CN116206776A

CN116206776A - Prediction model of blood concentration of polygenic SNP locus mediated antipsychotic drug, construction method and application thereof

Info

Publication number: CN116206776A
Application number: CN202310216898.5A
Authority: CN
Inventors: 薛付忠; 李吉庆; 于天贵
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2023-03-02
Filing date: 2023-03-02
Publication date: 2023-06-02

Abstract

The invention belongs to the technical fields of biomedicine and detection analysis, and particularly relates to a prediction model of blood concentration of an antipsychotic drug mediated by polygenic SNP loci, and a construction method and application thereof. The invention screens genetic and non-genetic factors which influence the blood concentration of the common antipsychotic, explores the influence of related factors on the in-vivo metabolic process of the antipsychotic, and builds a group pharmacokinetics model of the antipsychotic such as risperidone, quetiapine, clozapine, olanzapine and the like based on the integration of the genetic and non-genetic factors of a nonlinear mixed effect model; meanwhile, the invention also adopts the Super Learner prediction algorithm to construct a prediction model of the steady-state trough concentration of the antipsychotic drug, and provides quantitative scientific reference for a clinician to determine the administration dosage of the schizophrenic patient, thereby having good practical application value.

Description

Prediction model of blood concentration of polygenic SNP locus mediated antipsychotic drug, construction method and application thereof

Technical Field

The invention belongs to the technical fields of biomedicine and detection analysis, and particularly relates to a prediction model of blood concentration of an antipsychotic drug mediated by polygenic SNP loci, and a construction method and application thereof.

Background

The disclosure of this background section is only intended to increase the understanding of the general background of the invention and is not necessarily to be construed as an admission or any form of suggestion that this information forms the prior art already known to those of ordinary skill in the art.

In general, after an antipsychotic therapeutic effect prediction model is formulated, a clinician often gives a lower initial dose to a patient, then observes the therapeutic effect and adverse reaction of the patient after administration to adjust the dose, and if the dose is still ineffective after the dose has been increased to an effective dose, the clinician considers changing the drugs with different action mechanisms or considers the combined administration scheme to continue the treatment. The mode of adjusting the dosage or the medication scheme according to the curative effect and the adverse reaction in the treatment process can lead to the patient needing to take a long time to achieve effective treatment, and can also lead to the occurrence of the adverse reaction due to the overhigh blood concentration, thereby delaying the treatment of the patient and increasing the psychological and economic burden of the patient and the family. Thus, identifying factors that affect the pharmacokinetics of antipsychotics, providing a quantitative personalized dosing aid to a clinician at the beginning of a patient's admission therapy, is of great clinical therapeutic interest.

Numerous studies have shown that there are significant individual differences in the metabolic processes in antipsychotics, e.g., BIGO et al found that the rate of olanzapine clearance can vary by 4-10 times from individual to individual. The factors influencing the blood concentration of antipsychotics are manifold. Non-genetic factors such as age, sex, weight, smoking, liver function, kidney function and concomitant medication have significant effects on the blood concentration of antipsychotics. The cytochrome P450 enzyme superfamily is the main metabolic enzyme of antipsychotics, and many studies have reported that the gene polymorphism sites on CYP2D6, CYP1A2 and other genes are related to the blood concentration of the antipsychotics. In addition, membrane-associated transporters coded by the ABCB1 gene are mainly expressed in kidney, liver and blood brain barriers and participate in absorption, distribution and elimination of medicines, and various researches show that the ABCB1 gene polymorphism is related to blood concentration of antipsychotics such as olanzapine, clozapine and the like. At present, in clinical practice, drug dosage guidance is mostly carried out according to the metabolic typing of cytochrome P450 metabolic enzymes, quantitative blood concentration estimation cannot be provided, and quantitative tools capable of carrying out personalized drug dosage guidance by utilizing genetic and non-genetic factors affecting blood concentration are lacking. Population pharmacokinetics (Population pharmacokinetics, PPK) combines a compartment model of classical pharmacokinetics with statistical principles, and studies the overall regularity of drug operation in the patient population and the source of inter-individual differences in drug concentration after administration of clinically relevant doses of a drug are one of the methods by which personalized drug administration can be more precisely quantified. Several studies have established population pharmacokinetics models of antipsychotics such as olanzapine and risperidone, but the models only incorporate limited non-genetic factors, and no population pharmacokinetics model or blood concentration prediction model which comprehensively considers genetic and non-genetic factors exists.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a group medicine substitution model based on the blood concentration of an antipsychotic, and a construction method and application thereof. The invention screens genetic and non-genetic factors which influence the blood concentration of the common antipsychotic, explores the influence of related factors on the in-vivo metabolic process of the antipsychotic, and builds a group pharmacokinetics model of the antipsychotic such as risperidone, quetiapine, clozapine, olanzapine and the like based on the integration of the genetic and non-genetic factors of a nonlinear mixed effect model; meanwhile, the invention also adopts a Super Learner prediction algorithm to construct a prediction model of the steady-state trough concentration of the antipsychotic drug, and provides quantitative scientific reference for a clinician to determine the administration dosage of the schizophrenic patient. Based on the above results, the present invention has been completed.

The invention is realized by the following technical scheme:

in a first aspect of the present invention, there is provided a method of constructing a pharmacokinetics model of an antipsychotic drug population, the method comprising: collecting clinical information and blood samples of a subject, and obtaining corresponding data information; adopting a nonlinear mixed effect model for analysis, specifically adopting a pharmacokinetic model of primary absorption and elimination of a chamber as a basic structure model, and obtaining blood concentration simulation data to obtain corresponding model pharmacokinetic parameters; the random effect model is an addition model, covariates are introduced, model fitting is carried out, the influence of different factors on the pharmacokinetic parameters is investigated, and an antipsychotic drug population pharmacokinetics model is constructed;

Wherein the antipsychotic agent includes, but is not limited to risperidone, olanzapine, clozapine, and quetiapine.

The covariates include genetic and non-genetic factors, and can be quantitatively screened and evaluated using a linear model.

When the drug is risperidone, the covariates include, but are not limited to, body weight, smoking, drinking habits, combination of the Lianhua qingwen granule and silybin meglumine tablet, glomerular filtration hypofunction, and CYP2D6 enzyme activity score;

typical values of the population of the final models Ka, V, and CL are respectively: ka= 6.113h ^-1 V=25.694 x p (0.022 x body weight), cl=3.139 x p (0.155 x smoking-0.1158 x drinking-0.280 x glibenchmarking granules+0.473 x silybin meglumine tablets-0.141 x glomerular filtration hypofunction+0.066 x cyp2d6 enzyme activity score).

When the drug is olanzapine, the covariates include, but are not limited to, glutamic oxaloacetic transaminase, reduced glomerular filtration function, combined sodium valproate, oxcarbazepine, and SNP sites rs7916649, rs12768009.

Typical values of the population of the final models Ka, V, and CL are respectively: ka=0.144 h ^-1 V=71.197l, cl=6.120 exp (0.009 glutamic oxaloacetic transaminase-0.130 glomerular filtration hypofunction +0.253 sodium valproate +0.774 oxcarbazepine +0.292 rs7916649-0.155 rs12768009).

When the drug is clozapine, the covariates include, but are not limited to, age, sex, combined propranolol and SNP sites rs1135840 and rs1135822.

Typical values of Ka, V and CL in the final model are: ka=0.558 h ^-1 V=181.874 x p (1.371 x male), cl=42.472 x p (0.239 x male-0.009 x age-0.137 x propranolol-0.148 x rst1135840-0.474 x rst1135822).

When the drug is quetiapine, the covariates include, but are not limited to, age, sex, atorvastatin calcium combination, endogenous creatinine clearance, and rs2242480.

Typical values of the population of the final models Ka, V, and CL are respectively: ka=0.163 h ^-1 V=274.626l, cl=66.389 x p (-0.195 x female-0.007 x age +0.004 x intra-myogenic anhydride clearance-0.599 x atorvastatin calcium-0.119 x rs2242480).

It should be noted that, the key point of establishing the group pharmacokinetic model is that the selected index can accurately quantify and correctly reflect the relationship between the selected index and the disease treatment, and the index with obvious dose-effect relationship can only establish a good group pharmacokinetic model, and not randomly select one index related to the pathological or physiological condition of the disease and the drug effect as the covariate. According to the invention, a clinical large sample is used as a data base, reasonable indexes are selected from a plurality of indexes through scientific analysis to comprehensively quantitatively evaluate the overall efficacy, a pharmacokinetics system capable of quantitatively evaluating the participation of the medicines is scientifically designed, and the scientific accuracy of the index system is verified through a group pharmacokinetics model, so that the dynamic change of the overall efficacy space along with time can be effectively reflected.

In a second aspect of the invention, there is provided the use of an antipsychotic drug population pharmacokinetic model obtained by the above-described construction method in any one or more of the following:

(a) Preparing a product for accurately predicting the administration dosage of an antipsychotic;

(b) Preparing the personalized administration product of the antipsychotic.

(c) Is used for relevant basic research such as pharmacokinetics of antipsychotics and the like.

Wherein in said (c), said antipsychotic pharmacokinetics comprises an antipsychotic plasma concentration prediction.

In a third aspect of the present invention, there is provided an antipsychotic drug blood concentration predictive assessment system, the system comprising at least:

an acquisition unit configured to: acquiring subject related data information;

a data processing unit configured to: predicting the blood concentration of the antipsychotic drug of the subject based on the data information obtained by the obtaining unit and based on the built-in predictive evaluation model; the predictive evaluation model is obtained by training the pre-collected relevant data information of the patient by adopting a statistical algorithm.

An output unit configured to: and outputting the predicted blood concentration value of the antipsychotic of the subject according to the information of the data processing unit.

