormal data and deriving results that violate the pathological order, ensuring that the prediction conforms to the biological logic of "membrane potential decline→ROS explosion→ATP depletion→mtDNA release→mPTP opening", and improving the rationality of the prediction. Through a logic function, the linear result is mapped to a probability, making the output more intuitive and facilitating clinical understanding and application, providing a clear quantitative basis for myocardial injury risk assessment.

[0125] In a preferred embodiment, the method for calculating the myocardial injury risk probability of individual sub-models is as follows:

[0126] Let the mitochondrial damage node be , where , and define the trigger state as: node has triggered before time , then , and has not triggered, then ; the trigger time is , and if it has not triggered, then ;

[0127] Combining the pathological stage weight factor and the time decay factor, the dynamic weight of node at time is obtained:

[0128] ,

[0129] wherein is the time decay coefficient (the earlier the trigger time, the higher the weight), ensures that the weight of untriggered nodes is 0; is the current monitoring time; is the trigger time of node ;

[0130] A penalty is imposed on node combinations that violate the pathological order, denoted as the progressive association penalty term ; and then a logic function is used to map the linear combination result to a probability:

[0131] ,

[0132] ,

[0133] wherein is a linear prediction value; is an intercept term; is a regression coefficient of the node obtained by training historical data; is a myocardial injury risk probability at time .

[0134] In a preferred embodiment, the progressive correlation penalty term is specifically

[0135] The progressive correlation penalty term offsets the interference of abnormal data by imposing a penalty on "feature combinations that violate the pathological order", so that the model output conforms to the biological progressive relationship of "membrane potential decline → ROS burst → ATP depletion → mtDNA release → mPTP opening";

[0136] Let the node trigger time be (the expected order of pathology), the penalty term can be defined as , and the calculation method is as follows:

[0137] ,

[0138] wherein, is a penalty coefficient; is the maximum allowed time interval; is the trigger time of the node ; when the actual trigger time satisfies or the interval is too large , is increased, forcing the model to avoid "wrong association" of such samples during parameter optimization.

[0139] (4) Prediction result output

[0140] The prediction result output process of this embodiment 1 is the output of integrating multiple time series logistic regression sub-models. First, weights are assigned according to the prediction accuracy of each sub-model on the validation set, the weighted sum is calculated to obtain the basic integrated prediction probability, then the node comprehensive importance score is introduced, the basic result is corrected combined with the triggered node proportion and the correction coefficient, and the final risk prediction value is obtained.

[0141] In a preferred embodiment, the calculation method of the final risk prediction value is:

[0142] According to the prediction accuracy (such as AUC value) of each sub-model on the validation set, weights are assigned , the higher the accuracy of the sub-model, the greater the weight, satisfying ;

[0143] Calculate the weighted sum of the sub-model prediction values:

[0144] ,

[0145] wherein, is the base integrated prediction probability; is the total number of sub-models; is the sub-model index; is the current monitoring time; is the sub-model at time is the prediction probability;

[0146] introducing the node comprehensive importance score of the early stage output , the base result is corrected:

[0147] ,

[0148] wherein, is the correction coefficient; is the proportion of triggered nodes; is the maximum value of the node comprehensive importance.

[0149] According to the sub-model prediction accuracy, the weight is distributed and the weighted sum is calculated, which can make the sub-model with high accuracy play a greater role in the integrated result, integrate the advantages of multiple sub-models, reduce the limitations of a single model, reduce the prediction bias, improve the reliability and stability of the base integrated result, and provide better basis for subsequent correction.

[0150] The node comprehensive importance score is introduced for correction, the key node influence is highlighted by means of the maximum value of the node comprehensive importance, the degree of pathological process advancement is reflected by combining the proportion of triggered nodes, and the adjustment range is adjusted by the correction coefficient, so that the final prediction result integrates multiple model information and strengthens the role of key pathological characteristics, which is more in line with the pathological logic of new coronavirus myocardial injury, makes the prediction value more accurate, and provides a more scientific decision basis for early clinical intervention.

