An individualized medication prediction method based on artificial intelligence

Abstract

The present application relates to the technical field of artificial intelligence, and particularly relates to an individualized medication prediction method based on artificial intelligence, comprising: acquiring a plurality of standard historical medical data sets; acquiring metabolic capacity scores and drug metabolism rates corresponding to all data groups; training to obtain relationship models corresponding to all standard historical medical data sets; acquiring liver metabolism parameters, gene expression parameters and each drug name in a treatment scheme of a patient to obtain a drug metabolism rate; and determining a corresponding prediction drug amount determination method according to whether the drug types in the treatment scheme have a competitive relationship. The present application integrates drug metabolism enzyme-related gene expression and liver metabolism function, simultaneously incorporates drug absorption competition and metabolism enzyme competition when drugs are used in combination, corrects the final output prediction drug amount when drugs are used in combination, so that the model is more suitable for the actual drug use scene, and the accuracy of the prediction result is increased.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based method for personalized medication prediction. Background Technology

[0002] With the continuous development of medicine, the treatment of diseases has gradually shifted from the original mass treatment plan to a more individualized and refined approach. Compared with the traditional fixed-dosage dosing model based on population statistics, the individualized and refined dosing model is an important way to improve the efficacy of drug treatment and reduce adverse reactions caused by improper drug dosage. Based on this, artificial intelligence algorithms are being used more and more widely in the fields of medical data processing, personalized medicine, and drug dosage prediction.

[0003] However, predicting individual medication dosages based on artificial intelligence still has the following drawbacks: On the one hand, existing medication prediction methods only focus on individual medication situations. When predicting medication dosages for combined medication, they do not consider the absorption competition and metabolic enzyme competition during combined medication, resulting in a large deviation between the predicted drug metabolism rate and the actual metabolism rate in combined medication scenarios. On the other hand, existing medication dosage prediction methods still use population-average metabolic parameters or fixed coefficients to calculate medication dosages, which cannot truly achieve the goal of predicting medication dosages based on the patient's solid characteristics.

[0004] Therefore, the existing technology has defects and urgently needs improvement. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based personalized medication prediction method to address the following issues in existing technologies: Firstly, existing medication prediction methods focus only on individual medication situations and fail to consider absorption competition and metabolic enzyme competition during combined medication, leading to significant discrepancies between predicted and actual drug metabolism rates in combined medication scenarios. Secondly, existing medication dosage prediction methods still use population-average metabolic parameters or fixed coefficients to calculate medication dosage, failing to truly achieve the goal of predicting medication dosage based on the patient's solid characteristics.

[0006] This invention provides an artificial intelligence-based personalized medication prediction method, comprising: Step S1: Obtain historical medical datasets, preprocess the medical data, and divide the preprocessed historical medical datasets into several standard historical medical datasets according to the drug names; the historical medical datasets include several data groups, each data group including: patient gene expression parameter data, liver metabolism parameter data, blood drug concentration data, and corresponding clinical drug safety assessment data; the preprocessing includes: data screening and standardization. Step S2: For a single data group in a single standard historical medical dataset, a machine learning multi-feature fusion weighting strategy is used to calculate the metabolic capacity score based on the gene expression parameters and the liver metabolic parameters, and to calculate the patient's drug metabolism rate based on the blood drug concentration data. This step is repeated until the metabolic capacity score and drug metabolism rate corresponding to all data groups are obtained. Step S3: Based on the metabolic capacity scores and drug metabolism rates corresponding to all data groups, a traditional machine learning regression algorithm is used to train and model the relationship model. Step S2 and this step are repeated until the relationship model corresponding to all standard historical medical datasets is obtained. Step S4: Obtain the patient's liver metabolic parameters, gene expression parameters, and the names of each drug in the treatment plan. Determine the corresponding relationship model based on each drug name. Calculate the metabolic capacity score based on the liver metabolic parameters and blood drug concentration parameters. Input the metabolic capacity score into the relationship model to obtain the drug metabolism rate corresponding to each drug name for the patient. Step S5 involves determining the method for determining the predicted dosage based on whether there is competition between the types of drugs in the treatment plan. If drug competition exists, the drug metabolism rate corresponding to each drug name in the treatment plan is corrected based on the drug competition relationship. The predicted dosage for each drug name is then determined based on the corrected drug metabolism rate. If no drug competition exists, the predicted dosage for each drug is calculated based on the drug metabolism rate corresponding to each drug name in the treatment plan, the target blood concentration range of each drug, and the dosing interval. As a preferred technical solution for the artificial intelligence-based personalized medication prediction method, the gene expression parameter data includes at least two of the following: single nucleotide polymorphism site data of drug-metabolizing enzyme genes, gene copy number variation data, and quantitative gene expression level data. The liver metabolic parameter data include at least two of the following: liver function biochemical indicators, liver imaging scores, and liver reserve function scores.