Wherein the subject-related data includes, but is not limited to, subject general conditions, disease characteristics, antipsychotic medication conditions, and antipsychotic efficacy-related SNP sites;

the antipsychotic agents include, but are not limited to, risperidone, clozapine, olanzapine, and quetiapine.

When the antipsychotic drug is risperidone, the relevant data information includes, but is not limited to, dosage, weight, BMI, age, endogenous creatinine clearance, rs17327442, CYP2D6 enzyme activity score, sodium valproate, propranolol, rs7787082, benzofuranone, rs3789243, rs4244285, rs11528090, rs7779562, geocentric blood health soft capsule, rs762551, rs3743484, rs2235047, smoking, rs1135840, rs28371699, rs1058164, sex, back-racetam capsule, alcohol drinking, rs1081003, rs1065852, rs58440431, rs2004511, rs16947, rs75276289, rs1080996, silybin meglumine tablet, rs116917064, aspirin enteric tablet.

The prediction model is a 9-hydroxyrisperidone blood concentration prediction model constructed by adopting a super-strong learning integration algorithm.

In the process of constructing the predictive evaluation model, the relevant data information comprises, but is not limited to, dosage, frequency of administration, age, alkaline phosphatase, body weight, glutamic-pyruvic transaminase, glutamic-oxaloacetic transaminase, sodium valproate, CYP2D6 enzyme activity score, rs1065852, rs762551, rs3743484, aripiprazole, rs11528090, rs35280822, rs16947, rs79331140, rs1080996, rs7787082, propranolol, rs7779562, rs75276289, rs4646437, rs6583954, rs3758580, rs4244285, smoking, drinking wine, sex, rs12768009, rs7916649, rs2470890, rs17884832, rs5030865, rs17879992, rs17885098, 28371725, glomerular filtration hypofunction, phenolphthalein tablet, sulpiride, ziprasidone and radix Sanguisorbae white tablet.

The prediction model is an olanzapine blood concentration prediction model constructed by adopting a super-strong learning integration algorithm.

In the process of constructing the predictive evaluation model, the relevant data information comprises, but is not limited to, dosage, creatinine, age, glutamic-oxaloacetic transaminase, creatinine clearance, body weight, propranolol, risperidone, rs1080996, rs75276289, rs16947, rs11528090, rs762551, sex, frequency of administration, rs12535512, back-lasracetam capsule, rs1135840, rs1128503, rs12720464, rs3789243, CYP2D6 enzyme activity score, aspirin enteric tablet, rs3842, rs6979885, atorvastatin calcium tablet, rs4728709, and smoking.

The prediction model is a model for constructing the blood concentration prediction model of the clozapine by adopting a super-strong learning integration algorithm.

In the process of constructing the predictive evaluation model, the relevant data information comprises, but is not limited to, dosage, age, BMI, alkaline phosphatase, glutamic-oxaloacetic transaminase, creatinine, glutamic-pyruvic transaminase, weight, sodium valproate, CYP2D6 enzyme activity score, rs12535512, rs2246709, rs12768009, rs2242480, rs1081003, cefuroxime axetil, amisulpride, rs58440431, rs2004511, dioxinzkang soft capsule, sex, enalapril maleate tablet, atorvastatin calcium tablet, rs2470890, smoking, liver function abnormality, rs1135822, oxcarbazepine, buspirone, aripiprazole and heart stabilizing particles.

The prediction model is a quetiapine blood concentration prediction model constructed by adopting a super-strong learning integration algorithm.

In a fourth aspect of the invention, there is provided a computer readable storage medium having stored thereon a program which when executed by a processor performs the functions of the system according to the third aspect of the invention.

In a fifth aspect the present invention provides an electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the functions of the system according to the fourth aspect of the invention when the program is executed.

The beneficial technical effects of one or more of the technical schemes are as follows:

the technical scheme screens genetic and non-genetic factors influencing the blood concentration of the common antipsychotic drugs, explores the influence of related factors on the in-vivo metabolic process of the antipsychotic drugs, and builds a population pharmacokinetics model of the antipsychotic drugs such as risperidone, quetiapine, clozapine, olanzapine and the like based on the integration of the genetic and non-genetic factors of a nonlinear mixed effect model; meanwhile, the invention also adopts the ultra-strong learning integration algorithm to construct a prediction model of the steady-state trough concentration of the antipsychotic drug, and provides quantitative scientific reference for a clinician to determine the administration dosage of the schizophrenic patient, thereby having good practical application value.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.

FIG. 1 is a schematic representation of the time points for risperidone plasma concentration sampling in an example of the present invention.

FIG. 2 is a plot of goodness-of-fit of risperidone final model in an example of the present invention.

FIG. 3 is a residual diagram of a risperidone final model in an example of the invention.

FIG. 4 is a graph showing the pharmacokinetics of the final model population for a risperidone dosing regimen of 2mg (bid) in the examples of the present invention (2.5%, 50% and 97.5% scores for predicted plasma levels from top to bottom, respectively; grey circles are actual observations).

FIG. 5 is a comparison of RMSE for different amounts of risperidone candidates in the examples of the invention.

FIG. 6 is a graph showing predicted variable importance scores for risperidone based on random forests in an embodiment of the invention.

FIG. 7 is a scatter plot of predicted values and actual observed values of a 9-hydroxy risperidone blood pressure concentration prediction model based on a Super Learner in an embodiment of the present invention.

Fig. 8 is a plot of goodness-of-fit of olanzapine final model in an embodiment of the present invention.

Fig. 9 is a residual plot of the olanzapine final model in an embodiment of the present invention.

FIG. 10 is a graphical representation of the final model population pharmacokinetics at 5mg (bid) of olanzapine dosing regimen in the example of the present invention (2.5%, 50% and 97.5% quantiles from top to bottom, respectively, of predicted plasma concentration values; grey circles are actual observations).

Fig. 11 is RMSE based on a random forest prediction model for different numbers of candidate variables of olanzapine in an embodiment of the invention.

Figure 12 is a graph of predicted variable importance scores for olanzapine based on random forests in an embodiment of the present invention.

FIG. 13 is a plot of the predicted value of the olanzapine plasma concentration prediction model based on Super Learner versus the actual observed value scatter (A: training set ten-fold crossing; B: validation set) in an embodiment of the present invention.

Fig. 14 is a plot of goodness-of-fit of olanzapine final model in an embodiment of the present invention.

Fig. 15 is a residual plot of the olanzapine final model in an embodiment of the present invention.

FIG. 16 is a graphical representation of the final model population pharmacokinetics at 75mg (bid) of clozapine dosing regimen in an example of the invention (2.5%, 50% and 97.5% quantiles from top to bottom, respectively, of predicted plasma concentration values; grey circles are actual observations).

FIG. 17 is a comparison of RMSE for different numbers of candidate variables for clozapine in an example of the invention.

Figure 18 is a graph showing predicted variable importance scores for clozapine based on random forests in an embodiment of the invention.

FIG. 19 is a graph of predicted values and actual observed values of a model for predicting the blood concentration of clozapine based on a Super Learner in an embodiment of the present invention (A: training set ten-fold crossing; B: validation set).

FIG. 20 is a plot of goodness-of-fit of the quetiapine final model in an example of the present invention.

Fig. 21 is a residual plot of a quetiapine final model in an embodiment of the present invention.

FIG. 22 is a graphical representation of the final model population pharmacokinetics at 300mg (bid) in the quetiapine regimen of the present invention (2.5%, 50% and 97.5% quantiles from top to bottom, respectively, of predicted plasma concentration values; gray circles are actual observations).

FIG. 23 is a comparison of RMSE for different numbers of candidate variables for quetiapine in the examples of the invention.

FIG. 24 is a graph showing predicted variable importance scores for quetiapine based on random forests in an embodiment of the present invention.

FIG. 25 is a plot of predicted values and actual observed values of a quetiapine plasma concentration prediction model based on Super Learner (A: training set ten-fold crossing; B: validation set) in an embodiment of the present invention.

Detailed Description

It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, 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 belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof. It is to be understood that the scope of the invention is not limited to the specific embodiments described below; it is also to be understood that the terminology used in the examples of the invention is for the purpose of describing particular embodiments only, and is not intended to limit the scope of the invention. Experimental methods in the following embodiments, unless specific conditions are noted, are generally in accordance with conventional methods and conditions of biology within the skill of the art, and are fully explained in the literature.

The invention is further illustrated by the following examples, which are not to be construed as limiting the invention. It is to be understood that these examples are illustrative of the present invention and are not intended to limit the scope of the present invention. In addition, biomedical methods not described in detail in the examples are all conventional in the art, and specific operations may be referred to biomedical guidelines or product specifications.

Examples

1. Materials and methods

1.1 study object

The study was approved by the ethics committee of the university of Shandong, public health school. Subject group in the mountain eastern Yi city, pingyi county psychological hospitals incorporated patients with schizophrenia hospitalized in 2021 to 2022, specifically incorporated exclusion criteria were as follows:

inclusion criteria: (1) International statistical classification of diseases and related health problems (revision 10) (ICD-10) clinical diagnosis criteria for schizophrenia; (2) age 18-65 years; (3) The antipsychotic drugs such as risperidone, quetiapine, olanzapine or clozapine are taken according to the doctor's advice rule during the period of hospitalization; (4) can accept the regular blood concentration monitoring;

exclusion criteria: (1) Suffering from chronic enteritis, diarrhea, or other gastrointestinal disorders; (2) Has other important diseases such as serious cardiovascular and cerebrovascular diseases, cancer, etc.; (3) complete case data cannot be obtained; (4) extremely poor compliance and incapability of regularly taking medicines for a long time; (5) long-term drug injection is required to maintain therapeutic compliance; (6) The recent treatment with clozapine has been regularly carried out for therapeutic resistance (except for patients who have been treated for other reasons with clozapine).

