A biomarker-based prediction device for trauma-related acute kidney injury targets

Abstract

The present invention discloses a biomarker-based target prediction device for trauma-related acute kidney injury, which relates to the technical field of injury prediction. The device includes: obtaining a biomarker dataset for trauma-related acute kidney injury; performing treatment analysis of trauma-related acute kidney injury to determine treatment target information; obtaining multiple biomarker features and constructing a feature relationship matrix based on the multiple biomarker features; constructing a deep learning model to obtain multiple target prediction information; performing matching and screening on the multiple target prediction information to determine drug candidate molecule parameters; obtaining a treatment effect score, associating the treatment effect score with the multiple target prediction information, and generating a target prediction report. It solves the technical problems of low accuracy in early diagnosis and treatment target prediction of trauma-related acute kidney injury and low drug screening efficiency in the prior art, and achieves the technical effects of improving the target prediction accuracy, optimizing the drug screening process, and realizing personalized treatment.

Description

Technical Field

[0001] This application relates to the technical field of injury prediction, and particularly to a biomarker-based prediction device for trauma-associated acute kidney injury targets. Background Art

[0002] Trauma-associated acute kidney injury (T-AKI) is one of the most common functional injuries in trauma patients. Acute kidney injury (AKI) refers to a sharp decline in kidney function, usually manifested as symptoms such as an increase in serum creatinine levels and a decrease in urine output. Trauma-associated acute kidney injury, as the name implies, is an acute kidney injury caused by factors such as trauma or surgical trauma. Its pathogenesis is complex and involves multiple pathological changes in the kidney, including tubular injury, insufficient renal blood perfusion, immune responses, etc. With the continuous development of the application of biomarkers in disease diagnosis and treatment, and the rapid development of the biomedical field, targeted therapy has gradually become a new direction for disease treatment. However, currently, the treatment of trauma-related AKI mainly relies on supportive measures and lacks targeted drug treatment. Although some potential biomarkers and treatment targets have been identified through bioinformatics methods, further experimental verification and drug development are still needed. Traditional target screening usually relies on experimental data or existing clinical experience, making target prediction inaccurate and inefficient, thus affecting the treatment effect.

[0003] Therefore, in the current related technologies, there are technical problems such as low accuracy in the early diagnosis and treatment target prediction of trauma-related acute kidney injury and low drug screening efficiency. Summary of the Invention

[0004] By providing a biomarker-based prediction device for trauma-associated acute kidney injury targets, this application solves the technical problems of low accuracy in the early diagnosis and treatment target prediction of trauma-related acute kidney injury and low drug screening efficiency in the existing technology, and achieves the technical effects of improving the accuracy of target prediction, optimizing the drug screening process, and realizing personalized treatment.

[0005] The present application provides a biomarker-based target prediction device for trauma-related acute kidney injury. The device includes: a biomarker dataset acquisition module for screening based on a public database to obtain a biomarker dataset for trauma-related acute kidney injury; a treatment target information determination module for invoking a literature report library and combining it with the public database to conduct a treatment analysis of trauma-related acute kidney injury and determine treatment target information; a biomarker feature acquisition module for performing feature analysis based on the biomarker dataset and the treatment target information to obtain a plurality of biomarker features, and constructing a feature relationship matrix according to the plurality of biomarker features; a target prediction information acquisition module for constructing a deep learning model based on a self-attention mechanism, synchronizing the biomarker dataset to the deep learning model according to the feature relationship matrix for prediction, and obtaining a plurality of target prediction information, where the plurality of target prediction information includes a plurality of confidence scores; a verification result generation module for using a drug database to combine with the plurality of confidence scores to perform matching screening on the plurality of target prediction information, determining drug candidate molecule parameters, and verifying the plurality of target prediction information according to the drug candidate molecule parameters to generate a verification result; a treatment effect score acquisition module for performing a treatment evaluation on the plurality of target prediction information according to the verification result to obtain a treatment effect score, and associating the treatment effect score with the plurality of target prediction information to generate a target prediction report.

[0006] In a possible implementation manner, the biomarker feature acquisition module further performs the following processing: calculating a correlation coefficient between the biomarker dataset and the treatment target information to generate a plurality of correlation coefficients; performing functional annotation on the treatment target information according to the plurality of correlation coefficients to determine a plurality of target function annotation information; performing acute kidney injury analysis on the biomarker dataset based on the plurality of target function annotation information in combination with the plurality of correlation coefficients to determine a plurality of biomarker features; constructing a null two-dimensional matrix, converting the plurality of target function annotation information into matrix columns, and converting the plurality of biomarker features into matrix rows; filling the matrix columns and the matrix rows into the null two-dimensional matrix according to the plurality of correlation coefficients to generate the feature relationship matrix.

[0007] In a possible implementation, the target prediction information acquisition module further performs the following processes: performing weighted processing by regularization based on the self-attention mechanism to construct a self-attention layer, where the input end of the self-attention layer is communicatively connected to the output end of the input layer in the deep learning model; classifying based on the feature relationship matrix to determine multiple subspaces, and using the self-attention mechanism to connect the multiple subspaces to construct a multi-head attention mechanism; retrieving a target prediction task, introducing a loss function according to the target prediction task for classification regression, and constructing a fully connected layer; evaluating the self-attention layer, the multi-head attention mechanism, and the fully connected layer by cross-validation, and constructing the deep learning model according to the evaluation results; where the output end of the self-attention layer is communicatively connected to the input end of the multi-head attention mechanism, the input end of the fully connected layer is communicatively connected to the output end of the multi-head attention mechanism and the output end of the input layer, and the output end of the fully connected layer is the output end of the deep learning model.

[0008] In a possible implementation, the target prediction information acquisition module further performs the following processes: synchronizing the biomarker data set as the first vector input data to the input layer, and synchronizing the feature relationship matrix as the second vector input data to the input layer; using the self-attention layer to perform weighted calculation on the second vector input data to obtain multiple weight coefficients; performing multi-dimensional capture on the first vector input data and the second vector input data by the multi-head attention mechanism according to the multiple weight coefficients to generate multi-space attention parameters; generating multiple confidence scores according to the multi-space attention parameters; activating the fully connected layer according to the multiple confidence scores, and performing classification regression prediction on the first vector input data and the second vector input data according to the fully connected layer to generate a classification regression prediction result; matching the multiple confidence scores with the classification regression prediction result to generate the multiple target prediction information.