[0007] As a preferred technical solution for a personalized medication prediction method based on artificial intelligence, the step of calculating a metabolic capacity score based on the gene expression parameters and the liver metabolic parameters includes: Assign a corresponding weight coefficient to each parameter in the gene expression parameters, calculate the sum of the products of each parameter in the gene expression parameters and its corresponding weight coefficient, and record it as the gene expression score; Assign a corresponding weight coefficient to each parameter in the liver metabolic parameters, calculate the sum of the products of each parameter in the liver metabolic parameters and its corresponding weight coefficient, and record it as the liver metabolic score; The product of the gene expression score and the liver metabolism score is used as the metabolic capacity score.

[0008] As a preferred technical solution for personalized medication prediction based on artificial intelligence, the calculation of the patient's drug metabolism rate based on blood drug concentration data includes: The blood drug concentration data were fitted to obtain a blood drug concentration-time curve; The systemic clearance rate of the drug is obtained from the blood drug concentration-time curve and used as the patient's drug metabolism rate.

[0009] As a preferred technical solution for personalized medication prediction based on artificial intelligence, the metabolic capacity score and drug metabolism rate corresponding to all data groups are trained to obtain a relational model. The relational model is constructed by combining the correspondence between metabolic capacity score and drug metabolism rate with a regression algorithm. If the metabolic capacity score is input, the drug metabolism capacity is output.

[0010] As a preferred technical solution for a personalized medication prediction method based on artificial intelligence, the training steps of the relationship model include: The data sets in a single standard historical medical dataset are divided into training and validation sets in a 7:3 ratio. Stratified sampling is used in the partitioning process. Construct a traditional machine learning regression model, initialize the regression coefficients and intercept term in the linear regression model, and set the learning rate and number of iterations for the traditional machine learning regression model. The metabolic capacity score in each data group of the training set is used as the input feature, and the drug metabolism rate is used as the output label. The input is fed into the traditional machine learning regression model. The mean squared error is used as the loss function. The gradient descent method is combined to minimize the error between the predicted drug metabolism rate and the historical metabolism rate in the standard historical medical dataset. The model is iteratively optimized based on gradient descent until the loss function converges to obtain the relational model. The relational model is validated based on the validation set. If the deviation between the predicted drug metabolism rate output by the relational model and the historical drug metabolism rate in the validation set is less than a preset deviation value, the model is deemed qualified. If the deviation between the predicted drug metabolism rate output by the relational model and the historical drug metabolism rate in the validation set is greater than or equal to the preset deviation value, the model is deemed unqualified. In this case, the learning rate and number of iterations of the traditional machine learning regression model are adjusted and the model is retrained. The deviation value is the average absolute percentage error between the predicted drug metabolism rate and the historical drug metabolism rate.

[0011] As a preferred technical solution for a personalized medication prediction method based on artificial intelligence, the step of responding to the existence of drug competition, correcting the drug metabolism rate corresponding to each drug name in the treatment plan according to the drug competition relationship, and determining the predicted dosage corresponding to each drug name based on the corrected drug metabolism rate in the treatment plan, includes: Based on big data, information on drug interactions is collected. An interaction table is constructed with rows and columns representing drug types. In each cell, after determining the mutual influence of drugs based on the corresponding row and column, an absorption efficiency correction factor for the corresponding drug type is determined. The absorption efficiency correction factor is the ratio of the absorption rate of the drug types in combination to the absorption rate of the drug types in single administration. Calculate the enzyme competition effect correction factor for each drug based on the enzyme's affinity for each drug in which there is a drug competition relationship. The product of the absorption efficiency correction factor and the enzyme competition effect correction factor is used as the correction coefficient. The product of the correction coefficient and the drug metabolism rate output by the relation model is used as the final drug metabolism rate. The predicted dosage is calculated based on the target blood drug concentration range of the final drug metabolism rate and the dosing interval.