1.2 data collection

The basic information of schizophrenic patients collected from psychological hospitals in Pingyi county in this study included: (1) general demographic characteristics: age, sex, height, weight, educational level, marital status; (2) co-morbidity: diabetes, hypertension, coronary heart disease, cerebral apoplexy and other diseases; (3) lifestyle such as smoking and drinking; (4) features of schizophrenia: first onset date, disease course characteristics (acute, subacute, slow onset), family history of mental diseases, number of hospitalizations, number of days of hospitalization, etc.;

The study obtains the medical treatment condition of the patients with schizophrenia during the hospitalization period from the hospitalization doctor advice part in the electronic medical record system of the mental health center in Shandong province, wherein the treatment medicaments comprise antipsychotic medicaments such as risperidone, olanzapine, quetiapine, clozapine, amisulpride, aripiprazole, ziprasidone and the like; mood stabilizers such as lithium carbonate, sodium valproate, magnesium valproate, etc.; antidepressants such as citalopram, mirtazapine, duloxetine, etc.; bupiridone, lorazepam and other anxiolytic drugs. The collected treatment information includes: drug name, date of administration start and end, frequency of administration, dose administered, etc.

The study obtained laboratory test data during hospitalization of patients with mental disorders from the Pingyi county mental hospital laboratory information System (List), including liver function index: glutamic-pyruvic transaminase (AST), alkaline phosphatase (ALP); renal function index: creatinine (Cr), urea nitrogen/creatinine ratio (BUN/Cr); blood routine index: white blood cell count; neutrophil count, hemoglobin count, platelet count.

1.3 blood sample collection

1.3.2 blood concentration blood sample collection

1.3.2.1 regular blood concentration monitoring blood sample collection

2 ml of anticoagulated purple tube venous blood was collected for each case, typically at 6:30 a.m. blood samples were collected. In principle, it is required that the target drug has reached steady state, i.e. after 5-6 half-lives of continuous administration of a fixed dose. The half-life of each drug and the therapeutic reference concentration ranges are shown in table 1.

Table 1 half-life and therapeutic concentration reference ranges for four anti-spermatic drugs

1.3.2.2 population pharmacokinetics blood sample collection

The study describes the process of determining the blood sample collection and time of population pharmacokinetics using risperidone as an example. Referring to the related literature, a risperidone group medicine generation study room model is set as a one-room model primary absorption, and initial values of group medicine generation parameters are set as follows: cl=4.6L/h, k _a ＝3.1h ^-1 V=250l. According to the medicine taking mode of the single medicine taking of 8:00 and 20:00, under the condition that the blood concentration reaches a steady state, the recommended sampling time points of the blood concentration obtained by using the R-pack PopED optimized by the group pharmacokinetics test design are respectively as follows: 0.35h, 2h and 12h after taking the medicine. However, in clinical practice, patients with mental schizophrenia in Pingyi psychological hospitals generally take medicine for 10:30 a.m. and 20:00 a.m., and the ideal sampling requirements are difficult to achieve.

To adapt to clinical practice, and to increase the operability of the study protocol, we combined blood sample concentration sampling with other relevant test samples, the following two sampling protocols were set: after the blood concentration reached a steady state (risperidone and 9-hydroxyrisperidone were generally considered to reach a steady state of blood concentration after the same dose was continuously administered for 6 days), samples were taken at 6:30 and 9:30 am, i.e., 10.5h and 13.5h after 20:00 pm had been taken; samples were taken once at 11:30 noon after the last 10:30 dose. Considering that multiple samplings on the same day may lead to reduced patient compliance, blood sample collection may not be completed on the same day with the patient taking the drug daily and the drug taking time unchanged, e.g., after the patient reaches a steady state of blood concentration, a first blood sample is collected on the first day at 6:30 am, a second blood sample is collected on the following day at 9:30 am, and a third blood sample is collected on the following day at 11:30 am. The schematic diagram of the sampling point is shown in fig. 1.

Table 2 blood concentration sampling time points for antipsychotics

1.4 blood concentration determination

High Performance Liquid Chromatography (HPLC) is used for measuring the blood concentration of the antipsychotic. The FLC full-automatic two-dimensional liquid chromatography system and the high performance liquid chromatograph, mobile phase, chromatographic column, deproteinizing agent and the like matched with the FLC full-automatic two-dimensional liquid chromatography system are provided by Hunan De Mit instruments.

1.5 candidate SNP site selection and SNP site typing

Based on genes and SNP sites related to metabolism of antipsychotics listed in the pharmacogenetic and genomic pharmacological databases (PharmGKB), genes of metabolic enzymes such as CYP2D6, CPY3A4, CYP1A2, CYP3A5 and CYP2C19 and genes of ABCB1 transporters are determined as candidate genes in combination with pharmacological research literature, and SNP sites in the genes are selected as candidate SNP sites. Wherein all SNP loci related to drug metabolism reported in the literature in PharmGKB or single nucleotide polymorphism database (SNPedia) are selected; while the other sites of the candidate gene, if they are in the linkage disequilibrium (linkage disequilibrium, LD) region (r ² >0.8 A SNP with the highest Minimum Allele Frequency (MAF) is included as a candidate marker SNP. Genotyping was performed using ASA (Asian Screening Array) chip from Illumina corporation.

1.6 data sorting and analysis

1.6.1 data sort

Because the original data come from a plurality of information systems, for the convenience of statistical analysis, the basic information of a study object, the data of an inpatient doctor's advice, the data of monitoring the blood concentration of an antipsychotic drug and the genotyping data are combined by taking the inpatient number as a unique identification code. The antipsychotic drugs such as risperidone, olanzapine, quetiapine and clozapine and other combined doses are defined as follows: when the blood concentration of the target antipsychotic is detected, the required dosage in the prescription is ordered, the dosage units of the medicine are converted into mg/d, and the administration frequency of the antipsychotic is recorded. The other variable codes are shown in Table 3. The study fills up missing variables with a missing rate of less than 5% by using a multiple interpolation method.

Table 3 variable encoding table

Currently, in clinical practice, clinicians conduct medication guidance according to the metabolic typing of CYP2D6 gene, and for convenience of clinical application, the study refers to the metabolic typing method [ a1a, metabolic typing method of CYP2D6 gene in Chinese population related to psychotropic drugs ] based on CYP2D6 gene, which is proposed by the teachings of CPIC and Cui Yimin, and performs metabolic typing on study objects in the study. According to the genotyping results of SNP loci such as rs3892097, rs5030865, rs28371706, rs28371717, rs769258, rs28371725, rs1135822, rs72549348, rs267608297, rs267608289, rs1065852, rs16947, rs1135840, rs28371699 and rs28371701, the SNP locus alleles are marked as follows: a) Normal function alleles CYP2D6 x 1, # 2, # 33, # 35, # 39; b) Reduced function alleles CYP2D6, 9, 10, 14,

* 17. 29, 41, 49, 54, 59; c) Non-functional alleles CYP2D6 x 3, x 4, x 5, x 6, x 8, x 11, x 51, x 59; d) Alleles with unknown function CYP2D6 x 22, # 28, # 30, # 65. After determining the number of four functional allele mutations, the following formula is introduced to calculate an enzyme Activity Score (AS): as=a+d+0.5×b (where the 10 coefficient is 0.25).

1.6.2 data description

The study was statistically described in two ways: (1) General demographic characteristics, lifestyle habits, co-morbid and psychotic conditions of the subject are described overall; (2) Antipsychotic drug dosage and blood concentration distribution. If the continuous variable accords with normal distribution, the continuous variable is represented by mean value +/-standard deviation; if the normal distribution is not met, the continuity variable is expressed by a median (P25-P75 quartile); the classification variables are expressed by frequency numbers and composition ratios.

1.6.3 analysis of blood concentration influencing factors

Because the study subjects conduct blood concentration monitoring for multiple times in the hospitalization process, the study data are longitudinal data of repeated multiple measurements, and therefore, the study adopts a generalized estimation equation (Generalized estimating equations, GEE) model to screen non-genetic factors and SNP loci which influence the blood concentration of the antipsychotic. And simultaneously taking the dosage of the antipsychotic, the time interval from the previous dosage and the candidate variable into a GEE model, and analyzing the influence of the candidate variable on the blood concentration of the antipsychotic.

1.6.4 group pharmaceutical analysis

In order to further analyze the influence of the influence factors of the blood concentration of the antipsychotic on the metabolism of the antipsychotic, the study adopts a nonlinear mixed effect model to carry out group pharmacokinetics analysis. The basic flow of the population pharmacokinetics model construction is as follows:

the nonlinear mixture effect model can be represented by the following general formula:

wherein Y is _ij Is the j-th observed value of the concentration of the ith individual, X _ij Is the independent variable (such as observation time and dose) of a certain individual, and the pi-t-dimensional regression design matrix in the individual represents t independent in-individual variables;

is all pharmacokinetic parameters of a certain individual, epsilon _ij ：P _ij ^* 1 normal random error vector. f is Pi X1-dimensional X _ij And->

The functional form of which depends on the selected pharmacokinetic structural model;

(1) Determining a pharmacokinetic structural model:

the pharmacokinetic structural model describes the average effect of the pharmacokinetic population, and the differences in the route of administration and the in vivo process of the drug determine the diversity of the pharmacokinetic model, all of which can be described by a set of formulas and parameters.