[0009] In a possible implementation, the verification result generation module further performs the following processes: sorting the multiple target prediction information in descending order according to the multiple confidence scores to generate a target prediction sequence; calculating the mean according to the target prediction sequence to set a confidence threshold; comparing the multiple confidence scores with the confidence threshold to perform an initial screening on the multiple target prediction information, and storing the target prediction information corresponding to the confidence scores greater than or equal to the confidence threshold to generate an initial screening result; using the initial screening result as index information to perform a matching search on the drug database to generate a drug interaction relationship; screening the drug database based on the drug interaction relationship to obtain the drug candidate molecule parameters.

[0010] In a possible implementation manner, the verification result generation module further performs the following processing: classifying the drug candidate molecules according to the drug interaction relationship to generate a plurality of to-be-experimented data, where the plurality of to-be-experimented data includes in vitro to-be-experimented data and in vivo to-be-experimented data; performing in vitro experiment analysis by combining the in vitro to-be-experimented data with the drug candidate molecules to obtain in vitro experiment influence effect parameters; performing in vivo experiment analysis by combining the in vivo to-be-experimented data with the drug candidate molecules to obtain in vivo experiment influence effect parameters; performing in vitro pharmacodynamic evaluation on the plurality of target prediction information based on the in vitro experiment influence effect parameters to generate an in vitro pharmacodynamic score; performing in vivo pharmacodynamic evaluation on the plurality of target prediction information based on the in vivo experiment influence effect parameters to generate an in vivo pharmacodynamic score; performing safety verification on the plurality of target prediction information according to the in vitro pharmacodynamic score combined with the in vivo pharmacodynamic score to generate a target safety coefficient, and adding the target safety coefficient to the verification result.

[0011] In a possible implementation manner, the treatment effect score obtaining module further performs the following processing: setting a quality evaluation index based on the target safety coefficient; traversing the plurality of target prediction information to perform treatment scoring according to the quality evaluation index to generate an in vivo treatment quality score and an in vitro treatment quality score; performing in vivo efficacy analysis according to the in vivo treatment quality score combined with the in vivo pharmacodynamic score to generate an in vivo treatment effect score; performing in vitro efficacy analysis according to the in vitro treatment quality score combined with the in vitro pharmacodynamic score to generate an in vitro treatment effect score; respectively calculating the in vivo treatment effect score and the in vitro treatment quality score with the plurality of confidence scores of the plurality of target prediction information to generate the treatment effect score.

[0012] In a possible implementation manner, the treatment effect score obtaining module further performs the following processing: performing mapping analysis according to the in vivo treatment effect score and the plurality of target prediction information to generate a first mapping relationship parameter; performing mapping analysis according to the in vitro treatment effect score and the plurality of target prediction information to generate a second mapping relationship parameter; constructing a mapping relationship network according to the first mapping relationship parameter combined with the second mapping relationship parameter according to the plurality of confidence scores; performing transfer learning on the plurality of target prediction information according to the treatment effect score according to the mapping relationship network to generate a target learning result; feeding back the target learning result to the drug database for response to generate a treatment suggestion, and adding the treatment suggestion to the target prediction report.

[0013] A biomarker-based target prediction device for trauma-related acute kidney injury proposed in this application is used to obtain a biomarker dataset for trauma-related acute kidney injury; conduct treatment analysis of trauma-related acute kidney injury to determine treatment target information; obtain multiple biomarker features, and construct a feature relationship matrix based on the multiple biomarker features; construct a deep learning model to obtain multiple target prediction information; perform matching screening on the multiple target prediction information to determine drug candidate molecule parameters; obtain a treatment effect score, associate the treatment effect score with the multiple target prediction information, and generate a target prediction report. This solves the technical problems of low accuracy in early diagnosis and treatment target prediction of trauma-related acute kidney injury and low drug screening efficiency in the prior art, and achieves the technical effects of improving target prediction accuracy, optimizing the drug screening process, and realizing personalized treatment. Brief Description of the Drawings

[0014] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the device according to the embodiments of the present application. It should be understood that the operations in the front or below do not necessarily need to be executed precisely in sequence. On the contrary, according to needs, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0015] Figure 1 Structural schematic diagram of a biomarker-based target prediction device for trauma-related acute kidney injury provided by an embodiment of the present application;

[0016] Figure 2 Execution process schematic diagram of the biomarker feature acquisition module in a biomarker-based target prediction device for trauma-related acute kidney injury provided by an embodiment of the present application.

[0017] Explanation of reference numerals: Biomarker dataset acquisition module 10, treatment target information determination module 20, biomarker feature acquisition module 30, target prediction information acquisition module 40, verification result generation module 50, treatment effect score acquisition module 60. Detailed Embodiments

[0018] The above description is only an overview of the technical solutions of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specifically gives the detailed embodiments of this application.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following will further describe this application in detail with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0020] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first / second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, apparatus, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0021] An embodiment of this application provides a biomarker-based traumatic acute kidney injury target prediction device, as Figure 1 shown. The device includes:

[0022] A biomarker dataset acquisition module 10, configured to screen based on a public database to obtain a biomarker dataset for traumatic acute kidney injury.

[0023] Preferably, biomarker information related to trauma-associated acute kidney injury (T-AKI) is extracted from public databases (public, verified resource databases), which usually contain a large amount of biological data, including biological information at different levels such as gene expression, protein levels, metabolite changes, etc. Specifically, relevant biological data resources are screened from multiple public databases. For example, Gene Expression Omnibus (GEO) is a public database containing a large amount of gene expression data, covering many gene expression datasets related to diseases; KEGG (Kyoto Encyclopedia of Genes and Genomes) is an information database integrating genes, proteins, and metabolic pathways, which can provide a theoretical basis for the screening of biomarkers; Human Protein Atlas provides expression data of human proteins, which can reveal the relationship between proteins and diseases, etc. In public databases, through keyword retrieval, biomarkers related to acute kidney injury or trauma are located. For example, genes, proteins, or metabolites that are significantly changed in trauma or acute kidney injury patients are searched for, and then the screened relevant biomarker data are integrated to form a comprehensive dataset containing trauma-associated acute kidney injury markers, that is, a biomarker dataset, which includes not only the types of markers (such as certain genes or proteins), but also their expression levels under different pathological conditions, relevant biological information, etc.