[0012] As a preferred technical solution for personalized medication prediction based on artificial intelligence, the step of calculating the enzyme competition influence correction factor corresponding to each drug based on the enzyme affinity to each drug with a drug competition relationship, wherein: The enzyme competition effect correction factor is calculated according to the following formula: ; in: This is the enzyme competition effect correction factor for a single drug i that has a drug competition relationship. This refers to the enzyme affinity of an enzyme for a single drug i that is in competition with it for drug use. This represents the sum of enzyme affinities for all single drugs that compete with each other, and the number of n is equal to the number of drugs that compete with each other for the same enzyme.

[0013] As a preferred technical solution for a personalized medication prediction method based on artificial intelligence, the step of determining the predicted medication dosage based on the patient's drug metabolism rate includes: The predicted dosage is calculated based on the drug metabolism rate, target blood drug concentration range, and dosing interval output by the relational model.

[0014] As a preferred technical solution for a personalized medication prediction method based on artificial intelligence, the preprocessing of medical data includes data screening and standardization. The data filtering is performed in response to any of the following conditions being met: The data set contains incomplete data on gene expression parameters, liver metabolism parameters, blood drug concentration data, and clinical drug safety assessment data. In the dataset, the sequencing depth of gene expression parameters is lower than a preset depth; In the aforementioned clinical drug safety assessment, the patient experienced moderate or severe adverse reactions; The standardization process includes: Gene expression parameter data are numerically encoded and converted into a unified numerical data type. The biochemical indicators in liver metabolic parameters are standardized to obtain metabolic parameter values; Outlier detection and correction were performed on the blood drug concentration data, and interpolation was used to complete the blood drug concentration data for missing time points.

[0015] Compared with the prior art, the beneficial effects of the present invention are that by integrating the gene expression related to drug-metabolizing enzymes and liver metabolic function, the prediction results are made to conform to the physiological law that "genes determine the upper limit of metabolism, and the liver determines the degree of execution of the upper limit". At the same time, it incorporates the drug absorption competition and metabolic enzyme competition when using combination drugs, and corrects the final output predicted drug dosage when using combination drugs, making the model more in line with the actual drug use scenario and increasing the accuracy of the prediction results. Attached Figure Description

[0016] Figure 1 The flowchart of a personalized medication prediction method based on artificial intelligence provided by the present invention is shown. Detailed Implementation

[0017] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0019] like Figure 1 As shown, this invention discloses a personalized medication prediction method based on artificial intelligence, comprising: Step S1: Obtain historical medical datasets, preprocess the medical data, and divide the preprocessed historical medical datasets into several standard historical medical datasets according to the drug names; the historical medical datasets include several data groups, each data group including: patient gene expression parameter data, liver metabolism parameter data, blood drug concentration data, and corresponding clinical drug safety assessment data; the preprocessing includes: data screening and standardization. Step S2: For a single data group in a single standard historical medical dataset, a machine learning multi-feature fusion weighting strategy is used to calculate the metabolic capacity score based on the gene expression parameters and the liver metabolic parameters, and to calculate the patient's drug metabolism rate based on the blood drug concentration data. This step is repeated until the metabolic capacity score and drug metabolism rate corresponding to all data groups are obtained. Step S3: Based on the metabolic capacity scores and drug metabolism rates corresponding to all data groups, a traditional machine learning regression algorithm is used to train and model the relationship model. Step S2 and this step are repeated until the relationship model corresponding to all standard historical medical datasets is obtained. Step S4: Obtain the patient's liver metabolic parameters, gene expression parameters, and the names of each drug in the treatment plan. Determine the corresponding relationship model based on each drug name. Calculate the metabolic capacity score based on the liver metabolic parameters and blood drug concentration parameters. Input the metabolic capacity score into the relationship model to obtain the drug metabolism rate corresponding to each drug name for the patient. Step S5: Determine the method for determining the corresponding predicted dosage based on whether there is a competitive relationship between the types of drugs in the treatment plan. In response to the existence of a drug competition relationship, the drug metabolism rate corresponding to each drug name in the treatment plan is corrected according to the drug competition relationship. Based on the corrected drug metabolism rate corresponding to each drug name in the treatment plan, the predicted dosage corresponding to each drug name is determined. In response to the absence of a drug competition relationship, the predicted dosage of each drug is calculated based on the drug metabolism rate corresponding to each drug name in the treatment plan, the target blood concentration range of each drug, and the dosing interval.