Taking a single administration as an example, a pharmacokinetic model of primary absorption and elimination in a compartment can be expressed in the following form:

at time point t _j The cumulative dose formula is calculated as follows:

D[i，j]＝D[i，j-1]exp(-k _a t _j -t _j-1 ])+d[i，j]

At time point t _j The formula for calculating plasma concentration is as follows:

d[i，j]indicating that the ith subject was at time t _j Dosage of the medicine taken; f represents the bioavailability of the drug in the human population (the relative amount of drug absorbed into the systemic blood circulation after administration via the extravascular route). V [ i ]]Represents apparent distribution volume, K _a And K _a The first order absorption and elimination rate constants, respectively.

The pharmacokinetic structure model can be flexibly selected according to the characteristics and data characteristics of the medicine, and the pharmacokinetic structure models of risperidone, olanzapine, quetiapine and clozapine are all set as the pharmacokinetic model for primary absorption and elimination of one room.

(2) Fixed effect model: the fixed effect model is used for quantitatively examining the influence of fixed effects (such as age, sex, SNP locus and the like) on pharmacokinetic parameters, and the model structure comprises linear, multiplicative, saturated and indicating variable models. The most common at present are linear and multiplicative models.

1) Linear model: the formula is as follows:

in the middle of

Is a population typical value, θ ₁ Is a group standard value. When the fixed effect is not considered, it is equal to the population typical value; WT is patient weight, θ ₁ Is a fixed effect parameter of body weight.

2) Multiplication model: if the body weight range of the study subjects is large and the drug clearance is related to the body weight, the following formula is used:

taking the logarithm of the two sides of the formula to obtain a linear model with a logarithmic scale. In the middle of

Is a population typical value, θ ₁ Is a group standard value. WT is patient weight, θ ₁ Is a fixed effect parameter of body weight.

(3) Random effect model: the random effect model includes inter-individual variation and residual variation. The former describes the differences in pharmacokinetic parameters among individuals in a population; the latter describes the variability that the observations cannot explain. Generally, the pharmacokinetic parameters are assumed to be normal or lognormal distribution, and the model with the smallest objective function value is selected according to addition, proportion and power models. Inter-individual variation of pharmacokinetic parameters is expressed as a random variable η, with a mean of 0 and a variance of ω ² Expressed as η=n (0, ω ² ) The method comprises the steps of carrying out a first treatment on the surface of the The random variable epsilon represents the residual variation, expressed as eta = N (0, epsilon) ² ). Currently, an addition model or a proportion model is commonly used for representation:

1) Addition model: the formula is as follows:

wherein CL is _j For the true value of a certain individual parameter,

for parameter population typical values, +.>

For the concentration observations, +.>

Is a model predictor of concentration. />

ε _ij The individual and the random variation are respectively.

2) Proportion model: the formula is as follows:

/>

(3) Covariate screening: examine the influence of non-genetic factors such as age, sex, weight, combined medication and the like and genetic factors on pharmacokinetic parameters. Currently, covariate screening mainly employs hypothesis testing and forward increasing or backward decreasing. The basic principle of hypothesis testing is as follows:

the reliability of the nonlinear mixed effect model can be estimated by using a maximum likelihood method (maximum likelihood approach, ML). The maximum likelihood of a series of observations in a model can be expressed by:

the maximum likelihood may be expressed in terms of a minimum of-2 log (L). θ is the pharmacokinetic parameter of the model built, y _i Is the observed value of blood concentration, f (theta, x) _i ) In order to simulate the obtained blood concentration according to the established pharmacokinetic model, n is the number of observation points and sigma ² Representing the variance of the residual. The first part of the equation is constant, so the minimum depends on the second part. And the second part is also called an Objective Function (OFV) of the extended least squares method (extended least squares, ELS).

The maximum likelihood test, i.e., the significance test equivalent to the OFV change, employs a conventional statistical significance test method. Firstly, establishing a simplest model without any fixed effect factors, and then adding candidate covariates one by one in the simplest model to establish an incremental model. The variance of the parameters of the simplest model and the incremental model can be represented by the likelihood ratio of the two (L ₁ /L ₂ ) Has been demonstrated to be-2 log (L ₁ /L ₂ ) Following chi-square distribution, if the difference between the two model OPVs is greater than 3.84, according to the significance level of α=0.05, the candidate covariates are statistically significant for the parameter impact. And sequentially inspecting candidate fixing effects to finally obtain a final model. Forward increment, namely, on the basis of the simplest model, adding meaningful fixed effects one by taking delta OFV as a standard, and finally obtaining a full-scale regression model (Full regression model, FRM); on this basis, the final regression model is determined by using a backward progressive method to set the fixation effect to "0" one by one and eliminating the fixation effect without significance by hypothesis testing (Δ OFV).

The goodness of fit of the final population pharmacokinetic model is evaluated by RMSE and R2, and evaluated by drawing a diagnostic chart, including individual predictor versus observed value (IPRED vs DV, the closer the trend line is to y=x, the stronger the individual predictive capability of the final model), population predictor versus observed value (PRED vs DV, the closer the trend line is to y=x, the stronger the population predictive capability of the final model), weight residual versus population predictor (cwrevspred, the more biased the condition residual versus predicted value, the more biased the weight residual versus 0, the more biased the model population predictive capability), weight residual versus TIME scatter plot (cwrevs TIME, the more biased the condition weight residual versus blood sample collection TIME). The individual predictors refer to fitting results calculated from PK individual parameters of the model. The data points should be randomly and evenly distributed on both sides of a straight line with an intercept of 0 and a slope of 1. The closer the trend line is to the diagonal line, the more accurate the fit. The plot of the weight residual error to the group predicted value and the plot of the weight residual error to the time scatter can evaluate the advantages and disadvantages of the curve fitting overall, namely the weight residual error is between-6 and 6, which indicates that the model fitting effect is better. Meanwhile, the stability of the group pharmacokinetics model is verified by adopting a Bootstrap method, and the distribution of the group pharmacokinetics parameters and the stability of the model are inspected by carrying out Bootstrap verification on the model 500 times.

2. Test results

2.1 Risperidone study subject Condition and group pharmacokinetics study

2.1.1 subject base case

The study included 239 schizophrenic patients as subjects, and the general condition of the subjects and the distribution of laboratory test indicators are shown in Table 4. Subjects had an average age of 44.3±12.0 years; 132 men and 107 women, respectively accounting for 55.2% and 44.8%; the average value of the height and the weight is 163.6+/-8.7 cm and 68.2+/-13.7 kg respectively, and the average value of the BMI is 25.4+/-4.3 kg/m2; 68 smokers and 44 drinkers in the study subjects, accounting for 28.5% and 18.4% respectively; the prevalence of diabetes and hypertension is 7.9% and 6.3%, respectively. The mean values of the liver function and the kidney function detection indexes are in the normal reference value range, but 49 people have glomerular filtration hypofunction, 40 people have abnormal liver function, and the mean values respectively account for 20.5 percent and 16.7 percent.

The study included only schizophrenic patients with twice daily dosing frequency, and the average dose of each dose was 2.4+ -0.6 mg. In the steady state of blood concentration, the study collects 1115 risperidone blood concentrations, and after taking medicines for 1.5h, 5.0h, 9.5h and 12.5h, the mean blood concentration of the study subjects is 70.22+/-28.54, 53.23+/-14.68, 51.01 +/-24.33 and 44.89+/-18.43 ng/ml respectively. According to the upper limit of the reference range of risperidone treatment concentration of 60ng/ml, 244 cases of blood concentration after 9.5 hours of taking medicine and 11 cases of blood concentration after 12.5 hours of taking medicine exceed the upper limit of the reference range of risperidone treatment concentration, and the blood concentration respectively accounts for 26.5% and 23.4% of the blood concentration detection cases at the time point.

TABLE 4 general condition of study subjects and laboratory test indicators

^* In a steady state, the blood concentration is 1.5h, 5.0h, 9.5h and 12.5h away from the last administration time when the blood sample is collected; glomerular filtration function decline: creatinine clearance rate<80ml/min; liver function abnormality: glutamic-pyruvic transaminase>40U/L or glutamic oxaloacetic transaminase>40U/L；

2.1.2 Risperidone population pharmacokinetics study

In order to determine the mean value, standard deviation and discrete degree of various pharmacokinetic parameters of the schizophrenic patient population and study the influence of various factors on risperidone metabolism, risperidone population pharmacokinetics analysis is carried out in the study. And combining literature reports and the characteristics of the research data, finally determining a basic structure model as a pharmacokinetic structure model for primary absorption and elimination of a room, a random effect model as an addition model, and quantitatively examining the influence of a fixed effect (such as age, sex, SNP locus and the like) on pharmacokinetic parameters by adopting a linear model. Genetic and non-genetic factors related to population pharmacokinetic parameters screened based on a single-factor nonlinear mixed effect model are shown in table 5. The results show that with age and weight gain, the apparent distribution volume gradually increases, both of which have a statistical significance (P < 0.05) for the effect on apparent distribution volume. Genetic factors of statistical significance (P < 0.05) for the primary elimination rate effect include: rs28371699 (CYP 2D 6), rs1135840 (CYP 2D 6), rs1058164 (CYP 2D 6), rs75276289 (CYP 2D 6), rs16947 (CYP 2D 6), and rs72547513 (CYP 1 A2); non-genetic factors include: smoking, drinking, glomerular filtration hypofunction, weight, LIANHUAQINGWEN granule, silybin meglumine tablet, etc. In addition, the CYP2D6 enzyme activity score is positively correlated with the primary elimination rate, which increases with increasing CYP2D6 enzyme activity score (P < 0.05).