[0024] The treatment target information determination module 20 is used to call the literature report library and combine it with the public database to perform treatment analysis of trauma-associated acute kidney injury and determine the treatment target information.

[0025] Preferably, by combining the treatment research results in the literature and the biomarker data in the public database, a comprehensive analysis is carried out to identify potential treatment targets for trauma-related acute kidney injury (T-AKI). Among them, the literature report library refers to the data obtained by sorting out and summarizing the research results of trauma-related acute kidney injury, usually including a large number of scientific literatures, clinical studies, experimental studies, etc., such as databases like PubMed and Scholar, which contain various research results on trauma, acute kidney injury and their treatments. Specifically, the public database can not only find the biomarker information of trauma-related acute kidney injury, but also obtain the research results of targeted therapies related to these biomarkers. For example, a certain biomarker (such as KIM-1, IL-18, etc.) is closely related to acute kidney injury. There may be studies in the literature showing specific drugs or treatment methods for these biomarkers. By calling these literature reports and combining the biomarker information in the public database with the treatment strategies in the literature, potential drug targets or treatment targets can be identified. These targets may be molecules or proteins directly related to the pathogenesis of trauma-related acute kidney injury, and then the treatment target information is determined. Among them, the treatment target refers to the key molecule or signaling pathway that can be intervened by drugs or other treatment means during the occurrence of the disease. For example, the occurrence of trauma-related acute kidney injury is usually accompanied by processes such as oxidative stress, inflammatory response and cell death in the kidney. The targets related to these processes, such as specific receptors, key enzymes or inflammatory factors (such as TNF-α, IL-1β, etc.) in renal tubular cells, may be treatment targets.

[0026] The biomarker feature acquisition module 30 is used to perform feature analysis based on the biomarker data set and the treatment target information, obtain a plurality of biomarker features, and construct a feature relationship matrix according to the plurality of biomarker features.

[0027] Preferably, through in-depth analysis of the biomarker dataset and therapeutic target information, relevant features are extracted, and then a matrix describing the relationships between these features is constructed, that is, a feature relationship matrix is constructed, which helps to provide effective input features for subsequent deep learning models to optimize the prediction and screening of therapeutic targets. Specifically, through the comparative analysis of the biomarker dataset and therapeutic targets, key features are identified, that is, multiple biomarker features are obtained. Among them, biomarker features refer to the extracted specific attributes that can characterize the pathological state or be related to therapeutic targets. For example, some biomarkers may be upregulated in trauma-related acute kidney injury, while others may change after targeted therapy. Common biomarker features may include the expression differences of biomarkers in healthy and injured states, the dynamic changes of biomarkers over time, and the correlation between biomarkers and other biomarkers or therapeutic targets. Then, a feature relationship matrix is constructed using multiple biomarker features. Specifically, the rows and columns of the feature relationship matrix represent different biomarker features, and each element in the matrix represents the strength of the relationship between these two features, such as represented by correlation or synergy. Through the feature relationship matrix, the complex relationships between different biomarker features can be captured, thereby improving the accuracy and reliability of therapeutic target prediction.

[0028] The target prediction information acquisition module 40 is used to construct a deep learning model based on the self-attention mechanism, synchronize the biomarker dataset to the deep learning model according to the feature relationship matrix for prediction, and obtain multiple target prediction information, and the multiple target prediction information includes multiple confidence scores.

[0029] Preferably, the self-attention mechanism is used to process the biomarker dataset and the feature relationship matrix in the deep learning model, and then the therapeutic targets of trauma-related acute kidney injury (T-AKI) are predicted, and a corresponding confidence score is provided for each predicted target. Among them, the self-attention mechanism is an important technology in deep learning, especially having significant advantages in processing sequence data (such as text or time series data). Each biomarker in the biomarker dataset can be regarded as a feature or data at a time point. By modeling the relationships between these features through the self-attention mechanism, their correlations are automatically learned, and a deep learning model (such as using large model architectures like Transformer or their variants, such as BERT, GPT, etc. to construct the deep learning model) is obtained to better capture the roles of different biomarkers in trauma-related acute kidney injury. Specifically, synchronizing the biomarker dataset and the feature relationship matrix to the deep learning model, and using the deep learning model to process the independent information of each biomarker, the complex correlation relationships between them can also be captured, thereby improving the accuracy of target prediction.

[0030] Preferably, a deep learning model is then used to start predicting the treatment targets for trauma-related acute kidney injury. Specifically, based on the input biomarker data and its characteristic relationship matrix, the deep learning model predicts multiple potential treatment targets, which are usually molecules closely related to the occurrence, development, or treatment of the disease (such as genes, proteins, receptors, etc.). Each treatment target may contain multiple pieces of information, such as the predicted target name, the type of the target (such as gene, protein, enzyme, etc.), the biological processes that the target may be involved in, etc. And each treatment target is accompanied by a confidence score, indicating the reliability of the model's prediction for this target. Among them, the confidence score usually ranges from 0 to 1, and the closer it is to 1, the more confident the model is that the target is an effective treatment target. For example, when predicting the targets for trauma-related acute kidney injury, the model may predict target 1 (such as KIM-1) with a confidence score of 0.85, indicating that the model has 85% confidence in predicting KIM-1 as a target; target 2 (such as IL-18) with a confidence score of 0.92, indicating that the model has 92% confidence in predicting IL-18 as a target; target 3 (such as NGAL) with a confidence score of 0.75, indicating that the model has 75% confidence in predicting NGAL as a target. By using deep learning and the self-attention mechanism, analyze the biomarker data and its characteristic relationship, predict the potential treatment targets for trauma-related acute kidney injury, and provide a confidence score for each predicted target to help improve the accuracy and reliability of target prediction.

[0031] The verification result generation module 50 is used to match and screen the multiple target prediction information by combining with the drug database according to the multiple confidence scores, determine the drug candidate molecule parameters, and verify the multiple target prediction information according to the drug candidate molecule parameters to generate verification results.

