Drug combination data processing method and system based on organ adverse reaction

Abstract

The invention relates to a drug combination data processing method and system based on organ adverse reactions, and the method comprises the steps: carrying out the graph neural network fusion of multi-source patient medical data, and carrying out the cross-scale integration of organ function parameters, metabolic characteristics and hemodynamic data; extracting a core biomarker by utilizing principal component analysis and organ specificity screening, and generating a feature vector representing organ reaction; constructing a nonlinear drug interaction model by combining a dose-time sequence and in-vitro experimental data, and quantifying the strength of synergy and antagonism; gene expression profiles, pathological states and lifestyle data are introduced for triple dynamic calibration, and individualized adverse reaction probability optimization is achieved; monte Carlo sampling is adopted to simulate an organ toxicity scene, and intelligent classification decision is performed on a high-risk combination scheme based on a supervision rule base. The method breaks through the limitation of data islands, static evaluation and individualization deficiency in the traditional technology, and significantly improves the prediction precision and clinical applicability of the drug combination safety.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a method and system for processing drug combination data based on adverse organ reactions. Background Technology

[0002] With the development of precision medicine, the application of drug combination therapy in the treatment of complex diseases is becoming increasingly widespread. Drug combination data processing methods aim to assess the potential adverse reaction risks to human organs by analyzing the interaction effects of multiple drugs used simultaneously. These methods, by integrating patient physiological characteristics, pharmacokinetic data, and external environmental factors, construct predictive models to guide clinical medication decisions, and have become a key technical means to avoid drug-induced organ damage.

[0003] However, existing technologies have significant limitations: First, traditional methods struggle to effectively integrate heterogeneous medical data from multiple sources, such as organ function indicators, metabolic characteristics, and hemodynamic parameters, leading to prominent data silos and hindering cross-scale correlation analysis. Second, when assessing the risks of drug combination therapy, existing models lack dynamic individualized calibration capabilities, failing to comprehensively consider the impact of gene expression differences, pathological state evolution, and lifestyle changes on drug responses. Finally, current risk decision-making mechanisms largely rely on static threshold judgments, which cannot adapt to the dynamic changes in organ toxicity under complex dosage scenarios and are difficult to match with updated regulatory compliance standards in real time. These deficiencies result in insufficient accuracy in predicting adverse reactions to drug combination therapy regimens in clinical practice, particularly creating significant blind spots in the protective assessment of vulnerable organs such as the liver and kidneys. Summary of the Invention

[0004] Based on this, the purpose of this invention is to provide a method and system for processing drug combination data based on organ adverse reactions that can achieve deep integration of multi-source data, dynamic individualized calibration, and intelligent compliance decision-making.

[0005] The objective of this invention is achieved through the following solution:

[0006] In a first aspect, the present invention provides a method for processing drug combination data based on adverse organ reactions, comprising the following steps:

[0007] S1: Integrate and process the acquired multi-source patient medical data, fuse organ function parameters, metabolic characteristics and hemodynamic data through graph neural networks, perform unit unification transformation and construct a multi-scale data association matrix to generate a medical dataset in a unified format;

[0008] S2: Extract core features from the medical dataset, use principal component analysis to calculate the contribution weight of each feature vector, and screen key biomarkers that identify organs to generate a set of dimensionality-reduced feature vectors.

[0009] S3: Model drug interaction effects based on a set of dimensionality-reduced feature vectors, integrate dose-time series and in vitro experimental data to construct a nonlinear relationship model, calculate the strength of drug synergy through the random forest algorithm, and generate predicted values ​​of interaction effects.

[0010] S4: Perform individualized dynamic calibration on the predicted value of the interaction effect, adjust the prediction parameters by combining gene expression profile and pathological state data, introduce the acquired patient life data for probability distribution optimization and update, and generate individualized drug response prediction values.

[0011] S5: Based on individualized drug response prediction values, risk decision processing is performed. Monte Carlo sampling simulation of organ toxicity probability distribution is called. High-risk combinations are classified and labeled by flow cytometry according to the preset regulatory rule base, and compliance risk assessment results are generated. The compliance risk assessment results are used to indicate the safety level and clinical applicability of drug combination regimens.

[0012] In one embodiment, S1 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0013] S11: Perform data cleaning on the organ function parameters, metabolic characteristics and hemodynamic data of the acquired multi-source patient medical data, remove missing values ​​and outlier data points in the original data, and generate cleaned medical data.

[0014] S12: Perform unit standardization conversion on the cleaned medical data, converting data with different units of measurement into a preset standard unit system to generate standard medical multi-source data;

[0015] S13: By using graph neural networks to fuse standard medical multi-source data, establish the correlation between organ function and metabolic characteristics, and generate a medical dataset in a unified format. The medical dataset is used to indicate the integrated correlation results of cross-scale biomedical data.

[0016] In one embodiment, step S2 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0017] S21: Perform principal component analysis on the medical dataset to calculate the variance contribution of organ function indicators, metabolic activity parameters, and hemodynamic feature vectors, and generate a ranking of feature importance.

[0018] S22: Perform organ-specific screening on the feature importance ranking, assign screening weights to liver, kidney and heart-related features according to organ type, and generate a key biomarker identifier set;

[0019] S23: Dimensionality reduction is performed on the key biomarker identifier set, and high-contribution feature vectors in the biomarker set are retained to generate a dimensionality-reduced feature vector set. The dimensionality-reduced feature vector set is used to indicate the core feature combination most relevant to adverse organ reactions.

[0020] In one embodiment, step S3 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0021] S31: Integrate drug features into the dimensionality-reduced feature vector set, and construct a joint feature space of organ response features and drug characteristic parameters by combining dose-time series data and in vitro experimental data to generate a drug interaction feature set;

[0022] S32: Model the drug interaction feature set using a random forest, train multiple decision trees and calculate the optimal feature selection criterion for node splitting to generate a nonlinear relationship model;

[0023] S33: The drug interaction feature set is predicted and processed by a nonlinear relationship model to calculate the synergistic and antagonistic effects of the drug combination and generate interaction effect prediction values. The interaction effect prediction values ​​are used to indicate the actual impact of the drug combination regimen on organ adverse reactions.

[0024] In one embodiment, step S4 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0025] S41: Gene expression profile calibration is performed on the predicted values ​​of interaction effects. Drug clearance rate and volume of distribution parameters are adjusted according to the variations of drug metabolism-related genes to generate preliminary calibration values.

[0026] S42: Correct the initial calibration values ​​for pathological conditions, update the drug sensitivity and toxicity threshold parameters in combination with organ function grading indicators, and generate intermediate predicted values.

[0027] S43: Obtain patient lifestyle data, optimize lifestyle by combining intermediate predictive values, adjust the adverse reaction probability of intermediate predictive values ​​by introducing factors influencing lifestyle habits, and generate individualized drug response predictive values. Individualized drug response predictive values ​​are used to indicate the intensity of organ adverse reaction risk for a specific patient using a combination of drugs.