TABLE 5 genetic and non-genetic factors related to population pharmacokinetic parameters

^a Ka: first order absorption Rate constant (h ^-1 ) The method comprises the steps of carrying out a first treatment on the surface of the V: apparent distribution volume (L); CL: a first order cancellation rate constant (L/H);

covariates of the group pharmacokinetics model are screened by adopting forward increment and backward rejection methods, and the final group pharmacokinetics model is determined by combining relevant expertise, and the parameter estimation value of the 9-hydroxy risperidone final model is shown in Table 6. In the final population pharmacokinetic model, body weight has a significant effect on apparent distribution volume (P<0.05 A) is provided; smoking and drinking habit, combination of Lianhua Qingshen granule and silybin meglumine tablet, glomerular filtration hypofunction and CYP2D6 enzyme activity score have significant influence on primary elimination rate (P)<0.05). Typical values of the population of the final models Ka, V, and CL are respectively: ka= 6.113h ^-1 V=25.694 x p (0.022 x body weight), cl=3.139 x p (0.155 x smoking-0.1158 x drinking-0.280 x glibenchmarking granules+0.473 x silybin meglumine tablets-0.141 x glomerular filtration hypofunction+0.066 x cyp2d6 enzyme activity score). The verification result of Bootstrap 500 times shows that model parameter estimation based on measured data is basically similar to parameter estimation value based on sampling data, and the estimation value of main PK parameters obtained by a final model is in a quartile range of Bootstrap verification parameters, so that the model estimation is stable and has good reliability.

Table 6 9-final model parameter estimation and validation junctions for hydroxyrisperidone

^a CL: 9-hydroxy risperidone clearance rate ((L/H)); ka: absorption Rate (h) ^-1 ) The method comprises the steps of carrying out a first treatment on the surface of the V: apparent distribution volume (V);

the figure of merit of the final model of 9-hydroxy risperidone is shown in FIG. 2, RMSE and R for individual prediction and population prediction of 99-hydroxy risperidone ² 9.90ng/ml, 0.837 and 20.9ng/ml, 0.329, respectively. The plasma concentration observations are uniformly distributed on both sides of the trend line (y=x) and substantially coincide with the diagonal line. The residual plot of the final model of 9-hydroxy risperidone is shown in FIG. 3, which shows that the conditions are weightedThe residuals are uniformly distributed on both sides of y=0 and have no obvious correlation with time and population predictions. The final model population pharmacokinetics profile for risperidone dosing regimen at 2mg (bid) is shown in FIG. 4.

2.1.3 Risperidone conventional monitoring plasma concentration prediction

2.1.3.1 variable screening based on recursive feature elimination algorithm

We use a random forest based recursive feature elimination algorithm to screen the prediction variables for machine learning modeling. Ten fold cross validation RMSE based on random forest modeling with different numbers of candidate variables is shown in fig. 5. As can be seen from fig. 5, when the predicted variable of the first 36 bits of the variable importance ranking is included, RMSE of ten-fold cross-validation of the random forest algorithm is minimum, and then the predicted variables screened based on the recursive feature elimination algorithm of the random forest include dosage, weight, BMI, age, endogenous creatinine clearance, rs17327442, CYP2D6 enzyme activity score, sodium valproate, propranolol, rs7787082, benzomarie, rs3789243, rs4244285, rs11528090, rs7779562, geocentric blood health soft capsule, rs762551, rs3743484, rs2235047, smoking, rs1135840, rs28371699, rs1058164, gender, back-raxaracetam capsule, alcohol drinking, rs1081003, rs1065852, rs58440431, 2004511, rs 947, rs75276289, rs1080996, silybin meglumine tablet, rs116917064, aspirin enteric tablet. The prediction variables are used for constructing a prediction model based on a machine learning method such as random forest, bayesian additive regression tree and the like.

2.1.3.2 model fitting and evaluation

The study fits a prediction model based on a gradient lifting tree, a support vector machine, a random forest, a Bayesian additive regression tree and an xgboost algorithm, and simultaneously uses NNLS as a loss function, a set of weight vectors which minimize the risk of cross validation of the combination of the models are determined by using ten-fold cross validation, and the weights of the models are respectively 0.000, 0.156, 0.347, 0.243 and 0.254, so that a 9-hydroxy risperidone blood concentration prediction model based on a Super Learner is constructed, and the model is suitable for predicting the schizophrenia patient in a steady state in the morning of 6: the blood concentration of 9-hydroxy risperidone at 30 hours (the administration time is respectively 10 am and 21 pm). The individual algorithm weights and parameter settings in the Super filter are shown in Table 7.

Table 7 weights and parameter settings in Super Learner for individual algorithms

Note that: each algorithm parameter not listed in table 7 is set to the R-packet default value.

In the study, we randomly draw 300 times of conventional blood concentration monitoring records as a verification set, and the rest as a training set, and evaluate the blood concentration prediction model based on 9-basic risperidone of Super Learner. In the training set, the RMSE of the Super Learner, gradient lifting tree, support vector machine, random forest, bayesian additive regression tree, and xgBoost algorithm were 12.91ng/ml, 14.25ng/ml, 14.48ng/ml, 12.69ng/ml, 16.22ng/ml, and 11.31ng/ml, respectively. In the verification set, the RMSE for the above algorithm was 16.23ng/ml, 15.67ng/ml, 16.98ng/ml, 16.15ng/ml, 16.98ng/ml, 17.31ng/ml, respectively. R of Super Learner in training set and verification set ² 0.66 and 0.50, respectively, perform well (fig. 7).

TABLE 8 evaluation of the 9-base risperidone plasma concentration prediction model based on SuperLearner

2.2 study of olanzapine study subject base conditions and group pharmacokinetics

2.2.1 basic conditions of study object

This part of the study included 131 patients with schizophrenia treated with olanzapine as subjects, and the general condition of the subjects and the distribution of laboratory test indicators are shown in table 9. Subjects were 45.6±14.2 years of average age; 84 men and 47 women, with a ratio of 64.1% and 35.9%, respectively; the average value of the height and the weight is 163.4+/-9.2 cm and 67.4+/-14.8 kg respectively, and the average value of the BMI is 25.2+/-4.6 kg/m2; 47 smokers and 30 drinkers in the study subjects, accounting for 35.9% and 22.9% respectively; the prevalence of diabetes and hypertension is 5.3% and 9.9%, respectively. The mean values of the liver function and the kidney function detection indexes are in the normal reference value range, but 25 people have glomerular filtration hypofunction, and 22 people have abnormal liver function, and the mean values respectively account for 19.1 percent and 16.8 percent. Olanzapine was taken twice daily in 73 patients and once a night in 58 patients at 55.7% and 44.3%, respectively. The average dose of each administration is 6.8+/-2.4 mg.

In steady state plasma concentrations, 467 times of olanzapine plasma concentrations were collected in the study. After taking the medicines for 1.5 hours, 5.0 hours, 9.5 hours and 12.5 hours for patients taking the medicines twice a day, the average blood concentration values are 61.54 +/-18.32, 71.72 +/-22.99, 49.38+/-18.93 and 46.41+/-20.12 ng/ml respectively; the administration frequency is that the average value of the blood concentration is 34.61+/-15.39 after the patients take the medicine once every night for 9.5 hours. According to the upper limit of the reference range of olanzapine treatment concentration of 80g/ml, the blood concentration exceeds the upper limit of the reference range of the treatment concentration after 22 times of taking medicine for 9.5 hours, and the blood concentration at the time point accounts for 5.41% of the times of detection of the blood concentration.

TABLE 9 general condition of study subjects and laboratory test indicators

2.2.2 population pharmacokinetics studies of olanzapine

To determine the mean, standard deviation and degree of dispersion of various pharmacokinetic parameters of the schizophrenic patient population and to study the effect of various factors on clozapine metabolism, the study conducted a pharmacokinetics analysis of olanzapine population. And combining literature reports and the characteristics of the research data, finally determining a basic structure model as a pharmacokinetic structure model for primary absorption and elimination of a room, a random effect model as an addition model, and quantitatively examining the influence of a fixed effect (such as age, sex, SNP locus and the like) on pharmacokinetic parameters by adopting a linear model. Genetic and non-genetic factors related to population pharmacokinetic parameters screened based on a one-factor nonlinear mixed effect model are shown in table 10. The results show that male, glutamate oxaloacetic transaminase and combined sodium valproate, oxcarbazepine and lithium carbonate administration is associated with higher rates of olanzapine elimination in vivo (P < 0.05); while glomerular filtration hypofunction is associated with a reduced rate of elimination in olanzapine (P < 0.05). SNP loci such as rs4244285 (CYP 2C 19), rs12768009 (CYP 2C 19), rs7916649 (CYP 2C 19), rs3758580 (CYP 2C 19), rs3743484 (CYP 1A 2), rs11528090 (CYP 2C 19) and the like have statistical significance (P < 0.05) on the primary elimination rate.

Table 10 genetic and non-genetic factors related to the pharmacokinetics parameters of olanzapine population

^a CL: a first order cancellation rate constant (L/H); ^b glutamic-oxaloacetic transaminase is a continuous variable; gender, glomerular filtration hypofunction, sodium valproate, oxcarbazepine and lithium carbonate are classified as two classification variables (0/1); rs4244285:0-A/A,1-A/G,2-G/G; rs12768009:0-A/A,1-G/A,2-G/G; rs7916649:0-A/A,1-G/A,2-G/G; rs3743484:0-C/C,1-G/C,2-G/G; rs11528090:0-G/G,1-T/G,2-T/T;

covariates of the group pharmacokinetic model are screened by adopting forward increasing and backward eliminating methods, and the final group pharmacokinetic model is determined by combining relevant expertise, and the parameter estimation of the olanzapine final group pharmacokinetic model is shown in table 11. In the final population pharmacokinetics model, the primary elimination rate fixed effect model comprises glutamic oxaloacetic transaminase, glomerular filtration hypofunction, sodium valproate, oxcarbazepine, rs7916649 and rs12768009. Typical values of the population of the final models Ka, V, and CL are respectively: ka=0.144 h ^-1 V=71.197l, cl=6.120 x p (0.009 x oryzanol conversion)Ammoniase-0.130 glomerular filtration hypofunction +0.253 sodium valproate +0.774 oxcarbazepine +0.292 rs7916649-0.155 rs12768009). The verification result of Bootstrap 500 times shows that model parameter estimation based on measured data is basically similar to parameter estimation value based on sampling data, and the estimation value of main PK parameters obtained by a final model is in a quartile range of Bootstrap verification parameters, so that the model estimation is stable and has good reliability.