[0032] Preferably, the target prediction results are combined with the information in the drug database. Through screening and verification, drug candidate molecules with potential therapeutic effects on trauma-related acute kidney injury (T-AKI) are finally determined, and their relevance and efficacy with the predicted targets are evaluated. Specifically, the drug database is used in combination with confidence scores to match and screen the target prediction information. The drug database usually contains a large amount of drug molecule information, including approved drugs, drug molecules in clinical trials, and some potential new drug candidate molecules. Common drug databases include DrugBank, PubChem, ChEMBL, etc. In the drug database, drug candidate molecules related to these predicted targets are searched and screened. For example, the drug database may provide known inhibitors, agonists, or other related molecules of a certain target. By comparing the molecules in the drug database with the target prediction information, possible drug candidate molecules are screened, and then the parameters of the drug candidate molecules are determined. Generally, it refers to the relationship between the drug molecule and the target and its potential therapeutic effects, which may include the structural characteristics of the drug molecule, such as the chemical structure, molecular weight, lipophilicity (such as LogP), solubility, etc., drug-target interaction information, the binding affinity (such as IC50, Ki value, etc.) between the drug candidate molecule and the target, the binding site, etc., and the biological effects of the drug. Each drug molecule in the database is matched with the target prediction results according to these parameters to evaluate its therapeutic effect and possibility as a target.

[0033] Preferably, verifying multiple target prediction information according to the parameters of the drug candidate molecules means that after the parameters of the drug candidate molecules are determined, the target prediction information is verified to ensure that these candidate molecules can effectively act on the predicted targets, and then their potential as a treatment for T-AKI is verified. The verification process may include molecular docking, which evaluates the affinity between the drug molecule and the target by calculating and simulating the binding situation between the drug molecule and the target, and further confirms whether the drug can effectively bind and inhibit the target; biological verification, which verifies the effectiveness of the drug candidate molecule through in vitro (such as cell models) or in vivo (such as animal models) experiments in the laboratory, such as detecting the binding effect of the drug on the target and its alleviating effect on kidney injury; according to the results of the above verification process, the verification results of the drug candidate molecule are generated, including the target verification of the drug, verifying whether the drug can effectively bind to the predicted target and inhibit its function; evaluation of the therapeutic effect, how the drug treats trauma-related acute kidney injury, whether it can reduce kidney injury and improve renal function; finally, the effectiveness of the drug on the target is confirmed, thereby providing medical data support and decision-making.

[0034] A treatment effect score acquisition module 60, configured to perform treatment evaluation on the multiple target prediction information according to the verification result, obtain a treatment effect score, associate the treatment effect score with the multiple target prediction information, and generate a target prediction report.

[0035] Preferably, based on the verification result of the drug candidate molecule, the treatment effect of each target is further evaluated, and a detailed target prediction report is generated according to these evaluation results, and the treatment potential of each target and the effect of the corresponding drug candidate molecule are displayed. Specifically, the treatment evaluation of the multiple target prediction information according to the verification result generally includes whether the drug candidate molecule can effectively bind to the target and play a role, whether it can alleviate the pathological processes (such as renal tubular injury, inflammatory response, etc.) in trauma-related acute kidney injury (T-AKI), whether the target plays an important role in the pathological process, whether the drug improves renal function through the regulation of the target, and whether the drug candidate molecule can reduce injury, improve renal function or improve clinical symptoms by acting on the target, so as to obtain a treatment effect score, that is, a quantitative expression of the treatment effect of each target. For example, the drug is scored according to its effect on the target, such as the drug may be rated as effective, partially effective or ineffective, or the adaptability of the drug in clinical application, including patient acceptance, drug metabolism characteristics, etc.; then the treatment effect score is associated with the prediction information of each target as a target prediction report, which details the treatment effect, prediction information and possible drug candidate molecules of each target, and is used to continuously optimize the prediction model, help researchers understand the treatment potential of each target, and combine the treatment effect score of the target to select the most promising target and drug, so as to provide a scientific basis for targeted treatment.

[0036] A biomarker-based trauma-related acute kidney injury target prediction device according to an embodiment of the present invention is used to solve the technical problems of low accuracy in early diagnosis and treatment target prediction and low drug screening efficiency in the prior art, and achieves the technical effects of improving target prediction accuracy, optimizing the drug screening process and realizing personalized treatment. A biomarker-based trauma-related acute kidney injury target prediction device includes: a biomarker data set acquisition module 10, a treatment target information determination module 20, a biomarker feature acquisition module 30, a target prediction information acquisition module 40, a verification result generation module 50, and a treatment effect score acquisition module 60.

[0037] Next, the specific configuration of the biomarker feature acquisition module 30 will be described in detail. The biomarker feature acquisition module 30 may further include: calculating the correlation between the biomarker dataset and the therapeutic target information to generate multiple correlation coefficients; performing functional annotation on the therapeutic target information according to the multiple correlation coefficients to determine multiple target function annotation information; based on the multiple target function annotation information and combining the multiple correlation coefficients, performing acute kidney injury analysis on the biomarker dataset to determine multiple biomarker features; constructing a null two-dimensional matrix, converting the multiple target function annotation information into matrix columns, and converting the multiple biomarker features into matrix rows; filling the matrix columns and the matrix rows into the null two-dimensional matrix according to the multiple correlation coefficients to generate the feature relationship matrix.

[0038] Preferably, statistical methods (such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) are used to calculate the correlation between biomarkers and therapeutic targets, generating multiple correlation coefficients, which indicate the linear relationship between each biomarker and a specific therapeutic target. Among them, the correlation coefficient is a value between -1 and 1, representing the relationship strength and direction (positive correlation or negative correlation) between the two. For example, if the correlation coefficient of a certain biomarker and a target is close to 1, it indicates a strong positive correlation between them; if it is close to -1, it indicates a negative correlation; if it is close to 0, it indicates no obvious relationship between the two. According to the calculated correlation coefficients, the functions of each target are annotated, that is, the functional characteristics of the target are extracted according to the biological functions of the target (such as participating in inflammatory reactions, immune regulation, apoptosis, etc.), including annotating the role of the target in biological processes, related signal pathways, potential disease associations, etc., to determine multiple target function annotation information. Then, based on the target function annotation information and the correlation coefficients, acute kidney injury analysis is performed on the biomarker dataset, that is, analyzing which biomarkers are closely related to acute kidney injury, and further screening out the biomarker features closely associated with therapeutic targets or the kidney injury process.