[0151] Embodiment 2

[0152] Please refer to Figure 4 , the embodiment 2 provides a kind of early prediction system of new coronavirus myocardial injury based on mitochondrial function, comprising:

[0153] sample collection unit is used to determine sampling object and complete sample collection;

[0154] Node data acquisition unit is used to complete the node index data acquisition including mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level and mPTP opening degree;

[0155] Sub-model construction unit is used to complete feature importance evaluation for all node index data, output the importance score of each node and its interaction relationship, to calibrate pathological stage weight factor in time sequence logistic regression The time sequence logic regression layer retains a node progressive correlation penalty term, and a myocardial injury risk probability of a single sub-model is calculated by integrating a node trigger state, a time decay weight and the penalty term.

[0156] The prediction result output unit is configured to integrate output results of the plurality of time sequence logic regression sub-models to obtain a final risk prediction value.

[0157] Embodiment 3

[0158] The embodiment 3 also provides a computer readable storage medium, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement any step of a novel coronavirus myocardial injury early prediction method based on mitochondrial function.

[0159] The computer readable storage medium can include a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes.

[0160] For the computer readable storage medium provided in the present application, refer to the above method embodiments, which will not be repeated herein.

[0161] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A novel coronavirus myocardial injury early prediction method based on mitochondrial function, characterized in that, Comprise; S1. Determine the sampling object and complete sample collection; S2. Complete the node index data collection including mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level and mPTP opening degree; S3. Perform feature importance evaluation on all node indicator data, output the importance score of each node and its interaction relationship, to calibrate the pathological stage weight factor in the time series logistic regression ; In the time sequence logistic regression layer, the node progressive correlation penalty term is reserved, the node trigger state, the time decay weight and the penalty term are integrated, and the myocardial injury risk probability of a single sub-model is calculated; S4. The output results of multiple time sequence logistic regression sub-models are integrated to obtain the final risk prediction value; The calculation method of the myocardial injury risk probability of a single sub-model in S3 is: Let the mitochondria damage node be wherein , define the trigger state is: node triggered at time , then , if not triggered, then ; the trigger time is , if not triggered, then ; Combining the pathological stage weight factor and the time decay factor, the node is given a time decay weight at time : wherein, is a time decay coefficient, ensuring that the weight of a non-triggering node is 0; is a current monitoring time; is a triggering time of a node . A penalty is imposed for combinations of nodes that violate the pathological order, denoted as the progressive association penalty term and the linear combination result is mapped to a probability by a logistic function: wherein, is a linear prediction value; is an intercept term; is a regression coefficient for a node ; and is a probability of cardiac injury risk at time ; The progressive correlation penalty term in S3 is specifically: The progressive correlation penalty term offsets the interference of abnormal data by imposing a penalty on "feature combinations that violate the pathological order", so that the model output conforms to the biological progressive relationship of "membrane potential decline→ROS explosion→ATP depletion→mtDNA release→mPTP opening"; Let the node trigger time be The penalty term can be defined as The calculation method is as follows: wherein, is a penalty coefficient; is the maximum allowed time interval; is the triggering time of the node ; when the actual triggering time meets or the interval is too large , is increased, forcing the model to avoid "wrong associations" of such samples when optimizing parameters.

2. The method for early prediction of coronavirus myocardial injury based on mitochondrial function according to claim 1, characterized in that, The sampling object in S1 is a new coronavirus nucleic acid positive person, including asymptomatic, mild and ordinary patients, and excluding those who have already had severe myocardial injury; The new coronavirus nucleic acid positive person is included within 24 hours after the onset / confirmation.

3. The method according to claim 1, wherein the method is characterized by, The specific process of sample collection in S1 is: The collection nodes are divided into baseline period, dynamic monitoring period and endpoint period; The baseline period is the day of inclusion, and the first peripheral blood sample collection is completed in the baseline period and the initial mitochondrial state is recorded; The dynamic monitoring period is the 3rd day and the 7th day, and at least 1 sample is collected in the dynamic monitoring period respectively to track the time sequence linkage reaction of mitochondrial function injury; The endpoint period is the 14th day, and the last sample collection is completed in the endpoint period; When collecting samples, at least 2ml of peripheral blood is collected each time, and PBMC cells are separated by density gradient centrifugation within 1 hour and stored at below-80℃.