TABLE 11 results of final population pharmacokinetic model parameter estimation and validation of olanzapine

^a CL: olanzapine clearance rate (L/H); ka: absorption Rate (h) ^-1 ) The method comprises the steps of carrying out a first treatment on the surface of the V: apparent distribution volume (V);

^b rs7916649：0-A/A，1-G/A，2-G/G。

the goodness-of-fit plot of olanzapine final population pharmacokinetic model is shown in figure 8. The results show that the individual concentration predicted value, the population concentration predicted value and the measured concentration value have good fitting degree, and the RMSE and the R ² 13.946ng/ml, 0.469 and 16.9ng/ml, 0.212, respectively. The plasma concentration observations are uniformly distributed on both sides of the trend line (y=x) and substantially coincide with the diagonal line. The residual map of olanzapine final population pharmacokinetic model is shown in fig. 9, and the result shows that the conditional weighted residuals are uniformly distributed on both sides of y=0, and have no obvious correlation with time and population prediction values. The final model population pharmacokinetic profile for clozapine with a drug regimen of 75mg (bid) is shown in FIG. 10.

2.2.3 olanzapine conventional monitoring plasma concentration prediction

2.2.3.1 variable screening based on recursive feature elimination algorithm

We use a random forest based recursive feature elimination algorithm to screen the prediction variables for machine learning modeling. Ten fold cross validation RMSE based on random forest modeling with different numbers of candidate variables is shown in fig. 11. From fig. 11, it can be seen that the RMSE of the random forest algorithm cross-fold verification is minimum when the predicted variables of the first 42 bits of the variable importance ranking are included, and the random forest variable importance is shown in fig. 12. The predicted variables screened by the random forest-based recursive feature elimination algorithm include dosage, frequency of administration, age, alkaline phosphatase, weight, glutamic-pyruvic transaminase, glutamic-oxaloacetic transaminase, sodium valproate, CYP2D6 enzyme activity score, rs1065852, rs762551, rs3743484, aripiprazole, rs11528090, rs35280822, rs16947, rs79331140, rs1080996, rs7787082, propranolol, rs7779562, rs75276289, rs4646437, rs6583954, rs3758580, rs4244285, smoking, drinking, sex, rs12768009, rs7916649, rs2470890, rs17884832, rs5030865, 17879992, rs17885098, rs28371725, glomerular filtration hypofunction, phenolphthalein tablet, sulpiride, ziprasidone and radix Sanguisorbae white tablet. The prediction variables are used for constructing a prediction model based on a machine learning method such as random forest, bayesian additive regression tree and the like.

2.2.3.2 model fitting and evaluation

The method comprises the steps of fitting a prediction model based on a gradient lifting tree, a support vector machine, a random forest, a Bayes cumulative regression tree and an xgboost algorithm, simultaneously determining a group of weight vectors which minimize the risk of the combined cross validation of the model by using NNLS as a loss function and using ten-fold cross validation, wherein the weights of the model are respectively 0.424, 0.454, 0.000, 0.028 and 0.094, so that a model for predicting the blood concentration of the clozapine based on Super Learner is constructed, and the model is suitable for predicting the condition of the schizophrenia patient in the morning 6: olanzapine blood concentration at 30 hours (administration time is 10 am and/or 21 pm, respectively). The individual algorithm weights and parameter settings in the Super filter are shown in table 12.

Table 12 weights and parameter settings in Super Learner for individual algorithms

Note that: each algorithm parameter not listed in table 12 is set to the R-packet default value.

In the study, 100 times of routine blood concentration monitoring records are randomly extracted as a verification set, the rest is used as a training set, and the evaluation of olanzapine blood concentration prediction model based on Super Learner is shown in table 13. In the training set, the RMSE of the Super Learner, gradient lifting tree, support vector machine, random forest, bayesian additive regression tree, and xgBoost algorithm were 9.18ng/ml, 9.36ng/ml, 10.12ng/ml, 9.03ng/ml, 11.51ng/ml, and 7.31ng/ml, respectively. In the validation set, the RMSE for the above algorithm was 13.90ng/ml, 13.74ng/ml, 14.96ng/ml, 13.74ng/ml, 15.07ng/ml and 14.17ng/ml, respectively. R of Super Learner in training set and verification set ² 0.75 and 0.53, respectively, perform well (fig. 13).

TABLE 13 olanzapine plasma concentration prediction model evaluation based on SuperLearner

2.3 basic condition of clozapine study object and group pharmacokinetics study

2.3.1 basic conditions of study object

This part of the study included 136 schizophrenic patients receiving clozapine as subjects, and the general condition of the subjects and the distribution of laboratory test indicators are shown in Table 14. Subjects had an average age of 44.8±9.9 years; 70 men and 63 women, respectively accounting for 51.5% and 48.5%; the average value of the height and the weight is 163.7 +/-8.8 cm and 70.6+/-14.5 kg, and the average value of the BMI is 26.3+/-4.7 kg/m ² The method comprises the steps of carrying out a first treatment on the surface of the 35 smokers and 19 drinkers in the study subjects, accounting for 25.7% and 14.0% respectively; the prevalence of diabetes and hypertension is 11.0% and 8.1%, respectively. The mean values of the liver function and the kidney function detection indexes are in the normal reference value range, but 27 people have glomerular filtration hypofunction, and 30 people have abnormal liver function, and the mean values respectively account for 19.9 percent and 22.1 percent.

The study included patients with schizophrenia who took clozapine twice daily or once daily at a mean dose of 75.9.+ -. 35.5mg per administration. In the blood concentration steady state, the blood concentration of the clozapine of 665 persons is collected in the study, and after taking medicine for 1.5 hours, 5.0 hours, 9.5 hours and 12.5 hours, the mean value of the blood concentration of the study subjects is 588.2 +/-235.5, 460.2 +/-227.5, 337.4 +/-233.8 and 249.1+/-198.9 ng/ml respectively. According to the upper limit of the reference range of the treatment concentration of the clozapine of 600ng/ml, the blood concentration after taking the medicine for 1.5 hours for 7 times and the blood concentration after taking the medicine for 9.5 hours for 66 times respectively exceed the upper limit of the reference range of the treatment concentration, and the blood concentration respectively accounts for 53.8 percent and 10.69 percent of the detection examples of the blood concentration at the time point.

TABLE 14 general condition of study subjects and laboratory test indicators

2.3.2 clozapine population pharmacokinetics studies

In order to determine the mean value, standard deviation and discrete degree of various pharmacokinetic parameters of the schizophrenic patient population and to study the influence of various factors on clozapine metabolism, the study carries out the quetiapine population pharmacokinetics analysis. And combining literature reports and the characteristics of the research data, finally determining a basic structure model as a pharmacokinetic structure model for primary absorption and elimination of a room, a random effect model as an addition model, and quantitatively examining the influence of a fixed effect (such as age, sex, SNP locus and the like) on pharmacokinetic parameters by adopting a linear model. Genetic and non-genetic factors related to population pharmacokinetic parameters screened based on a one-factor nonlinear mixed effect model are shown in table 15. The results show that the apparent distribution volume of men is significantly greater than that of women (P > 0.05); whereas with age the apparent distribution volume gradually decreases (P < 0.05). Genetic factors of statistical significance (P < 0.05) for the primary elimination rate effect include: rs1135840 (CYP 2D 6), rs1065164 (CYP 2D 6), rs28371699 (CYP 2D 6), and rs1135822 (CYP 2D 6); non-genetic factors include: age, sex, smoking, glutamic-oxaloacetic transaminase level, and co-administration of propranolol. In addition, the CYP2D6 enzyme activity score is positively correlated with the primary elimination rate, which increases with increasing CYP2D6 enzyme activity score (P < 0.05).

TABLE 15 genetic and non-genetic factors related to the pharmacokinetics parameters of the clozapine population

V: apparent distribution volume (L); CL: a first order cancellation rate constant (L/H);

covariates of the group pharmacokinetic model are screened by adopting forward increasing and backward eliminating methods, and the final group pharmacokinetic model is determined by combining relevant expertise, and the parameter estimation of the final group pharmacokinetics model of clozapine is shown in Table 16. In the final population pharmacokinetic model, gender had a significant effect on apparent distribution volume (P<0.05 A) is provided; age, sex, combined propranolol and SNP sites rs1135840 and rs1135822 have significant influence on the primary elimination rate (P)<0.05). The parameter estimates of the final model are shown in table 18. The results show that population typical values of the final models Ka, V and CL are respectively: ka=0.558 h ^-1 V=181.874 x p (1.371 x male), cl=42.472 x p (0.239 x male-0.009 x age-0.137 x propranolol-0.148 x rst1135840-0.474 x rst1135822). The verification result of Bootstrap 500 times shows that model parameter estimation based on measured data is basically similar to parameter estimation value based on sampling data, and the estimation value of main PK parameters obtained by a final model is in a quartile range of Bootstrap verification parameters, so that the model estimation is stable and has good reliability.