[0020] Furthermore, this invention preprocesses the data and divides it according to the drug name to obtain an accurate standard historical medical dataset. Based on the standard historical medical dataset, a metabolic capacity score is calculated for each data group in the standard historical dataset. The patient's drug metabolism rate is calculated using blood drug concentration data. Through model training, a relational model representing the correspondence between the metabolic capacity score and the patient's drug metabolism rate is obtained. Thus, when medication is required repeatedly, the relational model corresponding to the drug name can be quickly obtained based on the drug name. The metabolic capacity score is calculated based on the patient's liver metabolic parameters and blood drug concentration parameters. The metabolic capacity score is input into the relational model to obtain the drug metabolism rate corresponding to each drug name for the patient, thereby calculating the patient's predicted medication dosage. Based on this, considering that when using combined medications, there may be mutual influences between the drugs, thus affecting the absorption and metabolism rate during the actual medication process, this invention also introduces an absorption efficiency correction factor and an enzyme competition influence correction factor. This makes the medication dosage prediction results more realistic. Starting from the perspective of low metabolism leading to long cumulative time and damage to the patient's body, this invention fundamentally avoids the risks caused by overdose or underdose from a metabolic perspective, reducing the adverse reaction rate of the predicted results.

[0021] Specifically, the gene expression parameter data includes at least two of the following: single nucleotide polymorphism site data of drug-metabolizing enzyme genes, gene copy number variation data, and quantitative gene expression level data. Liver metabolic parameter data include at least two of the following: liver function biochemical indicators, liver imaging scores, and liver reserve function scores.

[0022] It should be noted that, in practice, drug-metabolizing enzyme genes include, but are not limited to: CYP family genes (CYP3A4, CYP2D6, and CYP2C9); single nucleotide polymorphism (SNP) site data refers to the genotype data (wild-type, hybrid mutant, or homozygous mutant) of core mutation sites (such as CYP3A41G or CYP2D610) in drug-metabolizing enzyme genes. Copy number variation data refers to the copy number detection results of drug-metabolizing enzyme genes. Quantitative gene expression level data refers to the relative expression level of the mRNA corresponding to the drug-metabolizing enzyme gene, which can be detected using qPRC technology. In this embodiment of the invention, gene expression parameters are selected as two parameters: SNP site data of the CYP3A4 gene and relative mRNA expression level, obtained through peripheral blood gene sequencing and qPRC technology, respectively. Liver function biochemical indicators include, but are not limited to: alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin, albumin, and alkaline phosphatase. Liver imaging scores include, but are not limited to, liver ultrasound scores or liver CT scores. Liver reserve function scores are the 15-minute retention rate of the indocyanine green clearance test. In this embodiment of the invention, liver function biochemical indicators are selected as liver metabolic parameters. The liver function biochemical indicators selected are alanine aminotransferase (ALT) and aspartate aminotransferase (AST), which are obtained through clinical biochemical detection and Doppler ultrasound detection.

[0023] Specifically, a metabolic capacity score is calculated based on gene expression parameters and liver metabolic parameters, including: Assign a corresponding weight coefficient to each parameter in the gene expression parameters, calculate the sum of the products of each parameter in the gene expression parameters and its corresponding weight coefficient, and record it as the gene expression score; Assign a corresponding weight coefficient to each parameter in the liver metabolic parameters, calculate the sum of the products of each parameter in the liver metabolic parameters and its corresponding weight coefficient, and record it as the liver metabolic score; The product of gene expression score and liver metabolism score is used as the metabolic capacity score.

[0024] In detail, in the technical solution of this invention, the weighting coefficients corresponding to each parameter in the gene expression parameters and the weighting coefficients corresponding to each parameter in the liver metabolism parameters are configured by computer. In this embodiment of the invention, the weighting coefficients for the SNP site data of the CYP3A4 gene and the relative expression level of mRNA selected as gene expression parameters are 0.6 and 0.4, respectively. The gene expression score is: CYP3A4 gene SNP site data score × 0.6 + mRNA relative expression level score × 0.4. The liver metabolism parameters are selected as liver function biochemical indicators, specifically alanine aminotransferase (ALT) and aspartate aminotransferase (AST), with weighting coefficients configured to be 0.5 and 0.5, respectively. The liver metabolism parameter score is: ALT score × 0.5 + AST score × 0.5.

[0025] In detail, this invention quantifies gene expression scores and liver metabolism scores separately by weighted summation, thereby eliminating the barrier that prevents calculations between different dimensions. It realizes the data representation of the physiological mechanism that determines the upper limit of gene expression and the actual degree of drug metabolism determined by liver metabolism, providing a standardized data foundation for the training of relational models, thereby improving the accuracy of subsequent relational model training and prediction results.