[0039] Preferably, an empty two-dimensional matrix is constructed, converting the target function annotation information into matrix columns and the biomarker features into matrix rows. That is, the columns of the matrix represent the functional information of the therapeutic targets, and the rows represent the features of the biomarkers. The correlation coefficient serves as the element of the matrix, representing the relationship strength between the target and the biomarker. Then, according to the correlation coefficients (the relationship strength between the biomarker and the target), the elements in this two-dimensional matrix are filled to determine the feature relationship matrix, which contains the relationship information between the biomarker features and the target functions. The value of each matrix element (correlation coefficient) represents the similarity between the biomarker feature and the target function annotation. For example, the feature relationship matrix may be shown in the following table:

[0040]

[0041] Next, the specific configuration of the target prediction information acquisition module 40 will be described in detail. The target prediction information acquisition module 40 may further include: performing weighted processing with regularization based on the self-attention mechanism to construct a self-attention layer, where the input end of the self-attention layer is communicatively connected to the output end of the input layer in the deep learning model; classifying based on the feature relationship matrix to determine multiple subspaces, and using the self-attention mechanism to connect the multiple subspaces to construct a multi-head attention mechanism; retrieving the target prediction task, introducing a loss function for classification regression according to the target prediction task to construct a fully connected layer; evaluating the self-attention layer, the multi-head attention mechanism, and the fully connected layer using cross-validation, and constructing the deep learning model according to the evaluation results; where the output end of the self-attention layer is communicatively connected to the input end of the multi-head attention mechanism, the input end of the fully connected layer is communicatively connected to the output end of the multi-head attention mechanism and the output end of the input layer, and the output end of the fully connected layer is the output end of the deep learning model.

[0042] Preferably, the self-attention mechanism (Self-Attention Mechanism) dynamically adjusts the weights of input data (such as features, biomarker data, etc.) by calculating the relative importance between them, thereby helping the model to better focus on important features, and is used to calculate the relationships between different input features (such as biomarker features or target information). The regularization weighted processing controls the weights, reduces the network's dependence on the training data, enables the model to have better performance on new data, and helps the model to better generalize. The self-attention layer is constructed through the self-attention mechanism to process the features in the input data and weight them according to the relationships between them. The input end of the self-attention layer is connected to the input layer of the deep learning model to receive and process the input data; based on the feature relationship matrix, the model divides the input data (i.e., biomarker features) into multiple subspaces, each subspace contains a specific set of features, and the self-attention mechanism is used to connect the multiple subspaces to construct a multi-head attention mechanism, that is, multiple self-attention mechanisms are executed in parallel and their results are combined, so as to capture more patterns and information.

[0043] Preferably, the target prediction task is retrieved, that is, to predict the treatment targets related to acute kidney injury (AKI). According to the different target prediction tasks, a suitable loss function can be selected for classification and regression. For example, for the classification task, the cross-entropy loss function is used; for the regression task, the mean square error (MSE) loss function is used to evaluate the accuracy of target prediction. Depending on the task, it may be a classification task (predicting the target category) or a regression task (predicting the value or confidence of the target). A fully connected layer is constructed, which is usually used to integrate the outputs of different layers and make the final prediction, including the output ends of the self-attention layer and the multi-head attention mechanism, and finally generate the prediction result; cross-validation is used for evaluation, and a deep learning model is constructed according to the evaluation results, that is, the data set is divided into multiple subsets (such as 5-fold cross-validation), and the model is trained and evaluated on each subset to ensure that the model performs stably and has good generalization ability on different data subsets.

[0044] Preferably, the output end of the self-attention layer is connected to the input end of the multi-head attention mechanism; the input end of the fully connected layer is connected to the output end of the multi-head attention mechanism and the output end of the input layer. The model can utilize the information extracted from the input layer to the self-attention layer and the multi-head attention mechanism for further feature fusion to generate the final prediction result; the output end of the fully connected layer is the output end of the deep learning model, generating the final prediction result of the model, that is, the prediction of the target, and can perform target prediction accurately and efficiently.

[0045] Next, the specific configuration of the target prediction information acquisition module 40 will be described in detail. The target prediction information acquisition module 40 can further include: synchronizing the biomarker data set as the first vector input data to the input layer, and synchronizing the feature relationship matrix as the second vector input data to the input layer; using the self-attention layer to perform weighted calculation on the second vector input data to obtain multiple weight coefficients; through the multi-head attention mechanism, performing multi-dimensional capture on the first vector input data and the second vector input data according to the multiple weight coefficients to generate multi-space attention parameters; performing confidence scoring according to the multi-space attention parameters to generate multiple confidence scores; activating the fully connected layer according to the multiple confidence scores, and performing classification and regression prediction on the first vector input data and the second vector input data according to the fully connected layer to generate a classification and regression prediction result; matching the multiple confidence scores with the classification and regression prediction result to generate the multiple target prediction information.

[0046] Preferably, the biomarker dataset and the feature relationship matrix are synchronously input into the input layer as the first vector input data and the second vector input data respectively. The self-attention layer is used to weight the second vector input data (feature relationship matrix), calculate the weight coefficients of each feature (such as the correlation between the target and the biomarker), and obtain multiple weight coefficients. Then, the multi-head attention mechanism processes different parts of the input data in parallel through multiple independent heads. Each head can focus on different aspects or subspaces of the input data. Through the parallel calculation of multiple heads, the model can simultaneously focus on multiple dimensions in the biomarker data and the feature relationship matrix, capture the mutual relationships and complexities between different dimensions, and then generate multi-space attention parameters, which can describe the multi-dimensional relationships between biomarkers and different target functions.

[0047] Preferably, confidence scores are calculated based on the multi-space attention parameters, that is, confidence degree scores are calculated, which reflect the confidence level of the model in the prediction results of each target. The confidence degree score is a value between 0 and 1. The closer it is to 1, the higher the confidence of the model in predicting the target. A corresponding confidence degree score will be generated for each target, indicating the prediction confidence of the model for each target as a therapeutic target. Then, the fully connected layer is activated according to multiple confidence degree scores. The fully connected layer performs classification regression prediction, that is, predicts the therapeutic potential of each target based on the input biomarker features (the first vector) and target function information (the second vector). The goal of classification regression prediction is to output whether the target is an effective therapeutic target or the confidence degree score of the target, and then generate prediction results, including classification results, such as whether a certain target is an effective therapeutic target; regression results, the therapeutic effect score or confidence degree score of a certain target; finally, multiple confidence degree scores are matched and combined with the classification regression prediction results to generate the prediction information of the target. The prediction information of each target includes the target name and prediction category (for example, whether it is an effective therapeutic target), the confidence degree score of the target, that is, the confidence of the model in the therapeutic effect or effectiveness of the target, etc., as the final output result.