[0028] In one embodiment, the adjustment formula for the adverse reaction probability of a drug combination data processing method based on organ adverse reactions provided by the present invention is as follows:

[0029]

[0030] Among them, P final P represents the adjusted probability of adverse reactions.inter The intermediate predicted value is the pathology-corrected value, λ is the attenuation coefficient of lifestyle influence, and w k f represents the weight of the kth lifestyle factor. k Let m be the quantified value of the kth lifestyle factor, and m be the number of lifestyle factor types.

[0031] In one embodiment, step S5 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0032] S51: Perform Monte Carlo sampling on individualized drug response prediction values ​​to simulate organ toxicity responses under different dosage scenarios and generate toxicity probability distributions.

[0033] S52: Extract high-risk combinations from the toxicity probability distribution, screen for therapeutic drug combinations that exceed the preset organ toxicity threshold, and generate a high-risk drug combination set.

[0034] S53: Perform compliance classification processing on high-risk drug combination sets, match them with safety standards in the preset regulatory rule base, and generate compliance risk assessment results.

[0035] Secondly, the present invention provides a drug combination data processing system based on organ adverse reactions, the system being configured with the following modules:

[0036] The multi-source medical data integration module is used to integrate and process the acquired multi-source patient medical data. It fuses organ function parameters, metabolic characteristics and hemodynamic data through graph neural networks, performs unit unification transformation and constructs a multi-scale data association matrix to generate a medical dataset in a unified format.

[0037] The core feature extraction and dimensionality reduction module is used to extract core features from medical datasets, call principal component analysis to calculate the contribution weight of each feature vector, and screen key biomarkers that identify organs to generate a set of dimensionality-reduced feature vectors.

[0038] The drug interaction effect modeling module is used to model drug interaction effects based on a set of dimensionality-reduced feature vectors, integrate dose-time series and in vitro experimental data to construct a nonlinear relationship model, calculate the strength of drug synergy through the random forest algorithm, and generate predicted values ​​of interaction effects.

[0039] The individualized prediction calibration module is used to perform individualized dynamic calibration of the interaction effect prediction value. It adjusts the prediction parameters by combining gene expression profile and pathological state data, and introduces the acquired patient life data for probability distribution optimization and update to generate individualized drug response prediction values.

[0040] The risk decision assessment module is used to make risk decisions based on individualized drug response predictions. It calls Monte Carlo sampling to simulate organ toxicity probability distribution, performs flow cytometry classification and labeling of high-risk combinations according to a preset regulatory rule base, and generates compliance risk assessment results. The compliance risk assessment results are used to indicate the safety level and clinical applicability of drug combination regimens.

[0041] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned methods for processing drug combination data based on adverse organ reactions.

[0042] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described methods for processing drug combination data based on adverse organ reactions.

[0043] In summary, the drug combination data processing method based on organ adverse reactions provided in this application, based on the core feature extraction mechanism of principal component analysis and organ-specific screening, can achieve efficient dimensionality reduction and accurate identification of key biomarkers, providing a high-value feature foundation for drug interaction effect modeling. By integrating dose-time series and in vitro experimental data to construct a nonlinear relationship model, it can achieve dynamic quantitative prediction of drug synergistic / antagonistic effects, overcoming the limitations of static assessment in adapting to complex dosage scenarios. Combining an individualized dynamic calibration process with gene expression profiles, pathological states, and lifestyle data, it can achieve probability distribution optimization and update through the linkage of three factors, thereby solving the prediction bias problem caused by individual differences. By employing Monte Carlo sampling simulation and streaming classification decision-making based on regulatory rule bases, it can achieve dynamic scenario simulation of organ toxicity risk and intelligent compliance judgment, significantly improving the safety classification accuracy and clinical decision reliability of drug combination regimens.

[0044] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0045] Figure 1 A flowchart illustrating a method for processing drug combination data based on organ adverse reactions, provided in an embodiment of this application;

[0046] Figure 2 A schematic diagram of the process for generating personalized drug response prediction values ​​provided in an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of a drug combination data processing system based on organ adverse reactions, provided as another embodiment of this application. Detailed Implementation

[0048] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0050] In one embodiment, such as Figure 1 As shown, a method for processing drug combination data based on organ adverse reactions is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0051] S1: Integrate and process the acquired multi-source patient medical data, fuse organ function parameters, metabolic characteristics and hemodynamic data through graph neural networks, perform unit unification transformation and construct a multi-scale data association matrix to generate a medical dataset in a unified format.

[0052] Specifically, the system acquires multi-source patient medical data, including clinical diagnostic information from electronic medical records, organ function parameters such as liver and kidney function obtained from laboratory tests, metabolic characteristic data such as the concentration of metabolites in blood and urine, and hemodynamic data such as heart rate and blood pressure collected by monitoring instruments. The system uses a graph neural network for data fusion, abstracting various types of data into different types of nodes. Organ function parameter nodes, metabolic characteristic nodes, and hemodynamic data nodes correspond to different data attributes.

[0053] The system constructs connections between nodes based on physiological and pathological knowledge in the medical field, linking glomerular filtration rate (GFR) with serum creatinine concentration. Through a graph neural network message passing mechanism, it enables the interaction and fusion of features from different nodes, mapping multi-source heterogeneous data to a unified high-dimensional feature space. Preferably, the system performs standardized unit conversion, adhering to internationally accepted medical measurement unit standards, converting blood drug concentrations to μmol / L and blood pressure values ​​to mmHg. The system constructs a multi-scale data association matrix, based on the hierarchical relationships of human physiological structure. From molecular-level metabolic characteristics and cellular-level functional indicators to system-level hemodynamic data, it establishes cross-scale associations based on anatomical and physiological logic, generating a unified format medical dataset containing basic patient information, standardized data, and multi-scale associations.

[0054] S2: Extract core features from the medical dataset, use principal component analysis to calculate the contribution weight of each feature vector, and screen key biomarkers that identify organs to generate a set of dimensionality-reduced feature vectors.

[0055] Specifically, the system performs core feature extraction on the generated medical dataset and uses principal component analysis to calculate the covariance matrix of the dataset, reflecting the correlation between features. The system solves for the eigenvalues ​​and eigenvectors of the covariance matrix through eigenvalue decomposition or singular value decomposition, where each eigenvalue represents the variance explained by the corresponding principal component. The system calculates the contribution weight of each eigenvector and filters key biomarkers for identifying organs. A contribution weight threshold is set, and the original features corresponding to principal components with contribution weights greater than this threshold are selected as key biomarkers. For the liver, features such as alanine aminotransferase (ALT) and aspartate aminotransferase (AST) are selected. The system arranges the eigenvalues ​​corresponding to the selected key biomarkers in order, forming a dimensionality-reduced feature vector set. This set has a lower dimensionality than the original dataset, reducing data redundancy and retaining information that plays a crucial role in assessing organ adverse reactions (OrganADR).