TABLE 16 final model parameters estimation and validation results for clozapine

^a CL: clozapine clearance rate (L/H); ka: absorption Rate (h) ^-1 ) The method comprises the steps of carrying out a first treatment on the surface of the V: apparent distribution volume (V);

^b rs1135840：0-C/C，1-C/G，2-G/G；rs1135822：0-A/T，1-T/T；

the goodness-of-fit plot of the final model is shown in fig. 14. The results show that the individual concentration predicted value, the population concentration predicted value and the measured concentration value have better fitting degree, and the RMSE and the R ² 98.33ng/ml, 0.857 and 173.45ng/ml, 0.56, respectively. The plasma concentration observations are uniformly distributed on both sides of the trend line (y=x) and substantially coincide with the diagonal line. The residual diagram of the final model is shown in fig. 15. The result shows that the conditional weighted residuals are uniformly distributed on two sides of y=0, and have no obvious correlation with time and group prediction values, which indicates that the model fitting effect is better. The final model population pharmacokinetic profile for clozapine with a drug regimen of 75mg (bid) is shown in FIG. 16.

2.3.3 conventional monitoring of clozapine plasma concentration prediction

2.3.3.1 variable screening based on recursive feature elimination algorithm

We use a random forest based recursive feature elimination algorithm to screen the prediction variables for machine learning modeling. Ten fold cross validation RMSE based on random forest modeling with different numbers of candidate variables is shown in figure 17. As can be seen from fig. 17, when the predicted variable of the first 28 bits of the variable importance ranking is included, RMSE of ten-fold cross-validation of the random forest algorithm is minimum, and then the predicted variables screened based on the recursive feature elimination algorithm of the random forest include dosage, creatinine, age, glutamic-oxaloacetic transaminase, creatinine clearance, body weight, propranolol, risperidone, rs1080996, rs75276289, rs16947, rs11528090, rs762551, sex, frequency of administration, rs12535512, back-racetam capsule, rs1135840, rs1128503, rs12720464, rs3789243, CYP2D6 enzyme activity score, aspirin enteric tablet, rs3842, rs6979885, atorvastatin tablet, rs4728709, and smoking. The prediction variables are used for constructing a prediction model based on a machine learning method such as random forest, bayesian additive regression tree and the like.

2.3.3.2 model fitting and evaluation

The method comprises the steps of fitting a prediction model based on a gradient lifting tree, a support vector machine, a random forest, a Bayes cumulative regression tree and an xgboost algorithm, simultaneously determining a group of weight vectors which minimize the risk of the combined cross validation of the model by using ten-fold cross validation by taking NNLS as a loss function, wherein the weights of the model are respectively 0.271, 0.342, 0.213, 0.155 and 0.019, so that a model for predicting the blood concentration of the clozapine based on a Super Learner is constructed, and the model is suitable for predicting the condition that a patient suffering from schizophrenia is in a steady state in the morning of 6: the blood concentration of clozapine at 30 hours (the administration time is respectively 10 am and/or 21 pm). The individual algorithm weights and parameter settings in the Super filter are shown in table 17.

Table 17 weights and parameter settings in Super Learner for individual algorithms

Note that: each algorithm parameter not listed in table 19 is set to the R-packet default value.

In the study, 150 times of routine blood concentration monitoring records are randomly extracted as a verification set, the rest is used as a training set, and the evaluation of olanzapine blood concentration prediction model based on Super Learner is shown in table 18. In the training set, the RMSE of Super Learner, gradient lifting tree, support vector machine, random forest, bayesian additive regression tree, and xgboost algorithm were 68.88ng/ml, 69.26ng/ml, 72.76ng/ml, 64.41ng/ml, 80.73ng/ml, and 55.89ng/ml, respectively. In the validation set, the RMSE for the above algorithm was 88.51ng/ml, 92.08ng/ml, 90.45ng/ml, 90.68ng/ml, 88.98ng/ml and 96.31ng/ml, respectively. R of Super Learner in training set and verification set ² 0.72 and 0.58, respectively, perform well (fig. 19).

Table 18 olanzapine plasma concentration prediction model evaluation based on Super Learner

2.4 study of quetiapine subject base Condition and group pharmacokinetics

2.4.1 basic conditions of the study object

This part of the study included 177 patients with schizophrenia treated with quetiapine as subjects, and the general condition of the subjects and the distribution of laboratory test indicators are shown in Table 19. Subjects had an average age of 46.3±12.6 years; 104 men and 73 women, accounting for 58.8% and 41.2% respectively; the average value of the height and the weight is 163.3+/-8.6 cm and 71.5+/-16.5 kg respectively, and the average value of the BMI is 26.7+/-5.5 kg/m2; in the study subjects, the smokers 44 and the drinkers 35 respectively account for 24.9% and 19.8%; the prevalence of diabetes and hypertension is 11.3% and 13.6%, respectively. The mean values of the liver function and the kidney function detection indexes are in the normal reference value range, but 38 people have glomerular filtration hypofunction, and 21 people have abnormal liver function, and the mean values respectively account for 21.5 percent and 11.9 percent.

The study only included patients with schizophrenia who took quetiapine twice daily, with a mean dose of 242.9±88.3mg each. In the steady state of blood concentration, 551 blood levels of quetiapine are collected in the study, and after 1.5h, 5.0h, 9.5h and 12.5h of administration, the mean blood concentration of the study subjects is 334.5 +/-218.2, 286.3 +/-141.2, 209.1+/-144.5 and 141.9+/-125.5 ng/ml respectively. According to the upper limit of the reference range of the quetiapine treatment concentration of 500ng/ml, the blood concentration after 9 times of taking medicine for 1.5 hours and the blood concentration after 18 times of taking medicine for 9.5 hours exceed the upper limit of the reference range of the treatment concentration, and the blood concentration respectively accounts for 25.7% and 3.9% of the times of the blood concentration detection at the time point.

TABLE 19 general condition of study subjects and laboratory test indicators

2.4.2 quetiapine population pharmacokinetics studies

In order to determine the mean value, standard deviation and discrete degree of various pharmacokinetic parameters of the schizophrenic patient population and to study the influence of various factors on clozapine metabolism, the study carries out the quetiapine population pharmacokinetics analysis. And combining literature reports and the characteristics of the research data, finally determining a basic structure model as a pharmacokinetic structure model for primary absorption and elimination of a room, a random effect model as an addition model, and quantitatively examining the influence of a fixed effect (such as age, sex, SNP locus and the like) on pharmacokinetic parameters by adopting a linear model. Genetic and non-genetic factors related to population pharmacokinetic parameters screened based on a one-factor nonlinear mixed effect model are shown in table 20. The results showed that as the body weight increases, the apparent distribution volume gradually increases, and that the effect of body weight on the apparent distribution volume is all statistically significant (P < 0.05). Genetic factors of statistical significance (P < 0.05) for the primary elimination rate effect include: rs5030865 (CYP 2D 6), rs2069526 (CYP 1 A2), rs2242480 (CYP 3 A4), rs2235047 (ABCB 1); non-genetic factors include: age, sex, endogenous creatinine clearance, alkaline phosphatase, glutamic-oxaloacetic transaminase, back-racetam capsule, aripiprazole, atorvastatin calcium tablet, and blood creatinine level, etc.

TABLE 20 genetic and non-genetic factors related to the pharmacokinetics parameters of quetiapine populations

^a Ka: first order absorption Rate constant (h ^-1 ) The method comprises the steps of carrying out a first treatment on the surface of the V: apparent distribution volume (L); CL: first order cancellation rate constant (L/H)；

Covariates of the group pharmacokinetic model are screened by adopting forward increasing and backward eliminating methods, and the final group pharmacokinetic model is determined by combining relevant expertise, and the parameter estimation of the quetiapine final group pharmacokinetic model is shown in table 21. In the final population pharmacokinetic model, age, sex, combination of atorvastatin calcium tablet, endogenous creatinine clearance and rs2242480 had a significant effect on primary elimination rate (P<0.05). Typical values of the population of the final models Ka, V, and CL are respectively: ka=0.163 h ^-1 V=274.626l, cl=66.389 x p (-0.195 x female-0.007 x age +0.004 x intra-myogenic anhydride clearance-0.599 x atorvastatin calcium-0.119 x rs2242480). The verification result of Bootstrap 500 times shows that model parameter estimation based on measured data is basically similar to parameter estimation value based on sampling data, and the estimation value of main PK parameters obtained by a final model is in a quartile range of Bootstrap verification parameters, so that the model estimation is stable and has good reliability.

Table 21 quetiapine final population pharmacokinetic model parameter estimation

^a CL: quetiapine clearance rate (L/H); ka: absorption Rate (h) ^-1 ) The method comprises the steps of carrying out a first treatment on the surface of the V: apparent distribution volume (V);

^b rs2242480：0，C/C；1，T/C；2，T/T；。

the goodness-of-fit plot of the final model is shown in figure 20. The results show that the individual concentration predicted value and the population concentration predicted value have better fitting degree with the measured concentration value, and the RMSE and the R2 are 63.11ng/ml, 0.857 and 136.13ng/ml and 0.316 respectively. The plasma concentration observations are uniformly distributed on both sides of the trend line (y=x) and substantially coincide with the diagonal line. The residual diagram of the final model is shown in fig. 21. The result shows that the conditional weighted residuals are uniformly distributed on two sides of y=0, and have no obvious correlation with time and group prediction values, which indicates that the model fitting effect is better. The final model population pharmacokinetic profile for quetiapine dosing regimen at 300mg (bid) is shown in figure 22.