[0048] Next, the specific configuration of the verification result generation module 50 will be described in detail. The verification result generation module 50 may further include: sorting the multiple target prediction information in descending order according to the multiple confidence degree scores to generate a target prediction sequence; calculating the mean value according to the target prediction sequence to set a confidence degree threshold; comparing the multiple confidence degree scores with the confidence degree threshold to initially screen the multiple target prediction information, and storing the target prediction information corresponding to the confidence degree scores greater than or equal to the confidence degree threshold to generate an initial screening result; using the initial screening result as index information to perform a matching search on the drug database to generate a drug interaction relationship; and screening the drug database based on the drug interaction relationship to obtain the drug candidate molecule parameters.

[0049] Preferably, sorting multiple target prediction information in descending order according to multiple confidence scores means sorting the target prediction information in descending order according to the confidence score of each target (usually between 0 and 1). That is, the target prediction information with a higher confidence score will be ranked in the front, indicating a stronger relationship between the target and the biomarker and the therapeutic target. The accuracy of these target models is relatively high. After obtaining the target prediction sequence, the average value of the confidence scores of all targets is calculated to reflect the average confidence level of all target predictions, which is set as the confidence threshold to help remove those targets with low confidence and poor prediction accuracy. Compare the confidence score of each target with the set confidence threshold, and screen out the targets with a confidence score greater than or equal to the threshold to ensure that the target information with a high prediction confidence is retained, while the targets with low confidence are excluded as the initial screening result.

[0050] Preferably, input the initial screening result as index information into the drug database for matching and retrieval. By retrieving the drugs related to the targets in the initial screening result, the model will be able to find the drug molecules that have potential interaction relationships with these targets, and then generate a table of drug-target interaction relationships, that is, drug interactions, listing the drug molecules related to each target, such as target inhibitors, agonists, or drugs that can regulate the biological processes related to the target. Then, based on the drug interactions, further traverse the relevant drug molecules in the drug database to screen out the drugs most relevant to the targets. Specifically, match the screened high-confidence targets with the targets in the drug database, and through the target-drug interaction information of the drugs, screen out the drug candidate molecules that can bind to the targets and may have a therapeutic effect on TR-AKI, and then obtain the drug candidate molecule parameters, which may include chemical structure information, pharmacological parameters, and drug-target interactions, etc.

[0051] Next, the specific configuration of the verification result generation module 50 will be further described in detail. The verification result generation module 50 may further include: classifying the drug candidate molecules according to the drug interaction relationship to generate a plurality of to-be-experimented data, where the plurality of to-be-experimented data includes in vitro to-be-experimented data and in vivo to-be-experimented data; performing in vitro experiment analysis by combining the in vitro to-be-experimented data with the drug candidate molecules to obtain in vitro experiment impact effect parameters; performing in vivo experiment analysis by combining the in vivo to-be-experimented data with the drug candidate molecules to obtain in vivo experiment impact effect parameters; performing in vitro pharmacodynamic evaluation on the plurality of target prediction information based on the in vitro experiment impact effect parameters to generate an in vitro pharmacodynamic score; performing in vivo pharmacodynamic evaluation on the plurality of target prediction information based on the in vivo experiment impact effect parameters to generate an in vivo pharmacodynamic score; performing safety verification on the plurality of target prediction information by combining the in vitro pharmacodynamic score with the in vivo pharmacodynamic score to generate a target safety factor, and adding the target safety factor to the verification result.

[0052] Preferably, the drug interaction relationship indicates the potential effects (such as inhibition, activation, etc.) between the drug and the target. Classifying the drug candidate molecules according to the drug interaction relationship into a plurality of to-be-experimented data (in vitro to-be-experimented data, in vivo to-be-experimented data), where the in vitro to-be-experimented data refers to the experiments to be carried out on the drug candidate molecules, usually experiments carried out in cell, tissue or molecular models (for example, cell proliferation inhibition experiments, enzyme activity tests, etc.), and the in vivo to-be-experimented data refers to the experiments on the drug candidate molecules in animals or humans, usually experiments such as drug metabolism, pharmacodynamic evaluation, and toxicity tests carried out in mouse, rat or other animal models; performing in vitro experiment analysis by combining the in vitro to-be-experimented data with the drug candidate molecules, that is, performing in vitro experiments on the selected drug candidate molecules, using a cell culture system to verify the binding effect of the drug and the target, and evaluating the impact of the drug on TR-AKI-related cell types (such as renal tubular epithelial cells, endothelial cells, etc.). For example, the drug candidate molecules may be tested for their impact on cell proliferation, cell death, and the activity of specific proteins, etc., and finally obtain in vitro experiment impact effect parameters, such as IC50 value (drug half-maximal inhibitory concentration), EC50 value (effective concentration), etc., indicating the activity and effect of the drug in the in vitro environment.

[0053] Preferably, in vivo experimental analysis is carried out by combining in vivo experimental data to be tested with drug candidate molecules, usually in an animal model, that is, the efficacy of the drug is verified in the animal model, its improvement effect on trauma-related acute kidney injury is tested, and the effectiveness, safety and side effects of the drug are evaluated through drug administration tests to evaluate the pharmacokinetics (ADMET, such as absorption, distribution, metabolism, excretion, toxicity) and pharmacodynamics (such as anti-inflammatory, anti-tumor, antioxidant, etc.) of the drug, so as to obtain in vivo experimental impact effect parameters, such as plasma concentration curve, maximum drug concentration (Cmax), half-life (T1 / 2), bioavailability of the drug, etc. Then, based on the in vivo experimental impact effect parameters, in vivo pharmacodynamic evaluation is carried out on the multiple target prediction information, that is, by detecting biomarkers related to kidney function (such as blood urea nitrogen, etc.), the repair effect of the drug on acute kidney injury is evaluated. Note the distinction between the pharmacodynamic effect and the therapeutic effect here. The pharmacodynamic effect is the ideal effect of taking the drug, and the therapeutic effect is the actual effect after taking it; then, according to the in vitro pharmacodynamic score combined with the in vivo pharmacodynamic score, safety verification is carried out on the multiple target prediction information, that is, the in vitro pharmacodynamic score and the in vivo pharmacodynamic score are combined to evaluate the therapeutic effect and safety of the drug candidate molecule on the target, so as to obtain the target safety factor, which represents the safety of the drug to the target. Finally, the target safety factor is added to the verification result of the drug and comprehensively evaluated together with other experimental data of the drug (such as pharmacodynamic effect, drug-target interaction, etc.).