[0056] S3: Model drug interaction effects based on a set of dimensionality-reduced feature vectors, integrate dose-time series and in vitro experimental data to construct a nonlinear relationship model, calculate the strength of drug synergy through the random forest algorithm, and generate predicted values ​​of interaction effects.

[0057] Specifically, the system constructs a drug interaction effect model based on a set of dimensionality-reduced feature vectors, integrating dose-time series and in vitro experimental data. The dose-time series data includes drug doses at different administration time points and corresponding in vivo drug concentration monitoring values, while the in vitro experimental data covers the interaction detection results of different drug combinations in cell lines or tissue models, including cell viability and the amount of toxic products generated.

[0058] Preferably, the system performs linear interpolation on the dose-time series to unify the time intervals, and then concatenates it with features from in vitro experimental data to form the model input. The system constructs a nonlinear relationship model using a neural network based on radial basis functions. It optimizes network weights and biases through backpropagation to fit the complex nonlinear relationship between dose, time, and interaction effects during drug combination therapy. The system calls the random forest algorithm to construct 100 decision trees. Each decision tree is trained using bootstrap sampling, selecting samples from the training set. At each node split, a subset of features is randomly selected to determine the optimal split point. The system votes on the predicted drug synergy strength from all decision trees, takes the average to obtain the final synergy strength, and generates a predicted interaction effect value. This value is a continuous variable reflecting the potential impact of drug combination therapy on OrganADR.

[0059] S4: Perform individualized dynamic calibration on the predicted values ​​of interaction effects, adjust the prediction parameters by combining gene expression profiles and pathological state data, introduce the acquired patient life data for probability distribution optimization and update, and generate individualized drug response prediction values.

[0060] Specifically, the system acquires gene expression profile data and pathological status data. The gene expression profile data includes the gene expression levels of enzymes and transporters related to drug metabolism, while the pathological status data covers the patient's disease stage, complication type, and severity. Based on the gene expression profile data, the system determines the patient's drug metabolism type; patients with high CYP3A4 expression correspond to a higher drug metabolism rate, and the system adjusts the drug clearance parameter in the model accordingly. Combining the pathological status data, the system determines that patients with renal failure require a reduced metabolic rate parameter for drugs excreted through the kidneys, and adjusts the prediction parameters accordingly.

[0061] The system acquires patients' lifestyle data, including smoking history, alcohol consumption, dietary habits, and exercise frequency. After quantifying these data into numerical features, a Bayesian update algorithm is introduced. Using lifestyle data as prior information, the probability distribution of the predicted interaction effect is optimized and updated. The mean of the predicted probability distribution of hepatotoxicity for smoking patients is adjusted upward. The system ultimately generates an individualized drug response prediction value that fits the actual situation of a specific patient.

[0062] S5: Based on individualized drug response prediction values, risk decision processing is performed. Monte Carlo sampling simulation of organ toxicity probability distribution is called. High-risk combinations are classified and labeled by flow cytometry according to the preset regulatory rule base, and compliance risk assessment results are generated. The compliance risk assessment results are used to indicate the safety level and clinical applicability of drug combination regimens.

[0063] Specifically, the system performs risk decision-making based on individualized drug response predictions. The system uses Monte Carlo sampling to simulate organ toxicity probability distributions, setting the probability distribution types and parameters related to organ toxicity. Drug concentrations follow a normal distribution, and metabolite concentrations follow a log-normal distribution. The system performs random sampling based on a preset number of times, substituting parameter values ​​into the organ toxicity response model for each sample to calculate the corresponding toxicity value. Statistical analysis is then used to obtain the organ toxicity probability distribution. The system also calls upon a preset regulatory rule base, which contains rules regarding drug combination safety standards, contraindications, and dosage limits established by drug regulatory agencies in various countries.

[0064] Based on the organ toxicity probability distribution, the system marks drug combinations exceeding the safety threshold in the rule base as high-risk, and categorizes the remaining combinations into medium-risk and low-risk based on their toxicity probability. The system generates compliance risk assessment results, including the safety level of the drug combination regimen and corresponding clinical applicability recommendations. High-risk regimens are advised to be avoided, while medium-risk regimens require enhanced organ function monitoring, providing a basis for clinical medication decisions.

[0065] In summary, the drug combination data processing method based on organ adverse reactions provided in this application, based on the core feature extraction mechanism of principal component analysis and organ-specific screening, can achieve efficient dimensionality reduction and accurate identification of key biomarkers, providing a high-value feature foundation for drug interaction effect modeling. By integrating dose-time series and in vitro experimental data to construct a nonlinear relationship model, it can achieve dynamic quantitative prediction of drug synergistic / antagonistic effects, overcoming the limitations of static assessment in adapting to complex dosage scenarios. Combining an individualized dynamic calibration process with gene expression profiles, pathological states, and lifestyle data, it can achieve probability distribution optimization and update through the linkage of three factors, thereby solving the prediction bias problem caused by individual differences. By employing Monte Carlo sampling simulation and streaming classification decision-making based on regulatory rule bases, it can achieve dynamic scenario simulation of organ toxicity risk and intelligent compliance judgment, significantly improving the safety classification accuracy and clinical decision reliability of drug combination regimens.

[0066] The technical solution provided in this application utilizes graph neural network fusion processing of multi-source medical data to achieve cross-scale deep integration of organ function parameters, metabolic characteristics, and hemodynamic data, thereby addressing the lack of correlation analysis caused by data silos in traditional methods. This method systematically overcomes the shortcomings of existing technologies, such as data fragmentation, static modeling, and lack of individualization, forming a complete chain optimization from data integration → feature extraction → interactive modeling → dynamic calibration → risk decision-making. Ultimately, it achieves a synergistic improvement in the accuracy, clinical applicability, and regulatory compliance of OrganADR prediction, providing core technical support for precision medicine.

[0067] In one embodiment, S1 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0068] S11: Perform data cleaning processing on the organ function parameters, metabolic characteristic data and hemodynamic data of the acquired multi-source patient medical data, process the missing values ​​and abnormal data points in the original data, and generate cleaned medical data.

[0069] Specifically, the system traverses the original data using a missing value detection algorithm based on field integrity verification, scanning data records row by row to identify the location of missing values ​​and their corresponding sample numbers in each field. For continuous data, the system calculates the central tendency of all non-missing values ​​in the field and fills in the missing values; the central tendency is determined by statistically analyzing the distribution characteristics of the valid data in the field. For discrete data, the system counts the frequency of each categorical value in the field and selects the categorical value with the highest frequency to fill in the missing items. For time series data, the system extracts valid data points before and after the missing points, constructs a linear function based on the time interval, and calculates the imputed values ​​for the missing points using the function.