[0026] Furthermore, the patient's drug metabolism rate is calculated based on blood drug concentration data, including: The blood drug concentration data were fitted to obtain the blood drug concentration-time curve; The systemic clearance rate of the drug is obtained from the blood drug concentration-time curve and used as the patient's drug metabolism rate.

[0027] In implementation, blood drug concentration data of patients at 0.5h, 1h, 2h, 4h, 8h, and 12h after medication were obtained from historical medical datasets. A nonlinear least squares method was used to fit the blood drug concentration-time curve based on a one-compartment open model. The fitting formula is as follows: ,in Let be the blood drug concentration at time t. , , , All are fitting parameters. Integrate the fitted blood drug concentration-time curve and calculate the area under the curve (AUC) of the region formed by the blood drug concentration-time curve and the horizontal axis from 0 to infinity. Then, use the formula... ,in: For whole-body clearance rate, This is the dosage, in mg. For bioavailability, The horizontal axis represents the area formed by the blood drug concentration-time curve and the horizontal axis, ranging from 0 to infinity. Bioavailability is the relative amount of a drug that, after entering the body at a certain dose, can reach the site of action and exert its therapeutic effect, i.e., the rate at which it does so. The calculated systemic clearance rate is taken as the patient's drug metabolism rate. The fitting of the blood drug concentration-time curve, integration, and calculation of systemic clearance rate in the above process are all existing technologies and will not be elaborated upon here.

[0028] In detail, this invention uses the blood drug concentration-time curve as a basis to calculate the drug clearance rate as the drug metabolism rate, which is consistent with the core indicator for characterizing the metabolic rate in clinical pharmacokinetics and conforms to the actual situation.

[0029] Furthermore, the metabolic capacity scores and drug metabolism rates corresponding to all data sets are trained to obtain a relational model. The relational model is constructed by combining the correspondence between metabolic capacity scores and drug metabolism rates with a regression algorithm. If the metabolic capacity score is input, the drug metabolism capacity is output.

[0030] Furthermore, the training steps for the relational model include: The data sets in a single standard historical medical dataset are divided into training and validation sets in a 7:3 ratio. Stratified sampling is used in the partitioning process. Construct a traditional machine learning regression model, initialize the regression coefficients and intercept term in the linear regression model, and set the learning rate and number of iterations for the traditional machine learning regression model. In the training set, the metabolic capacity score of each data group is used as the input feature, and the drug metabolism rate is used as the output label. The input is fed into the traditional machine learning regression model. The mean squared error is used as the loss function. The gradient descent method is combined to minimize the error between the predicted drug metabolism rate and the historical metabolism rate in the standard historical medical dataset. The model is iteratively optimized based on gradient descent until the loss function converges to obtain the relational model. The relational model is validated using the validation set. If the deviation between the predicted drug metabolism rate output by the relational model and the historical drug metabolism rate in the validation set is less than a preset deviation value, the model is deemed qualified. If the deviation between the predicted drug metabolism rate output by the relational model and the historical drug metabolism rate in the validation set is greater than or equal to the preset deviation value, the model is deemed unqualified. In this case, the learning rate and number of iterations of the traditional machine learning regression model are adjusted and the model is retrained. The deviation is the average absolute percentage error between the predicted drug metabolism rate and the historical drug metabolism rate.

[0031] Specifically, in response to the existence of drug competition, the drug metabolism rates corresponding to each drug name in the treatment regimen are adjusted according to the drug competition relationship. Based on the adjusted drug metabolism rates corresponding to each drug name in the treatment regimen, the predicted dosage for each drug name is determined, including: Based on big data, information on drug interactions is collected. An interaction table is constructed with rows and columns representing drug types. In each cell, after determining the mutual influence of drugs based on the corresponding row and column, an absorption efficiency correction factor for the corresponding drug type is determined. The absorption efficiency correction factor is the ratio of the absorption rate of the drug types in combination to the absorption rate of the drug types in single administration. Calculate the enzyme competition effect correction factor for each drug based on the enzyme's affinity for each drug in which there is a drug competition relationship. The product of the absorption efficiency correction factor and the enzyme competition effect correction factor is used as the correction coefficient. The product of the correction coefficient and the drug metabolism rate output by the relation model is used as the final drug metabolism rate. The predicted dosage is calculated based on the target blood drug concentration range of the final drug metabolism rate and the dosing interval.