4. The method according to claim 1, wherein the method is characterized by, The pathological stage weight factor in S3 The calibration method is: Mitochondrial damage nodes wherein , 1 to 5 correspond to membrane potential drop, ROS burst, ATP depletion, mtDNA release, mPTP opening in turn; and then calculate two types of importance scores: single node importance and interaction relationship importance ; For each node , integrate its individual importance and the importance of the associated interactions to get a comprehensive score: wherein, is a weight coefficient; is a node a normalized individual node importance score; is a node a node a normalized interaction importance score; denotes a node the strongest interaction contribution to other nodes; Based on the progressive relationship of "membrane potential drop → ROS burst → ATP depletion → mtDNA release → mPTP opening", set the initial weight ; combining the integrated score with the initial pathology weight to obtain a calibrated weight : wherein, is a tuning coefficient; is the average overall score of all nodes; The calibrated weight is normalized to a total sum of 1 to facilitate model parameter convergence: wherein is the normalized pathology stage weight; is the sum of weights for all nodes, is the node index.

5. The method for early prediction of coronavirus myocardial injury based on mitochondrial function according to claim 4, characterized in that, The calculation method of the two types of importance scores is: Calculate the importance of a single node: wherein, is the number of decision trees, is the node In the first reduced impurity; The synergistic effect of each pair of nodes is evaluated to determine the extent to which the prediction result is improved, and the pair of nodes , the synergistic effect is calculated: wherein, is and single node importance while simultaneously being a feature; positive values indicate a synergistic early warning value; At the same time, further normalize and perform normalization processing: wherein, are the maximum and minimum of all single-node raw scores, respectively; are the maximum and minimum of all interaction pair raw scores, respectively; is the node is the normalized single-node importance score. is the node is the node is the normalized interaction relationship importance score.

6. The method for early prediction of coronavirus myocardial injury based on mitochondrial function according to claim 4 or 5, characterized in that, The calculation method of the final risk prediction value in S4 is: According to the prediction accuracy of each sub-model on the validation set, weights are allocated The higher the accuracy of the sub-model, the greater the weight, satisfying The weighted sum of the prediction values of the sub-models is calculated: wherein, is the base integrated prediction probability; is the total number of submodels; is the submodel index; is the current monitoring time; is the submodel at time is the prediction probability; Node comprehensive importance score of introducing previous output Correcting base result: wherein, is a correction factor; is the proportion of triggered nodes; is the maximum value of the integrated importance of the nodes. 7.A system for early prediction of myocardial injury caused by a novel coronavirus based on mitochondrial function, characterized in that, Comprise; A sample collection unit for determining the sampling object and completing sample collection; A node data collection unit for completing node index data collection including mitochondrial membrane potential, ROS generation level, ATP content, mtDNA release level and mPTP opening degree; The sub-model construction unit is configured to perform feature importance evaluation on all node index data, and output importance scores of each node and interaction relationship thereof, so as to calibrate pathological stage weight factors in the time sequence logistic regression ; In the time sequence logistic regression layer, the node progressive correlation penalty term is reserved, the node trigger state, the time decay weight and the penalty term are integrated, and the myocardial injury risk probability of a single sub-model is calculated; A prediction result output unit for integrating the output results of multiple time sequence logistic regression sub-models to obtain the final risk prediction value; The calculation method of the myocardial injury risk probability of a single sub-model in the sub-model construction unit is: Let the mitochondria damage node be wherein , define the trigger state is: node triggered at time , then , if not triggered, then ; the trigger time is , if not triggered ; Combining the pathological stage weight factor and the time decay factor, the node is given a time decay weight at time : wherein, is a time decay coefficient, ensuring that the weight of a non-triggering node is 0; is the current monitoring time; is the triggering time of a node . A penalty is imposed for combinations of nodes that violate the pathological order, denoted as the progressive association penalty term and the linear combination result is mapped to a probability by a logistic function: wherein, is a linear prediction value; is an intercept term; is a regression coefficient for a node ; and is a probability of cardiac injury risk at time ; The progressive correlation penalty term in the sub-model construction unit is specifically: The progressive correlation penalty term offsets the interference of abnormal data by imposing a penalty on "feature combinations that violate the pathological order", so that the model output conforms to the biological progressive relationship of "membrane potential decline→ROS explosion→ATP depletion→mtDNA release→mPTP opening". Let the node trigger time be The penalty term can be defined as The calculation method is as follows: wherein, is a penalty coefficient; is the maximum allowed time interval; is the triggering time of the node When the actual triggering time meets or the interval is too large , is increased, forcing the model to avoid "wrong associations" of such samples when optimizing parameters.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the early prediction method of novel coronavirus myocardial injury based on mitochondrial function according to any one of claims 1-6.

Images