[0070] Preferably, the system can use a standardized deviation algorithm to calculate the deviation of each data point from the mean of its field. When the deviation exceeds a preset range, it is marked as a candidate outlier. Simultaneously, the system can use the interquartile range method to perform secondary verification on the candidate outliers, calculating the quartile values ​​of the field data to determine the normal data range. Data points exceeding this range are confirmed as outliers. For confirmed outliers, the system retrieves historical data sequences of similar tests for the same patient. If the outlier occurs in isolation, the system calculates the average of a preset number of valid data points before and after the outlier and replaces it. If outliers occur consecutively, the system adds a quality flag to the data, triggering a manual review process. The system generates cleaned medical data, including processed organ function parameters, metabolic characteristic data, and hemodynamic data. The data retains the original test timestamps, data acquisition device identifiers, and cleaning operation records.

[0071] S12: Perform unit standardization conversion on the cleaned medical data, converting data with different units of measurement into a preset standard unit system to generate standard medical multi-source data.

[0072] Specifically, the system calls a preset standard unit system, which is based on internationally accepted medical metrology units and covers basic and derived units in fields such as human physiological indicators, metabolite detection, and hemodynamic monitoring. It includes standard units for basic quantities such as length, mass, time, and amount of substance, as well as definitions for derived units such as concentration, pressure, and enzyme activity.

[0073] For example, the system identifies the unit of measurement information for each test item in the cleaned medical data through a field parsing mechanism and compares it with the standard unit system. For units of measurement with conversion relationships, the system calls its internal conversion formula library to perform the conversion. This library is built based on international medical testing unit conversion standards and includes unit conversion logic for various commonly used clinical tests. For non-physical quantity units, the system performs standardized conversion through a terminology mapping table. This mapping table is built based on an authoritative medical terminology system and includes the correspondence between medical testing terms and standardized codes, covering data types such as blood typing and imaging examination descriptions. After the conversion is complete, the system adds a "unit conversion identifier" field to the data, recording the original unit of measurement, the target unit of measurement, and the conversion process information, thereby generating standard medical multi-source data.

[0074] S13: By using graph neural networks to fuse standard medical multi-source data, establish the correlation between organ function and metabolic characteristics, and generate a medical dataset in a unified format. The medical dataset is used to indicate the integrated correlation results of cross-scale biomedical data.

[0075] Specifically, the system constructs a graph neural network model, mapping organ function parameters, metabolic characteristic data, and hemodynamic data from standard medical multi-source data into three types of nodes. Organ function nodes include parameter sets such as liver and kidney function indicators and myocardial enzyme profiles; metabolic characteristic nodes include parameter sets such as glucose metabolism and lipid metabolism metabolite concentrations; and hemodynamic nodes include parameter sets such as vascular pressure and flow rate. The system defines the connections between nodes based on a physiological and pathological association rule base: between organ function nodes and metabolic characteristic nodes, connections are established based on organ metabolic pathways, such as connecting parameters reflecting liver metabolic function with parameters of liver synthetic metabolites; between metabolic characteristic nodes and hemodynamic nodes, connections are established based on the mechanism by which metabolites affect the circulatory system; and between organ function nodes and hemodynamic nodes, connections are established based on the association mechanism between organ perfusion and functional status.

[0076] The system inputs node feature vectors into a graph neural network and uses the GraphSAGE (Graph Sample and Aggregate) algorithm to sample the neighboring nodes of each node, aggregating the features of neighboring nodes with the node's own features to generate an updated node embedding vector. The system optimizes model parameters using a loss function and adjusts network weights through backpropagation until the feature association error on the validation set is within acceptable limits. After training, the system outputs a fused node embedding matrix, associates the original parameter values ​​with the patient's unique identifier, and generates a standardized medical dataset. This dataset includes a node association strength matrix, cross-scale feature mapping relationships, and original data indices, used to present the integrated association results between organ function, metabolic characteristics, and hemodynamic data.

[0077] In one embodiment, step S2 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0078] S21: Perform principal component analysis on the medical dataset to calculate the variance contribution of organ function indicators, metabolic activity parameters, and hemodynamic feature vectors, and generate a ranking of feature importance.

[0079] Specifically, the system calculates the covariance matrix of three types of eigenvectors, reflecting the degree of linear correlation between different features. The system then performs eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors. Here, the eigenvalues ​​represent the total amount of information carried by the corresponding principal components, while the eigenvectors reflect the linear combination relationship between the principal components and the original features.

[0080] The system arranges the feature values ​​in descending order and calculates the proportion of each feature value to the total number of feature values. This proportion represents the variance contribution of the corresponding principal component. Based on the variance contribution, the system determines the loading coefficients of each original feature in the principal components. The absolute value of the loading coefficients reflects the degree of influence of the original features on the principal components. The system calculates the comprehensive contribution value of each original feature based on the product of the loading coefficients and the variance contribution. It then sorts the organ function indicators, metabolic activity parameters, and hemodynamic feature vectors from highest to lowest comprehensive contribution value, generating a feature importance ranking.

[0081] S22: Perform organ-specific screening based on the importance ranking of features, assign screening weights to liver, kidney and heart-related features according to organ type, and generate a set of key biomarkers.

[0082] Specifically, the system incorporates a mapping table between organ types and features. This table is constructed based on the correspondence between organs and biomarkers in medical literature and includes feature entries related to the liver, kidneys, and heart. The system assigns screening weights to different organs, based on their vulnerability to adverse drug reactions and their importance in clinical research. The system matches features ranked by importance against the mapping table to determine the organ category to which each feature belongs and calculates the weighted contribution value of the feature. This weighted contribution value is the product of the feature's overall contribution value and the corresponding organ screening weight. The system sets a screening threshold, retaining features with weighted contribution values ​​higher than the threshold and labeling these features as organ-specific biomarkers. The system integrates all labeling results to generate a key biomarker identifier set, which includes the organ type and weighted contribution value corresponding to each biomarker.

[0083] S23: Dimensionality reduction is performed on the key biomarker identifier set, retaining the high-contribution feature vectors in the biomarker set to generate a dimensionality-reduced feature vector set. The dimensionality-reduced feature vector set is used to indicate the core feature combination most relevant to OrganADR.

[0084] Specifically, the system performs dimensionality reduction on the key biomarker identifier set, extracts the contribution value of each feature vector in the key biomarker identifier set, and calculates the contribution value based on the variance contribution in principal component analysis and the weighted score in organ-specific screening. The system sorts the feature vectors in the biomarker set from high to low according to the contribution value and calculates the cumulative contribution, which is the sum of the contributions accumulated from the feature with the highest contribution value.