[0032] Taking the combined use of nifedipine and metoprolol as an example, big data analysis revealed a mutual influence between the two drugs (i.e., competition for resources in absorption rate and metabolic enzymes during combined use; this process can be obtained through existing medical data, instructions for use, and drug structures). Clinical data from 500 cases of combined use of nifedipine and metoprolol, as well as individual use, were collected, and absorption efficiency was statistically analyzed, yielding the following data: The clinical data of nifedipine monotherapy were preprocessed to remove outliers, and the average absorption rate was taken as the absorption rate of nifedipine monotherapy, which was found to be 1.2 mg / h. Clinical data on the combined use of nifedipine and metoprolol were preprocessed to remove outliers, and the average absorption rate of nifedipine was taken as the absorption rate of nifedipine and metoprolol in combination, which was found to be 1.02 mg / h. When nifedipine is used in combination with metoprolol, the correction factor for nifedipine is: 1.02 / 1.2 = 0.85.

[0033] Similarly, when nifedipine and metoprolol are used in combination, the correction factor for metoprolol is: 0.72 / 0.80=0.90.

[0034] After calculation, in the interaction table, the cell with nifedipine on the ordinate and metoprolol on the abscissa is defined as the effect of nifedipine on metoprolol absorption, and the corresponding data for this cell is 0.85. The cell with metoprolol on the ordinate and nifedipine on the abscissa is defined as the effect of metoprolol on nifedipine absorption, and the corresponding data for this cell is 0.90. That is, in response to combined drug use, the correction factor for the effect of nifedipine on metoprolol is 0.85, and the correction factor for the effect of metoprolol on nifedipine is 0.90. This process is repeated to obtain the interaction table for all drugs in combined drug use. If the number of drugs with mutual influence is greater than two, for each individual drug A, the values ​​corresponding to the effects of the other drugs on drug A in the interaction table are assigned corresponding weight coefficients. The values ​​of the weight coefficients are configured by the system, the attending physician, or the actual situation. The sum of the products of the values ​​corresponding to the effects of the other drugs on drug A and their corresponding weight coefficients is calculated and used as the final absorption efficiency correction factor for drug A.

[0035] Furthermore, based on the enzyme's affinity for each drug with which it competes for efficacy, the enzyme competition effect correction factor for each drug is calculated, where: The enzyme competition effect correction factor is calculated using the following formula: ; in: This is the enzyme competition effect correction factor for a single drug i that has a drug competition relationship. This refers to the enzyme affinity of an enzyme for a single drug i that is in competition with it for drug use. This represents the sum of enzyme affinities for all single drugs that compete with each other, and the number of n is equal to the number of drugs that compete with each other for the same enzyme.

[0036] It should be noted that enzyme affinity represents the degree of affinity of the same enzyme for different drugs. The higher the enzyme affinity, the more competitive the drug is against the enzyme. The enzyme affinity of a drug in competition is only related to the structure of the drug itself and can be obtained directly from experiments or indirectly through big data queries.

[0037] In detail, by establishing an interaction table, this invention can correct the drug metabolism rate when using combination therapy, thereby determining the dosage of combination therapy based on the corrected drug metabolism rate, effectively improving the safety of combination therapy and avoiding the safety risks caused by neglecting drug interactions in existing individualized drug prediction in the field of combination therapy.

[0038] Furthermore, the predicted dosage is determined based on the patient's drug metabolism rate, including: The predicted dosage is calculated based on the drug metabolism rate, target blood drug concentration range, and dosing interval output by the relational model.

[0039] It should be noted that the target blood drug concentration range and dosing interval can both be obtained from the treatment plan. The predicted dosage is calculated based on the drug metabolism rate, target blood drug concentration range, and dosing interval output by the relational model. This step is directly calculated using existing pharmacokinetic methods and is therefore not elaborated upon here.

[0040] Furthermore, the medical data undergoes preprocessing, which includes data screening and standardization. Data filtering occurs in response to any of the following conditions being met: The dataset contains incomplete data on gene expression parameters, liver metabolism parameters, blood drug concentration data, and clinical drug safety assessment data. In the dataset, the sequencing depth of gene expression parameters is lower than the preset depth; In clinical drug safety assessments, patients experienced moderate or severe adverse reactions; In this embodiment of the invention, the preset depth is 30×.

[0041] Standardization processes include: Gene expression parameter data are numerically encoded and converted into a unified numerical data type. The biochemical indicators in liver metabolic parameters are standardized to obtain metabolic parameter values; Outlier detection and correction were performed on the blood drug concentration data, and interpolation was used to complete the blood drug concentration data for missing time points.