[0054] Next, the specific configuration of the treatment effect score acquisition module 60 will be described in detail. The treatment effect score acquisition module 60 may further include: setting a quality evaluation index based on the target safety factor; traversing the multiple target prediction information and performing treatment scoring according to the quality evaluation index to generate an in vivo treatment quality score and an in vitro treatment quality score; performing in vivo efficacy analysis according to the in vivo treatment quality score combined with the in vivo pharmacodynamic score to generate an in vivo treatment effect score; performing in vitro efficacy analysis according to the in vitro treatment quality score combined with the in vitro pharmacodynamic score to generate an in vitro treatment effect score; respectively calculating the in vivo treatment effect score and the in vitro treatment quality score with the multiple confidence scores of the multiple target prediction information to generate the treatment effect score.

[0055] Preferably, quality assessment indicators are set according to the target safety factor to measure the therapeutic potential of target prediction. Then, multiple target prediction information is traversed, and according to the set quality assessment indicators, the therapeutic potential of each target is scored, which may include a comprehensive evaluation of aspects such as the safety of the target, in vitro and in vivo pharmacodynamics, and toxicity. Furthermore, an in vivo treatment quality score and an in vitro treatment quality score are generated. Among them, the in vivo treatment quality score takes into account factors such as the effectiveness, safety, and pharmacokinetics of the drug in vivo, and the in vitro treatment quality score is scored based on in vitro experimental data to evaluate the effect and toxicity of the drug in vitro. Then, in vivo efficacy analysis is performed according to the in vivo treatment quality score combined with the in vivo pharmacodynamic score to generate an in vivo treatment effect score. For example, a five-point scale (0-5) or a percentage scoring system is used to represent the efficacy to generate the in vivo treatment effect score. For example, 0 indicates no effect, 1-2 indicates mild efficacy, 3-4 indicates moderate efficacy, and 5 indicates complete cure. Then, in vitro efficacy analysis is performed according to the in vitro treatment quality score combined with the in vitro pharmacodynamic score to generate an in vitro treatment effect score, reflecting the effect of the drug in in vitro experiments. The in vivo treatment effect score (since the in vivo experimental results are closer to clinical reality, a higher weight can be assigned), the in vitro treatment quality score (in vitro experiments provide preliminary data on the action of the drug, but due to its limitations, a moderate weight can be assigned), and multiple confidence scores of multiple target prediction information (according to the training results of the deep learning model, the confidence score has a greater impact on the credibility of the target, so a higher weight is assigned) are calculated to obtain a treatment effect score, indicating the overall effect and reliability of the drug in treating this target.

[0056] Next, the specific configuration of the treatment effect score obtaining module 60 will be further described in detail. The treatment effect score obtaining module 60 may further include:

[0057] Mapping analysis is performed according to the in vivo treatment effect score and multiple target prediction information, that is, through statistical or mathematical methods (such as regression analysis, clustering analysis, correlation analysis, etc.), to determine how each target performs in vivo according to the in vivo pharmacodynamic score, and then determine the first mapping relationship parameter, which reflects the relationship between the in vivo treatment effect score and each target, indicating the therapeutic potential or effect intensity of the target in vivo. Similarly, mapping analysis is performed according to the in vitro treatment effect score and multiple target prediction information to generate a second mapping relationship parameter, reflecting the performance of the target in in vitro experiments, which may describe the effectiveness of the target, the response to the drug, etc. Then, the mapping relationship parameters in vivo and in vitro (i.e., the first and second mapping relationship parameters) are combined with the confidence score of the target to construct a mapping relationship network, indicating the relationship between the treatment effects (in vivo and in vitro data) of different targets and the model's prediction information of the targets. In this way, the model can comprehensively evaluate the therapeutic potential of the targets from different data sources (in vivo and in vitro experimental data and prediction information).

[0058] Preferably, migrating learning is performed on multiple target prediction information according to the treatment effect score according to the mapping relationship network, which refers to using the treatment effect score (scores in vivo and in vitro) and the information in the mapping relationship network to optimize the learning of the target, that is, migrating the experimental data in vivo and in vitro (the structure constructed through the mapping relationship network) into the learning of the target to update the understanding of the target. Among them, migrating learning is a machine learning method that helps improve the learning efficiency in another related task by using the knowledge obtained in one task. The output is the target learning result, which represents information such as the treatment effect, potential, relevance, and confidence of each target. Then, the target learning result is fed back to the drug database for response, that is, according to the treatment effect of the target and the drug recommendation result, personalized treatment suggestions are put forward, such as drug screening, recommending which drugs are most likely to effectively treat the target, and then helping doctors select appropriate treatment plans, such as suggesting a certain treatment method or combination of drugs based on the characteristics and treatment effect of the target. These treatment suggestions are added to the target prediction report, which includes detailed information about each target, treatment effect, recommended drugs and their candidate molecules, experimental data support, etc., so as to efficiently screen out effective treatment targets and provide a scientific basis for drug development, clinical treatment plan formulation, and personalized treatment.