[0085] The system sets a cumulative contribution threshold. When the total accumulated contribution reaches this threshold, subsequent feature vectors are no longer included, thus retaining high-contribution feature vectors. For feature vectors with the same contribution value, the system performs a secondary sorting based on their known association records with OrganADR, prioritizing feature vectors with clearly defined association records. The system generates a dimensionality-reduced feature vector set, which contains the selected feature vectors, their corresponding contribution values, and their respective organ category identifiers. This set is stored in matrix form, with each row corresponding to one feature vector and columns containing the feature identifier, contribution value, and associated organ information. This dimensionality-reduced feature vector set is used for extracting core features related to OrganADR in subsequent drug interaction effect modeling.

[0086] In one embodiment, step S3 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0087] S31: Integrate drug features into the dimensionality-reduced feature vector set, and construct a joint feature space of organ response features and drug characteristic parameters by combining dose-time series data and in vitro experimental data, thereby generating a drug interaction feature set.

[0088] Specifically, the system performs drug feature integration on the dimensionality-reduced feature vector set, extracting organ response features from the set. These organ response features include core feature vectors related to the liver, kidneys, and heart, along with their contribution information. The system imports dose-time series data, which includes time-series parameters such as single-dose dosage, daily dosing frequency, duration of continuous dosing, and dosing interval for each drug. Simultaneously, the system loads in vitro experimental data, covering experimental results such as cell viability, enzyme inhibition rate, transporter binding rate, and metabolite formation rate at different drug concentrations.

[0089] For example, the system fuses organ response features with drug characteristic parameters using a feature concatenation algorithm. These drug characteristic parameters include dynamic change parameters from dose-time series data and physicochemical properties from in vitro experimental data. The system constructs a joint feature space, the dimensions of which are the sum of the organ response feature dimension and the drug characteristic parameter dimension. Each dimension corresponds to a specific feature variable, and the variable values, after standardization, fall within the [0,1] interval. The system generates a drug interaction feature set, stored in the form of a three-dimensional tensor. The first dimension corresponds to the patient sample number, the second dimension to the feature variable identifier, and the third dimension to the time node. The tensor elements are the standardized values ​​of the feature variables at specific time nodes. The feature set contains cross-correlation information between organ responses and drug characteristics.

[0090] S32: Model a random forest on the drug interaction feature set, train multiple decision trees and calculate the optimal feature selection criterion for node splitting to generate a nonlinear relationship model.

[0091] Specifically, the system performs random forest modeling on the drug interaction feature set, sets a parameter for the number of decision trees, and initializes multiple untrained decision tree structures based on this parameter. Preferably, the system can use bootstrap sampling to extract multiple training sample subsets from the drug interaction feature set, with each subset used to train a decision tree. The system randomly selects a feature subset for each decision tree, and the size of the feature subset is the square root of the total number of features in the drug interaction feature set.

[0092] The system calculates the optimal feature selection criterion for node splitting. For each candidate splitting feature, the system calculates the reduction in impurity of that feature in the current node's sample set. Impurity is represented by the Gini index or entropy value. The feature with the largest reduction in impurity is determined as the optimal splitting feature. The system recursively performs node splitting until the number of node samples is less than a preset threshold or the tree depth reaches its maximum limit. After training all decision trees, the system integrates the structural parameters of multiple decision trees to generate a nonlinear relationship model. The model parameters include the splitting feature sequence, splitting threshold, leaf node output values, and tree hierarchical structure for each tree.

[0093] S33: The drug interaction feature set is predicted and processed by a nonlinear relationship model to calculate the synergistic and antagonistic effects of the drug combination and generate interaction effect prediction values. The interaction effect prediction values ​​are used to indicate the actual impact of the drug combination regimen on organ adverse reactions.

[0094] Specifically, the system performs predictive processing on the drug interaction feature set using a nonlinear relationship model. Sample data from the drug interaction feature set is input into the nonlinear relationship model, and each sample data is processed independently by each decision tree. The decision tree reaches the corresponding leaf node along the splitting path based on the sample feature value and outputs the tree's predicted value. The system collects the predicted values ​​from all decision trees and uses a simple majority voting method to determine the final prediction result. For regression tasks, the average value calculation method is used.

[0095] The system calculates the synergistic effect strength of drug combinations. When the prediction results show that the organ response strength of the combination therapy is greater than the sum of the individual drug response strengths, the synergistic effect strength is represented by a positive value, which is the ratio of the difference between the combined response strength and the sum of the individual drug response strengths to the sum of the individual drug response strengths. The system calculates the antagonistic effect strength. When the prediction results show that the organ response strength of the combination therapy is less than the sum of the individual drug response strengths, the antagonistic effect strength is represented by a negative value, which is the ratio of the difference between the sum of the individual drug response strengths and the combined response strength to the sum of the individual drug response strengths. The system generates interaction effect prediction values, which are standardized results of the synergistic or antagonistic effect strengths, represented by values ​​in the range [-1, 1]. Larger positive values ​​indicate stronger synergistic effects, and smaller negative values ​​indicate stronger antagonistic effects. The prediction values ​​are used to quantify the actual impact of drug combination regimens on OrganADR.

[0096] In one embodiment, such as Figure 2 As shown, S4 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0097] S41: Gene expression profile calibration is performed on the predicted values ​​of interaction effects. The drug clearance rate and volume of distribution parameters are adjusted according to the variations of drug metabolism-related genes to generate preliminary calibration values.

[0098] Specifically, the system imports the patient's gene expression profile data, which includes gene sequence information related to drug metabolism, including genes encoding drug-metabolizing enzymes, transport proteins, and receptors. The system performs variant detection on the gene expression profile data, identifying key gene variant sites in drug metabolism pathways, such as single nucleotide polymorphisms and insertion / deletion variants, by comparing them with a reference genome sequence. The system queries a pre-defined genotype-phenotype association database based on the gene variant type. This database stores the correspondence between different gene variants and metabolic phenotypes, determining the corresponding drug metabolism phenotype for different variants, such as rapid metabolizer, intermediate metabolizer, or slow metabolizer. The system calls upon a pharmacokinetic parameter library to extract baseline values ​​for drug clearance and volume of distribution. Based on the differences in metabolic phenotypes caused by gene variants, the system adjusts the baseline parameters, introducing a drug clearance adjustment formula:

[0099] CL = CL0 × G

[0100] Where CL is the adjusted drug clearance rate, CL0 is the baseline clearance rate in the pharmacokinetic parameter library, and G is the genotype correction factor, with G>1 for fast metabolizers, G<1 for slow metabolizers, and G=1 for intermediate metabolizers. A distribution volume adjustment formula is also introduced:

[0101] V d =V d0 ×(2-G)

[0102] Among them, V d V is the adjusted distribution volume. d0 Let G be the baseline distribution volume and G be the gene correction factor. The system substitutes the adjusted parameters into the calculation model for the interaction effect prediction value and recalculates to obtain the preliminary calibration value, which reflects the influence of gene factors on drug response.