[0042] In this embodiment of the invention, gene expression parameter data are numerically encoded. Wild-type, heterozygous mutant, and homozygous mutant SNPs of the CYP3A4 gene are encoded as 0, 1, and 2, respectively, thus converting them into uniform numerical data. Liver metabolic parameter data, including alanine aminotransferase (ALT) and aspartate aminotransferase (AST), are standardized using a min-max method. By defining a specific numerical range and the relationship between the value and the score, unitless metabolic parameter values ​​are obtained based on the values ​​of ALT and AST within the defined range.

[0043] The information involved in this application (including but not limited to historical medical data) has been authorized by the user or by all parties, and the collection, use, and processing of such data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the "patient gene expression parameter data" and "liver metabolic parameter data" involved in this disclosure were obtained with full authorization.

[0044] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0045] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A personalized medication prediction method based on artificial intelligence, characterized in that, include: Step S1: Obtain historical medical datasets, preprocess the medical data, and divide the preprocessed historical medical datasets into several standard historical medical datasets according to the names of the medications used. The historical medical dataset includes several data groups, each containing: patient gene expression parameter data, liver metabolism parameter data, blood drug concentration data, and corresponding clinical drug safety assessment data; the preprocessing includes: data screening and standardization. Step S2: For a single data group in a single standard historical medical dataset, a machine learning multi-feature fusion weighting strategy is used to calculate the metabolic capacity score based on the gene expression parameters and the liver metabolic parameters, and to calculate the patient's drug metabolism rate based on the blood drug concentration data. This step is repeated until the metabolic capacity score and drug metabolism rate corresponding to all data groups are obtained. Step S3: Based on the metabolic capacity scores and drug metabolism rates corresponding to all data groups, a traditional machine learning regression algorithm is used to train and model the relationship model. Step S2 and this step are repeated until the relationship model corresponding to all standard historical medical datasets is obtained. Step S4: Obtain the patient's liver metabolic parameters, gene expression parameters, and the names of each drug in the treatment plan. Determine the corresponding relationship model based on each drug name. Calculate the metabolic capacity score based on the liver metabolic parameters and blood drug concentration parameters. Input the metabolic capacity score into the relationship model to obtain the drug metabolism rate corresponding to each drug name for the patient. Step S5: Determine the method for determining the corresponding predicted dosage based on whether there is a competitive relationship between the types of drugs in the treatment plan. In response to the existence of a drug competition relationship, the drug metabolism rate corresponding to each drug name in the treatment plan is corrected according to the drug competition relationship. Based on the corrected drug metabolism rate corresponding to each drug name in the treatment plan, the predicted dosage corresponding to each drug name is determined. In response to the absence of a drug competition relationship, the predicted dosage of each drug is calculated based on the drug metabolism rate corresponding to each drug name in the treatment plan, the target blood concentration range of each drug, and the dosing interval.

2. The personalized medication prediction method based on artificial intelligence according to claim 1, characterized in that, The gene expression parameter data includes at least two of the following: single nucleotide polymorphism site data of drug-metabolizing enzyme genes, gene copy number variation data, and quantitative gene expression level data. The liver metabolic parameter data include at least two of the following: liver function biochemical indicators, liver imaging scores, and liver reserve function scores.

3. The personalized medication prediction method based on artificial intelligence according to claim 2, characterized in that, The calculation of metabolic capacity score based on the gene expression parameters and the liver metabolic parameters includes: Assign a corresponding weight coefficient to each parameter in the gene expression parameters, calculate the sum of the products of each parameter in the gene expression parameters and its corresponding weight coefficient, and record it as the gene expression score; Assign a corresponding weight coefficient to each parameter in the liver metabolic parameters, calculate the sum of the products of each parameter in the liver metabolic parameters and its corresponding weight coefficient, and record it as the liver metabolic score; The product of the gene expression score and the liver metabolism score is used as the metabolic capacity score.

4. The personalized medication prediction method based on artificial intelligence according to claim 1, characterized in that, The calculation of the patient's drug metabolism rate based on blood drug concentration data includes: The blood drug concentration data were fitted to obtain a blood drug concentration-time curve; The systemic clearance rate of the drug is obtained from the blood drug concentration-time curve and used as the patient's drug metabolism rate.