[0059] Although the present application makes various references to certain modules in the device according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The included individual units and modules are only divided according to functional logic, but are not limited to the above division as long as the corresponding functions can be achieved. In addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0060] The above specific implementation manners do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A biomarker-based prediction device for trauma-related acute kidney injury targets, characterized in that, The device includes: a biomarker dataset acquisition module for screening based on a public database to obtain a biomarker dataset for trauma-related acute kidney injury; a treatment target information determination module for calling a literature report library and combining with the public database to perform treatment analysis of trauma-related acute kidney injury to determine treatment target information; a biomarker feature acquisition module for performing feature analysis based on the biomarker dataset and the treatment target information to obtain multiple biomarker features, and constructing a feature relationship matrix according to the multiple biomarker features, including: performing correlation calculation according to the biomarker dataset and the treatment target information to generate multiple correlation coefficients; performing functional annotation on the treatment target information according to the multiple correlation coefficients to determine multiple target functional annotation information; performing acute kidney injury analysis on the biomarker dataset based on the multiple target functional annotation information and the multiple correlation coefficients to determine multiple biomarker features; constructing a null two-dimensional matrix, converting the multiple target functional annotation information into matrix columns, and converting the multiple biomarker features into matrix rows; filling the matrix columns and the matrix rows into the null two-dimensional matrix according to the multiple correlation coefficients to generate the feature relationship matrix; a target prediction information acquisition module for constructing a deep learning model based on a self-attention mechanism, synchronizing the biomarker dataset to the deep learning model according to the feature relationship matrix for prediction to obtain multiple target prediction information, and the multiple target prediction information includes multiple confidence scores; a verification result generation module for using a drug database and combining with the multiple confidence scores to perform matching screening on the multiple target prediction information to determine drug candidate molecule parameters, and verifying the multiple target prediction information according to the drug candidate molecule parameters to generate a verification result; The using a drug database and combining with the multiple confidence scores to perform matching screening on the multiple target prediction information to determine drug candidate molecule parameters includes: sorting the multiple target prediction information in descending order according to the multiple confidence scores to generate a target prediction sequence; performing mean calculation according to the target prediction sequence to set a confidence threshold; comparing the multiple confidence scores with the confidence threshold to perform initial screening on the multiple target prediction information, and storing the target prediction information corresponding to the confidence scores greater than or equal to the confidence threshold to generate an initial screening result; using the initial screening result as index information to perform matching retrieval on the drug database to generate a drug interaction relationship; screening by traversing the drug database based on the drug interaction relationship to obtain the drug candidate molecule parameters; a treatment effect score acquisition module for performing treatment evaluation on the multiple target prediction information according to the verification result to obtain a treatment effect score, and associating the treatment effect score with the multiple target prediction information to generate a target prediction report.

2. The biomarker-based traumatic acute kidney injury target prediction device according to claim 1, wherein Constructing a deep learning model based on a self-attention mechanism includes: Based on the self-attention mechanism, regularization is used for weighted processing to construct a self-attention layer, and the input end of the self-attention layer is communicatively connected to the output end of the input layer in the deep learning model; Based on the feature relationship matrix for classification, multiple subspaces are determined, and the self-attention mechanism is used to connect the multiple subspaces to construct a multi-head attention mechanism; Retrieve the target prediction task, introduce a loss function for classification regression according to the target prediction task, and construct a fully connected layer; Evaluate the self-attention layer, the multi-head attention mechanism, and the fully connected layer using cross-validation, and construct the deep learning model according to the evaluation results; Among them, the output end of the self-attention layer is communicatively connected to the input end of the multi-head attention mechanism, the input end of the fully connected layer is communicatively connected to the output end of the multi-head attention mechanism and the output end of the input layer, and the output end of the fully connected layer is the output end of the deep learning model.

3. The biomarker-based prediction device for trauma-related acute kidney injury targets according to claim 2, wherein, The step of synchronizing the biomarker dataset to the deep learning model according to the feature relationship matrix for prediction to obtain multiple target prediction information includes: Synchronize the biomarker dataset as the first vector input data to the input layer, and synchronize the feature relationship matrix as the second vector input data to the input layer; Use the self-attention layer to perform weighted calculation on the second vector input data to obtain multiple weight coefficients; Through the multi-head attention mechanism, perform multi-dimensional capture on the first vector input data and the second vector input data according to the multiple weight coefficients to generate multi-space attention parameters; Perform confidence scoring according to the multi-space attention parameters to generate multiple confidence scores; Activate the fully connected layer according to the multiple confidence scores, and perform classification regression prediction on the first vector input data and the second vector input data according to the fully connected layer to generate a classification regression prediction result; Match the multiple confidence scores with the classification regression prediction result to generate the multiple target prediction information.

4. The biomarker-based prediction device for trauma-related acute kidney injury targets according to claim 1, characterized in that Verify the multiple target prediction information according to the drug candidate molecule parameters to generate a verification result, including: Classify the drug candidate molecules according to the drug interaction relationship to generate multiple data to be experimented, and the multiple data to be experimented include in vitro data to be experimented and in vivo data to be experimented; Based on the in vitro data to be experimented, combine with the drug candidate molecules for in vitro experiment analysis to obtain in vitro experiment influence effect parameters; Based on the in vivo data to be experimented, combine with the drug candidate molecules for in vivo experiment analysis to obtain in vivo experiment influence effect parameters; Based on the in vitro experiment influence effect parameters, perform in vitro efficacy evaluation on the multiple target prediction information to generate in vitro efficacy scores; Based on the in vivo experiment influence effect parameters, perform in vivo efficacy evaluation on the multiple target prediction information to generate in vivo efficacy scores; According to the in vitro efficacy score combined with the in vivo efficacy score, perform safety verification on the multiple target prediction information to generate a target safety coefficient, and add the target safety coefficient to the verification result.

5. The biomarker-based traumatic acute kidney injury target prediction device according to claim 4, wherein Performing a treatment evaluation on the multiple target prediction information according to the verification result to obtain a treatment effect score, including: Setting a quality evaluation index based on the target safety factor; Traversing the multiple target prediction information and performing a treatment score according to the quality evaluation index to generate an in-vivo treatment quality score and an in-vitro treatment quality score; Performing an in-vivo efficacy analysis based on the in-vivo treatment quality score in combination with the in-vivo pharmacodynamic score to generate an in-vivo treatment effect score; Performing an in-vitro efficacy analysis based on the in-vitro treatment quality score in combination with the in-vitro pharmacodynamic score to generate an in-vitro treatment effect score; Calculating the in-vivo treatment effect score and the in-vitro treatment quality score respectively with the multiple confidence scores of the multiple target prediction information to generate the treatment effect score; 6. The biomarker-based prediction device for trauma-related acute kidney injury targets according to claim 5, wherein Associating the treatment effect score with the multiple target prediction information to generate a target prediction report, including: Performing a mapping analysis based on the in-vivo treatment effect score and the multiple target prediction information to generate a first mapping relationship parameter; Performing a mapping analysis based on the in-vitro treatment effect score and the multiple target prediction information to generate a second mapping relationship parameter; Constructing a mapping relationship network according to the first mapping relationship parameter in combination with the second mapping relationship parameter according to the multiple confidence scores; Performing transfer learning on the multiple target prediction information based on the treatment effect score according to the mapping relationship network to generate a target learning result; Feeding back the target learning result to the drug database for response to generate a treatment suggestion, and adding the treatment suggestion to the target prediction report.

Images