[0103] S42: Correct the initial calibration values ​​for pathological conditions, update the drug sensitivity and toxicity threshold parameters in combination with organ function grading indicators, and generate intermediate predicted values.

[0104] Specifically, the system acquires preliminary calibration values, which have already incorporated the influence of gene expression profiles. The system imports the patient's organ function classification indicators, determined based on clinical examination data, covering the functional status classification of major organs such as the liver, kidneys, and heart, including Child-Pugh classification for liver function, chronic kidney disease staging for kidney function, and NYHA classification for heart function. Based on the organ function classification, the system queries preset pathological state-parameter association rules, established based on clinical research data, to determine the drug sensitivity coefficient and toxicity threshold correction coefficient for the corresponding classification. The system updates the drug sensitivity parameters in the preliminary calibration values, introducing a drug sensitivity coefficient adjustment formula:

[0105] S = S0 × F

[0106] Where S is the adjusted drug sensitivity coefficient, S0 is the baseline sensitivity coefficient in the initial calibration value, and F is the pathological state factor, where F>1 when organ function grade decreases and F<1 when organ function grade increases. The system synchronously adjusts the toxicity threshold parameter, introducing the toxicity threshold adjustment formula:

[0107] T = T0 × (1 / F)

[0108] Where T is the adjusted toxicity threshold, T0 is the baseline toxicity threshold, and F is the pathological state factor. The system substitutes the updated parameters into the initial calibration values ​​and generates intermediate predicted values ​​through model calculations. These values ​​integrate the combined effects of genes and pathological states.

[0109] S43: Obtain patient lifestyle data, optimize lifestyle by combining intermediate predictive values, adjust the adverse reaction probability of intermediate predictive values ​​by introducing factors influencing lifestyle habits, and generate individualized drug response predictive values. Individualized drug response predictive values ​​are used to indicate the intensity of organ adverse reaction risk for a specific patient using a combination of drugs.

[0110] Specifically, the system collects patients' lifestyle data, obtained through structured questionnaires or wearable devices, covering categories such as smoking behavior, alcohol consumption frequency, dietary structure, exercise intensity, and sleep patterns. The system quantifies the lifestyle data, introducing a formula for quantifying lifestyle factors:

[0111]

[0112] Among them, f k Let x be the quantified value of the kth lifestyle factor. k This is the original measurement value of this factor. σ represents the mean of this factor in the healthy population. k This represents the standard deviation of this factor in healthy individuals. The system calls upon a lifestyle influence coefficient library, obtained through retrospective clinical data statistics, which includes the weights w of various lifestyle factors. k The system incorporates a lifestyle-related attenuation coefficient λ, where the weights reflect the degree of influence of different factors on drug response, and the attenuation coefficient reflects the cumulative attenuation rate of the overall lifestyle effect. The system then quantifies the lifestyle factors f. k Weight w k Substituting the terms into the summation term in the formula, the comprehensive impact value of lifestyle is calculated. Preferably, the adjustment formula for the probability of adverse reactions is:

[0113]

[0114] Among them, P final P represents the adjusted probability of adverse reactions. inter The intermediate predicted value is the pathology-corrected value, λ is the attenuation coefficient of lifestyle influence, and w k f represents the weight of the kth lifestyle factor. k Let be the quantified value of the k-th lifestyle factor, and m be the number of lifestyle factor types. When the lifestyle is health-oriented, If the value is positive, the exp term is less than 1. Below P inter When the lifestyle is risk-prone, If the value is negative, the exp term is greater than 1. Higher than P interThe system generates a personalized drug response prediction value through the above calculations. This value integrates the influence of genes, pathology, and lifestyle, and is used to indicate the risk intensity of organ adverse reactions when a specific patient uses a combination of drugs.

[0115] In one embodiment, step S5 of the drug combination data processing method based on organ adverse reactions provided by the present invention specifically includes the following steps:

[0116] S51: Monte Carlo sampling is performed on the individualized drug response prediction values ​​to simulate organ toxicity responses under different dosage scenarios and generate a toxicity probability distribution.

[0117] Specifically, the system performs Monte Carlo sampling on the individualized drug response prediction values ​​to extract key parameters, including drug metabolism rate, organ tolerance threshold, dose-response curve slope, and individual variability coefficient. The system sets sampling parameters, including the number of samples, dose range, and step size. The dose range covers 0.5 to 2 times the commonly used clinical dose, and the step size is 1 / 10 of the smallest clinical dosing unit.

[0118] For example, the system generates random samples conforming to a uniform distribution using a pseudo-random number generator. Each sample contains a random dose value for each drug, with the dose value falling within a preset dose range. The system substitutes the random samples into a drug-organ interaction model, which is built based on pharmacokinetic-pharmacodynamic principles and includes mathematical expressions for each stage of drug absorption, distribution, metabolism, and excretion. The system calculates the organ toxicity response value corresponding to each random sample. The calculation process involves the drug concentration accumulation in the target organ, the binding rate to cellular targets, and the cascade amplification effect of subsequent pathological reactions. The system performs statistical analysis on all sampling results, uses a kernel density estimation algorithm to fit the probability distribution curve of the toxicity response values, and generates a toxicity probability distribution. This distribution is stored in the form of a probability density function, containing the probability of occurrence and 95% confidence intervals corresponding to different toxicity levels.

[0119] S52: Extract high-risk combinations from the toxicity probability distribution, screen for therapeutic drug combinations that exceed the preset organ toxicity threshold, and generate a high-risk drug combination set.

[0120] Specifically, the system performs high-risk combination extraction processing on the toxicity probability distribution and loads preset organ toxicity thresholds. These thresholds are divided according to organ type: the liver toxicity threshold is set based on the multiple of increase in serum transaminase, the kidney toxicity threshold is set based on the decrease in creatinine clearance, and the cardiotoxicity threshold is set based on the QT interval prolongation value. The threshold data are from the clinical safety standards published by the National Adverse Drug Reaction Monitoring Center.

[0121] For example, the system iterates through all drug combinations in the toxicity probability distribution and calculates the cumulative probability that the toxicity response value of each combination exceeds the corresponding organ toxicity threshold at each dose point. The system sets a high-risk judgment criterion; when the cumulative probability exceeds a preset probability threshold, the drug combination is marked as a high-risk candidate. The system performs repeatability verification on high-risk candidate combinations, confirming the stability of the high-risk attribute by doubling the number of samples. The system generates a set of high-risk drug combinations, which includes the generic name of the drug, the dosage, the type of organ involved, the probability of exceeding the toxicity threshold, and the corresponding dose point. The data is stored in tabular form, and each record is associated with a unique identifier of the original individualized drug response prediction value.

[0122] S53: Perform compliance classification processing on high-risk drug combination sets, match them with safety standards in the preset regulatory rule base, and generate compliance risk assessment results.