5. The personalized medication prediction method based on artificial intelligence according to claim 1, characterized in that, The metabolic capacity score and drug metabolism rate corresponding to all data groups are trained to obtain a relational model. The relational model is constructed by combining the correspondence between metabolic capacity score and drug metabolism rate with a regression algorithm. If the metabolic capacity score is input, the drug metabolism capacity is output.

6. The personalized medication prediction method based on artificial intelligence according to claim 5, characterized in that, The training steps for the relation model include: The data sets in a single standard historical medical dataset are divided into training and validation sets in a 7:3 ratio. Stratified sampling is used in the partitioning process. Construct a traditional machine learning regression model, initialize the regression coefficients and intercept term in the linear regression model, and set the learning rate and number of iterations for the traditional machine learning regression model. The metabolic capacity score in each data group of the training set is used as the input feature, and the drug metabolism rate is used as the output label. The input is fed into the traditional machine learning regression model. The mean squared error is used as the loss function. The gradient descent method is combined to minimize the error between the predicted drug metabolism rate and the historical metabolism rate in the standard historical medical dataset. The model is iteratively optimized based on gradient descent until the loss function converges to obtain the relational model. The relational model is validated based on the validation set. If the deviation between the predicted drug metabolism rate output by the relational model and the historical drug metabolism rate in the validation set is less than a preset deviation value, the model is deemed qualified. If the deviation between the predicted drug metabolism rate output by the relational model and the historical drug metabolism rate in the validation set is greater than or equal to the preset deviation value, the model is deemed unqualified. In this case, the learning rate and number of iterations of the traditional machine learning regression model are adjusted and the model is retrained. The deviation value is the average absolute percentage error between the predicted drug metabolism rate and the historical drug metabolism rate.

7. The personalized medication prediction method based on artificial intelligence according to claim 1, characterized in that, In response to the existence of drug competition, the drug metabolism rate corresponding to each drug name in the treatment regimen is corrected according to the drug competition relationship, and the predicted dosage corresponding to each drug name is determined based on the corrected drug metabolism rate in the treatment regimen, including: Based on big data, information on drug interactions is collected. An interaction table is constructed with rows and columns representing drug types. In each cell, after determining the mutual influence of drugs based on the corresponding row and column, an absorption efficiency correction factor for the corresponding drug type is determined. The absorption efficiency correction factor is the ratio of the absorption rate of the drug types in combination to the absorption rate of the drug types in single administration. Calculate the enzyme competition effect correction factor for each drug based on the enzyme's affinity for each drug in which there is a drug competition relationship. The product of the absorption efficiency correction factor and the enzyme competition effect correction factor is used as the correction coefficient. The product of the correction coefficient and the drug metabolism rate output by the relation model is used as the final drug metabolism rate. The predicted dosage is calculated based on the target blood drug concentration range of the final drug metabolism rate and the dosing interval.

8. The personalized medication prediction method based on artificial intelligence according to claim 7, characterized in that, The step involves calculating the enzyme competition effect correction factor for each drug based on the enzyme's affinity for each drug with which there is a drug competition relationship, wherein: The enzyme competition effect correction factor is calculated according to the following formula: ； in: This is the enzyme competition effect correction factor for a single drug i that has a drug competition relationship. This refers to the enzyme affinity of an enzyme for a single drug i that is in competition with it for drug use. This represents the sum of enzyme affinities for all single drugs that compete with each other, and the number of n is equal to the number of drugs that compete with each other for the same enzyme.

9. The personalized medication prediction method based on artificial intelligence according to claim 1, characterized in that, The step of determining the predicted dosage based on the patient's drug metabolism rate includes: The predicted dosage is calculated based on the drug metabolism rate, target blood drug concentration range, and dosing interval output by the relational model.

10. The personalized medication prediction method based on artificial intelligence according to claim 1, characterized in that, The medical data preprocessing includes data filtering and standardization. The data filtering is performed in response to any of the following conditions being met: The data set contains incomplete data on gene expression parameters, liver metabolism parameters, blood drug concentration data, and clinical drug safety assessment data. In the dataset, the sequencing depth of gene expression parameters is lower than a preset depth; In the aforementioned clinical drug safety assessment, the patient experienced moderate or severe adverse reactions; The standardization process includes: Gene expression parameter data are numerically encoded and converted into a unified numerical data type. The biochemical indicators in liver metabolic parameters are standardized to obtain metabolic parameter values; Outlier detection and correction were performed on the blood drug concentration data, and interpolation was used to complete the blood drug concentration data for missing time points.

Images