2.4.3 prediction of the blood concentration of quetiapine for conventional monitoring

2.4.3.1 variable screening based on recursive feature elimination algorithm

We use a random forest based recursive feature elimination algorithm to screen the prediction variables for machine learning modeling. Ten fold cross validation RMSE based on random forest modeling with different numbers of candidate variables is shown in figure 23. From the figure, when the predicted variable with the top 31 bits of the variable importance ranking is included, the RMSE of ten-fold cross validation of the random forest algorithm is minimum, and the predicted variable importance score based on the random forest is shown in the figure 24. The predicted variables screened by the random forest-based recursive feature elimination algorithm include dosage, age, BMI, alkaline phosphatase, glutamic-oxaloacetic transaminase, creatinine, glutamic-pyruvic transaminase, body weight, sodium valproate, CYP2D6 enzyme activity score, rs12535512, rs2246709, rs12768009, rs2242480, rs1081003, cefuroxime axetil, amisulpride, rs58440431, rs2004511, diexin soft capsules, gender, enalapril maleate tablets, atorvastatin calcium tablets, rs2470890, smoking, liver function abnormality, rs1135822, oxcarbazepine, buspirone, aripiprazole and heart stabilizing particles. The prediction variables are used for constructing a prediction model based on a machine learning method such as random forest, bayesian additive regression tree and the like.

2.4.3.2 model fitting and evaluation

The study fits a prediction model based on a gradient lifting tree, a support vector machine, a random forest, a Bayesian additive regression tree and an xgboost algorithm, and simultaneously uses NNLS as a loss function, a set of weight vectors which minimize the risk of cross validation of the combination of the models are determined by using ten-fold cross validation, and the weights of the models are respectively 0.000, 0.083, 0.674, 0.243 and 0.000, so that a 9-hydroxy risperidone blood concentration prediction model based on Super Learner is constructed, and the model is suitable for predicting the schizophrenia patient in a steady state in the morning of 6: the blood concentration of 9-hydroxy risperidone at 30 hours (the administration time is respectively 10 am and 21 pm). The individual algorithm weights and parameter settings in the Super filter are shown in table 22.

Table 22 weight and parameter settings in Super Learner for individual algorithms

Note that: each algorithm parameter not listed in table 22 is set to the R-packet default value.

In the study, 100 times of routine blood concentration monitoring records are randomly extracted as a verification set, the rest is used as a training set, and the evaluation of the quetiapine blood concentration prediction model based on the Super Learner is shown in Table 23. In the training set, the RMSE of Super Learner, gradient lifting tree, support vector machine, random forest, bayesian additive regression tree, and xgboost algorithm were 68.43ng/ml, 67.61ng/ml, 75.28ng/ml, 63.27ng/ml, 88.60ng/ml, and 48.66ng/ml, respectively. In the validation set, the RMSE for the above algorithm was 85.27ng/ml, 81.09ng/ml, 84.18ng/ml, 86.90ng/ml, 89.35ng/ml and 90.23ng/ml, respectively. R of Super Learner in training set and verification set ² 0.65 and 0.46, respectively, perform well (fig. 25).

Table 23 model evaluation of the quetiapine plasma concentration prediction model of Super Learner

It should be noted that the above examples are only for illustrating the technical solution of the present invention and are not limiting thereof. Although the present invention has been described in detail with reference to the examples given, those skilled in the art can make modifications and equivalents to the technical solutions of the present invention as required, without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a pharmaceutical model of an antipsychotic drug population, the method comprising: collecting clinical information and blood samples of a subject, and obtaining corresponding data information; adopting a nonlinear mixed effect model for analysis, specifically adopting a pharmacokinetic model of primary absorption and elimination of a chamber as a basic structure model, and obtaining blood concentration simulation data to obtain corresponding model pharmacokinetic parameters; the random effect model is an addition model, covariates are introduced, model fitting is carried out, the influence of different factors on the pharmacokinetic parameters is investigated, and an antipsychotic drug population pharmacokinetics model is constructed;

wherein the antipsychotic agent comprises risperidone, olanzapine, clozapine and quetiapine;

The covariates include genetic and non-genetic factors, and are quantitatively screened and evaluated using a linear model.

2. The method of claim 1, wherein when the drug is risperidone, the covariates comprise body weight, smoking, drinking habits, combination of the even flower plague-clearing granules and silybin meglumine tablets, glomerular filtration hypofunction, and CYP2D6 enzyme activity scores;

typical values of the population of the final models Ka, V, and CL are respectively: ka= 6.113h ^-1 V=25.694 x p (0.022 x body weight), cl=3.139 x p (0.155 x smoking-0.1158 x drinking-0.280 x glibenchmarking granules+0.473 x silybin meglumine tablets-0.141 x glomerular filtration hypofunction+0.066 x cyp2d6 enzyme activity score);

when the medicine is olanzapine, the covariates comprise glutamic-oxaloacetic transaminase, glomerular filtration hypofunction, sodium valproate, oxcarbazepine and SNP loci rs7916649 and rs12768009;

typical values of the population of the final models Ka, V, and CL are respectively: ka=0.144 h ^-1 V=71.197l, cl=6.120 exp (0.009 glutamic oxaloacetic transaminase-0.130 glomerular filtration hypofunction +0.253 sodium valproate +0.774 oxcarbazepine +0.292 rs7916649-0.155 rs12768009);

when the drug is clozapine, the covariates comprise age, sex, combined propranolol and SNP loci rs1135840 and rs1135822;

Typical values of Ka, V and CL in the final model are: ka=0.558 h ^-1 V=181.874 x p (1.371 x male), cl=42.472 x p (0.239 x male-0.009 x age-0.137 x propranolol-0.148 x rst1135840-0.474 x rst1135822);

when the drug is quetiapine, the covariates include age, sex, atorvastatin calcium combination, endogenous creatinine clearance and rs2242480;

3. Use of an antipsychotic drug population pharmacokinetic model obtained by the construction method of claim 1 or 2 in any one or more of the following:

(b) Preparing a personalized product of the antipsychotic;

(c) Is used for relevant basic researches such as pharmacokinetics of antipsychotics and the like;

further, in said (c), said antipsychotic pharmacokinetics comprises an antipsychotic plasma concentration prediction.

4. An antipsychotic drug blood concentration predictive assessment system, the system comprising at least:

An acquisition unit configured to: acquiring subject related data information;

a data processing unit configured to: predicting the blood concentration of the antipsychotic drug of the subject based on the data information obtained by the obtaining unit and based on the built-in predictive evaluation model; the prediction evaluation model is obtained by training the pre-collected relevant data information of the patient by adopting a statistical algorithm;

an output unit configured to: outputting a predicted blood concentration value of the antipsychotic of the subject according to the information of the data processing unit;

wherein the subject-related data includes SNP loci related to general conditions, disease characteristics, antipsychotic medication conditions and antipsychotic therapeutic effects of the subject;

the antipsychotic agent comprises risperidone, clozapine, olanzapine and quetiapine.

5. The system of claim 4, wherein, when the antipsychotic is risperidone, the predictive evaluation model construction process, the relevant data information includes, but is not limited to, dosage, body weight, BMI, age, endogenous creatinine clearance, rs17327442, CYP2D6 enzyme activity score, sodium valproate, propranolol, rs7787082, benalasol, rs3789243, rs4244285, rs11528090, rs7779562, dioxepinacol soft capsule, rs762551, rs3743484, rs2235047, smoking, rs1135840, rs28371699, rs1058164, sex, ziprasidan capsule, drinking wine, rs1081003, rs1065852, rs58440431, rs2004511, rs16947, rs75276289, rs1080996, silybin meglumine tablet, rs116917064, aspirin enteric tablet;

6. The system of claim 4, wherein the antipsychotic drug is olanzapine, and wherein the relevant data information during the predictive assessment model construction process comprises dosing, dosing frequency, age, alkaline phosphatase, body weight, glutamic-pyruvic transaminase, glutamic-oxaloacetic transaminase, sodium valproate, CYP2D6 enzyme activity score, rs1065852, rs762551, rs3743484, aripiprazole, rs11528090, rs35280822, rs16947, rs79331140, rs1080996, rs7787082, propranolol, rs7779562, rs75276289, rs4646437, rs6583954, rs3758580, rs4244285, smoking, drinking, gender, rs12768009, rs7916649, rs2470890, rs17884832, rs5030865, rs17879992, rs17885098, 28371725, glomerular hypofunction, phenolphthalein, sulpiride, ziprasidone, and raw white tablet;

7. The system of claim 4, wherein the antipsychotic drug is clozapine and the predictive assessment model is constructed by the relevant data information including dosage, creatinine, age, glutamate oxaloacetate, creatinine clearance, body weight, propranolol, risperidone, rs1080996, rs75276289, rs16947, rs11528090, rs762551, gender, frequency of use, rs12535512, ziracetam capsule, rs1135840, rs1128503, rs12720464, rs3789243, CYP2D6 enzyme activity score, aspirin enteric tablet, rs3842, rs6979885, atorvastatin calcium tablet, rs4728709, smoking;

8. The system of claim 4, wherein the antipsychotic agent is quetiapine and the relevant data information during the predictive assessment model construction includes dosage, age, BMI, alkaline phosphatase, glutamic-oxaloacetate, creatinine, glutamic-pyruvic transaminase, body weight, sodium valproate, CYP2D6 enzymatic activity score, rs12535512, rs2246709, rs12768009, rs2242480, rs1081003, cefuroxime axetil, amisulpride, rs58440431, rs2004511, geotricin soft capsule, gender, enalapril maleate tablet, atorvastatin calcium tablet, rs2470890, smoking, liver function abnormality, rs1135822, oxcarbazepine, buspirone, aripiprazole, and heart stabilizing particles;

9. A computer readable storage medium, on which a program is stored, which program, when being executed by a processor, implements the functions of the system according to any of claims 4-8.

10. An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor implements the functions of the system of any of claims 4-8 when the program is executed by the processor.