[0123] Specifically, the system performs compliance classification processing on high-risk drug combination sets. The system loads a preset regulatory rule base, which is divided into basic rules, special population rules, and combination drug rules. The basic rules include drug contraindications and precautions stipulated in the National Pharmacopoeia, the special population rules cover drug restrictions for patients with hepatic or renal insufficiency, the elderly, and children, and the combination drug rules include the classification and management requirements of drug interactions.

[0124] Preferably, the system uses a string matching algorithm to associate drug names in the high-risk drug combination set with drug entries in the rule base, extracting relevant rule clauses. The system employs a rule matching degree calculation method to quantify the degree of compliance between high-risk combinations and each rule clause. The matching degree calculation involves the degree of agreement between drug type, dosage, route of administration, and patient characteristics. Based on the matching results, the system classifies high-risk combinations into prohibited, restricted, and monitored categories, associating the classification results with the corresponding rule clause number and evidence level. The system generates a compliance risk assessment result, which includes the high-risk combination's classification identifier, the specific rule clause violated, the quantified risk level, and alternative medication recommendations. The result data structure is compatible with the data interface format of the hospital information system.

[0125] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0126] Based on the same inventive concept, this application also provides a system for processing drug combination data based on organ adverse reactions (ADDRs) to implement the aforementioned method for processing drug combination data based on organ adverse reactions. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the system for processing drug combination data based on organ adverse reactions provided below can be found in the limitations of the method for processing drug combination data based on organ adverse reactions described above, and will not be repeated here.

[0127] Preferably, such as Figure 3 As shown, the present invention provides a drug combination data processing system 600 based on organ adverse reactions, which is configured with the following modules:

[0128] The multi-source medical data integration module 610 is used to integrate and process the acquired multi-source patient medical data. It fuses organ function parameters, metabolic characteristics and hemodynamic data through graph neural networks, performs unit unification transformation and constructs a multi-scale data association matrix to generate a medical dataset in a unified format.

[0129] The core feature extraction and dimensionality reduction module 620 is used to extract core features from medical datasets, call principal component analysis to calculate the contribution weight of each feature vector, and screen key biomarkers that identify organs to generate a set of dimensionality-reduced feature vectors.

[0130] The drug interaction effect modeling module 630 is used to model drug interaction effects based on a set of dimensionality-reduced feature vectors, integrate dose-time series and in vitro experimental data to construct a nonlinear relationship model, calculate the strength of drug synergy through the random forest algorithm, and generate predicted values ​​of interaction effects.

[0131] The individualized prediction calibration module 640 is used to perform individualized dynamic calibration of the interaction effect prediction value, adjust the prediction parameters by combining gene expression profile and pathological state data, introduce the acquired patient life data for probability distribution optimization and update, and generate individualized drug response prediction values.

[0132] The risk decision assessment module 650 is used to process risk decisions based on individualized drug response prediction values. It calls Monte Carlo sampling to simulate organ toxicity probability distribution, performs flow cytometry classification and labeling of high-risk combinations according to a preset regulatory rule base, and generates compliance risk assessment results. The compliance risk assessment results are used to indicate the safety level and clinical applicability of drug combination regimens.

[0133] Preferably, the multi-source medical data integration module 610 provided in this application is configured with the following units:

[0134] The medical data cleaning unit is used to clean organ function parameters, metabolic characteristic data and hemodynamic data of acquired multi-source patient medical data, remove missing values ​​and outlier data points in the original data, and generate cleaned medical data.

[0135] The unified data unit is used to unify the units of cleaned medical data, converting data with different units of measurement into a preset standard unit system to generate standard medical multi-source data.

[0136] The graph neural network fusion unit is used to fuse standard medical multi-source data through graph neural networks, establish the correlation between organ function and metabolic characteristics, and generate a medical dataset in a unified format. The medical dataset is used to indicate the integration and correlation results of cross-scale biomedical data.

[0137] Preferably, the core feature extraction and dimensionality reduction module 620 provided in this application is configured with the following units:

[0138] Principal component analysis ranking unit is used to perform principal component analysis on medical datasets, calculate the variance contribution of organ function indicators, metabolic activity parameters and hemodynamic feature vectors, and generate feature importance ranking;

[0139] The organ feature screening unit is used to perform organ-specific screening by ranking the importance of features, assigning screening weights to liver, kidney and heart-related features according to organ type, and generating a set of key biomarker identifiers.

[0140] The biomarker dimensionality reduction unit is used to reduce the dimensionality of the key biomarker identifier set, retain the high-contribution feature vectors in the biomarker set, and generate a dimensionality-reduced feature vector set. The dimensionality-reduced feature vector set is used to indicate the core feature combination most relevant to adverse organ reactions.

[0141] Preferably, the drug interaction effect modeling module 630 provided in this application is configured with the following units:

[0142] The drug feature integration unit is used to integrate drug features from the dimensionality-reduced feature vector set, and to construct a joint feature space of organ response features and drug characteristic parameters by combining dose-time series data and in vitro experimental data, thereby generating a drug interaction feature set.

[0143] The random forest modeling unit is used to perform random forest modeling on the drug interaction feature set, train multiple decision trees and calculate the optimal feature selection criterion for node splitting, and generate a nonlinear relationship model.

[0144] The interaction effect prediction unit is used to predict the drug interaction feature set through a nonlinear relationship model, calculate the synergistic and antagonistic effects of the drug combination, and generate interaction effect prediction values. The interaction effect prediction values ​​are used to indicate the actual impact of the drug combination regimen on organ adverse reactions.

[0145] Preferably, the individualized prediction calibration module 640 provided in this application is configured with the following units:

[0146] The gene expression calibration unit is used to calibrate the gene expression profile of the interaction effect prediction value, adjust the drug clearance rate and volume of distribution parameters according to the variation of drug metabolism-related genes, and generate preliminary calibration values.

[0147] The pathological state correction unit is used to correct the initial calibration value for pathological state, update the drug sensitivity and toxicity threshold parameters in combination with organ function grading indicators, and generate intermediate prediction values.

[0148] The lifestyle optimization unit is used to acquire patient lifestyle data, combine intermediate predicted values ​​to optimize lifestyle, introduce lifestyle influencing factors to adjust the adverse reaction probability of intermediate predicted values, and generate individualized drug response predicted values. Individualized drug response predicted values ​​are used to indicate the intensity of organ adverse reaction risk for a specific patient using a combination drug regimen.

[0149] Preferably, the risk decision assessment module 650 provided in this application is configured with the following units:

[0150] The Monte Carlo toxicity sampling unit is used to perform Monte Carlo sampling on individualized drug response predictions to simulate organ toxicity responses under different dosage scenarios and generate toxicity probability distributions.

[0151] The high-risk combination extraction unit is used to perform high-risk combination extraction processing on the toxicity probability distribution, screen the therapeutic drug combinations that exceed the preset organ toxicity threshold, and generate a high-risk drug combination set.

[0152] The compliance classification and assessment unit is used to classify high-risk drug combination sets for compliance, match them with safety standards in the preset regulatory rule base, and generate compliance risk assessment results.

[0153] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method for processing drug combination data based on organ adverse reactions.

[0154] In one embodiment, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for processing drug combination data based on organ adverse reactions.

[0155] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0156] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0157] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for processing drug combination data based on organ adverse reactions, characterized in that, Includes the following steps: S1: Integrate and process the acquired multi-source patient medical data, fuse organ function parameters, metabolic characteristics and hemodynamic data through graph neural networks, perform unit unification transformation and construct a multi-scale data association matrix to generate a medical dataset in a unified format; S2: Extract core features from the medical dataset, use principal component analysis to calculate the contribution weight of each feature vector, and screen key biomarkers that identify organs to generate a set of dimensionality-reduced feature vectors. S3: Based on the set of reduced-dimensional feature vectors, model the drug interaction effect, integrate dose-time series and in vitro experimental data to construct a nonlinear relationship model, calculate the strength of drug synergy through random forest algorithm, and generate predicted values ​​of interaction effect; S4: Perform individualized dynamic calibration on the predicted value of the interaction effect, adjust the prediction parameters by combining gene expression profile and pathological state data, introduce the acquired patient life data for probability distribution optimization and update, and generate individualized drug response prediction value. S5: Based on the individualized drug response prediction value, risk decision processing is performed, Monte Carlo sampling simulation of organ toxicity probability distribution is called, and high-risk combinations are classified and labeled by flow cytometry according to the preset regulatory rule base to generate compliance risk assessment results. The compliance risk assessment results are used to indicate the safety level and clinical applicability of the drug combination regimen.

2. The method according to claim 1, characterized in that, S1 includes: S11: Perform data cleaning on the organ function parameters, metabolic characteristics and hemodynamic data of the acquired multi-source patient medical data, remove missing values ​​and outlier data points in the original data, and generate cleaned medical data. S12: Perform unit standardization conversion on the cleaned medical data, converting data with different units of measurement into a preset standard unit system to generate standard medical multi-source data; S13: The standard medical multi-source data is fused and processed by a graph neural network to establish the correlation between organ function and metabolic characteristics, and to generate a medical dataset in a unified format. The medical dataset is used to indicate the integration and correlation results of cross-scale biomedical data.

3. The method according to claim 1, characterized in that, S2 includes: S21: Perform principal component analysis on the medical dataset to calculate the variance contribution of organ function indicators, metabolic activity parameters and hemodynamic feature vectors, and generate a ranking of feature importance. S22: Perform organ-specific screening on the importance ranking of the features, assign screening weights to liver, kidney and heart-related features according to organ type, and generate a key biomarker identifier set; S23: Perform dimensionality reduction processing on the key biomarker identifier set, retain the high contribution feature vectors in the biomarker set, and generate a dimensionality-reduced feature vector set. The dimensionality-reduced feature vector set is used to indicate the core feature combination most relevant to adverse organ reactions.

4. The method according to claim 1, characterized in that, S3 includes: S31: Integrate drug features into the dimensionality-reduced feature vector set, and construct a joint feature space of organ response features and drug characteristic parameters by combining dose-time series data and in vitro experimental data to generate a drug interaction feature set; S32: Perform random forest modeling on the drug interaction feature set, train multiple decision trees and calculate the optimal feature selection criterion for node splitting to generate a nonlinear relationship model; S33: The drug interaction feature set is predicted and processed by the nonlinear relationship model to calculate the synergistic and antagonistic effects of the drug combination and generate an interaction effect prediction value. The interaction effect prediction value is used to indicate the actual impact of the drug combination regimen on organ adverse reactions.

5. The method according to claim 1, characterized in that, S4 includes: S41: Gene expression profile calibration is performed on the predicted interaction effect values. The drug clearance rate and volume of distribution parameters are adjusted according to the variations of drug metabolism-related genes to generate preliminary calibration values. S42: Correct the preliminary calibration value for pathological condition, update the drug sensitivity and toxicity threshold parameters in combination with organ function grading indicators, and generate intermediate prediction values. S43: Obtain patient lifestyle data, optimize lifestyle by combining the intermediate predicted value, introduce lifestyle influencing factors to adjust the adverse reaction probability of the intermediate predicted value, and generate an individualized drug response predicted value. The individualized drug response predicted value is used to indicate the intensity of organ adverse reaction risk for a specific patient using a combination drug regimen.

6. The method according to claim 5, characterized in that, The formula for adjusting the probability of adverse reactions is: Among them, P final P represents the adjusted probability of adverse reactions. inter The intermediate predicted value is the pathology-corrected value, λ is the attenuation coefficient of lifestyle influence, and w k f represents the weight of the kth lifestyle factor. k Let m be the quantitative value of the kth lifestyle factor, and m be the number of lifestyle factor types.

7. The method according to any one of claims 1-6, characterized in that, S5 includes: S51: Perform Monte Carlo sampling on the individualized drug response prediction values ​​to simulate organ toxicity responses under different dosage scenarios and generate a toxicity probability distribution. S52: Perform high-risk combination extraction processing on the toxicity probability distribution, screen for therapeutic drug combinations that exceed the preset organ toxicity threshold, and generate a high-risk drug combination set; S53: Perform compliance classification processing on the high-risk drug combination set, match the safety standards in the preset regulatory rule base, and generate compliance risk assessment results.

8. A drug combination data processing system based on organ adverse reactions, characterized in that, The system includes: The multi-source medical data integration module is used to integrate and process the acquired multi-source patient medical data. It fuses organ function parameters, metabolic characteristics and hemodynamic data through graph neural networks, performs unit unification transformation and constructs a multi-scale data association matrix to generate a medical dataset in a unified format. The core feature extraction and dimensionality reduction module is used to extract core features from the medical dataset, call principal component analysis to calculate the contribution weight of each feature vector, and screen key biomarkers that identify organs to generate a set of dimensionality-reduced feature vectors. The drug interaction effect modeling module is used to model drug interaction effects based on the dimensionality-reduced feature vector set, integrate dose-time series and in vitro experimental data to construct a nonlinear relationship model, calculate the strength of drug synergy through random forest algorithm, and generate predicted values ​​of interaction effects. The individualized prediction calibration module is used to perform individualized dynamic calibration on the predicted value of the interaction effect, adjust the prediction parameters by combining gene expression profile and pathological state data, introduce the acquired patient life data for probability distribution optimization and update, and generate individualized drug response prediction values. The risk decision assessment module is used to perform risk decision processing based on the individualized drug response prediction value, call Monte Carlo sampling to simulate organ toxicity probability distribution, perform flow cytometry classification and labeling of high-risk combinations according to the preset regulatory rule base, and generate compliance risk assessment results. The compliance risk assessment results are used to indicate the safety level and clinical applicability of the drug combination regimen.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images