Formula determination method of long-acting injection, electronic equipment and storage medium

Abstract

The invention provides a formula determination method of a long-acting injection, electronic equipment and a storage medium, and relates to the technical field of pharmaceutical preparations. The method comprises the following steps: constructing a physiological pharmacokinetic model of a reference pharmaceutical preparation, and constructing a pharmacodynamic model of the reference pharmaceutical preparation; reversely deducing an in-vitro target release curve of the reference pharmaceutical preparation according to the physiological pharmacokinetic model and the pharmacodynamic model; according to a property prediction model obtained through pre-training, obtaining an in-vitro predicted release curve of the at least one alternative preparation formula; and screening out a target preparation formula from at least one alternative preparation formula according to the in-vitro predicted release curve and the in-vitro target release curve of each alternative preparation formula. According to the application, the physiological pharmacokinetic model, the pharmacodynamic model and the property prediction model are integrated, the design and screening of a new preparation formula are guided, an intelligent and closed-loop long-acting injection development process is realized, the development period is greatly shortened, the success rate is improved, and the problem of long development period of the long-acting injection is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pharmaceutical preparations, in particular to a formula determination method of long-acting injectables. BACKGROUND

[0002] Long-acting injectables (LAIs) are an important means of drug delivery for chronic diseases, with advantages such as prolonged efficacy and reduced dosing frequency, and are widely used in the fields of arthritis, diabetes, cancer and mental illness. However, the development of traditional LAI formulations highly depends on trial and error experiments, with a cycle of 8-10 years, and lacks clear in vitro-in vivo correlation (IVIVC), resulting in low development efficiency.

[0003] Therefore, how to improve the development efficiency of long-acting injectables is a problem to be solved. SUMMARY

[0004] The present application aims at the deficiencies in the prior art, and provides a formula determination method of long-acting injectables, an electronic device and a storage medium, so as to solve the technical problems in the prior art.

[0005] To achieve the above-mentioned purpose, the technical solutions adopted by the embodiments of the present application are as follows:

[0006] In a first aspect, the embodiments of the present application provide a formula determination method of long-acting injectables, which comprises:

[0007] constructing a physiological pharmacokinetic model of a reference drug formulation, and constructing a pharmacodynamics model of the reference drug formulation, wherein the physiological pharmacokinetic model is used to characterize the dynamic change process of drug release, absorption, distribution, metabolism and excretion in the body, and the pharmacodynamics model is used to characterize the relationship between different drug concentrations and drug effects;

[0008] According to the physiological pharmacokinetic model and the pharmacodynamics model, the in vitro target release curve of the reference drug formulation is back calculated;

[0009] According to a pre-trained property prediction model, an in vitro predicted release curve of at least one alternative formulation is obtained;

[0010] According to the in vitro predicted release curve of each alternative formulation and the in vitro target release curve, a target formulation is screened from the at least one alternative formulation.

[0011] Optionally, the constructing of the physiological pharmacokinetic model of the reference drug formulation comprises:

[0012] obtaining drug parameters of the reference drug formulation and physiological parameters of a test subject;

[0013] establishing a base model of at least one compartment structure, the base model comprising: a drug release model, a disposition model of a drug in each compartment structure;

[0014] inputting a drug parameter of the reference drug preparation, a physiological parameter of a test subject into the base model of each compartment structure, and establishing a physiologically-based pharmacokinetic model of the reference drug preparation.

[0015] Optionally, the constructing the pharmacodynamics model of the reference drug preparation comprises:

[0016] obtaining a pre-constructed placebo effect model, and determining a change relationship between time and pharmacodynamic effect under the placebo effect model, wherein the placebo effect model is used to characterize a non-drug induced pain relief effect in a simulation experiment;

[0017] obtaining a pre-constructed drug effect model, and determining a change relationship between drug exposure and pharmacodynamic effect of the reference drug preparation under the drug effect model;

[0018] obtaining a pharmacodynamics model of the reference drug preparation according to the change relationship between time and pharmacodynamic effect under the placebo effect model and the change relationship between drug exposure and pharmacodynamic effect of the reference drug preparation under the drug effect model.

[0019] Optionally, the backstepping the in-vitro target release curve of the reference drug preparation according to the physiologically-based pharmacokinetic model and the pharmacodynamics model comprises:

[0020] obtaining pre-set target pharmacodynamic response information;

[0021] simulating blood concentration change and pharmacodynamic response information under at least one release curve according to the physiologically-based pharmacokinetic model and the pharmacodynamics model;

[0022] identifying a release curve window satisfying the target pharmacodynamic response information from the at least one release curve;

[0023] obtaining pre-defined key release parameters of the in-vitro target release curve;

[0024] obtaining the in-vitro target release curve of the reference drug preparation according to the key release parameters of the in-vitro target release curve and the release curve window.

[0025] Optionally, the obtaining the in-vitro predicted release curve of at least one alternative preparation formula according to the pre-trained property prediction model comprises:

[0026] obtaining formula parameters of each of the alternative preparation formulas;

[0027] inputting the formulation parameters of each of the alternative formulation recipes into the property prediction model to obtain an in-vitro predicted release curve of each of the alternative formulation recipes.

[0028] Optionally, the screening of the target formulation recipe from the at least one alternative formulation recipe according to the in-vitro predicted release curve of each of the alternative formulation recipes and the in-vitro target release curve comprises:

[0029] determining a similarity factor between the in-vitro predicted release curve of each of the alternative formulation recipes and the in-vitro target release curve using a preset closed-loop optimization strategy, and taking the alternative formulation recipe corresponding to the maximum similarity factor as the target formulation recipe.

[0030] Optionally, the training process of the property prediction model comprises:

[0031] obtaining a training data set, the training data set comprising a plurality of historical formulation recipes and release curve data corresponding to each historical formulation recipe, and the training data set further comprising a plurality of candidate formulation recipes and actual in-vitro release curves of each candidate formulation recipe;

[0032] preprocessing the training data set to obtain a preprocessed training data set;

[0033] iteratively training the initial property prediction model using the preprocessed training data set until a new initial property prediction model obtained by current training meets a preset condition, and then taking the new initial property prediction model as the property prediction model.

[0034] Optionally, the method further comprises:

[0035] adopting a preset molecular dynamics model to simulate and verify the interaction mechanism between the drug and the polymer in the target formulation recipe to obtain a verification result of the target formulation recipe.

[0036] In a second aspect, the embodiments of the present application further provide an electronic device, comprising a processor, a storage medium and a bus, the storage medium storing machine readable instructions executable by the processor, the processor and the storage medium being in communication through the bus when the electronic device is running, and the processor executing the machine readable instructions to perform the steps of the method provided in the first aspect.

[0037] In a third aspect, the embodiments of the present application further provide a computer readable storage medium, the storage medium storing a computer program, the computer program being executed by a processor to perform the steps of the method provided in the first aspect.

[0038] The present application has the following beneficial effects:

[0039] The application provides a long-acting injection formula determination method, an electronic device and a storage medium. The method comprises: constructing a PBPK model of a reference drug preparation and a PD model of the reference drug preparation, i.e., constructing a PBPK model and a PD model of a known drug preparation serving as a reference standard in the development process of a long-acting injection, to simulate the release behavior of the drug in a test subject and the change of drug efficacy, and finally to inversely deduce an in-vitro release target to guide the intelligent development of the long-acting injection; then, starting from the pharmacokinetics (PK) and pharmacodynamics (PD) behavior of the drug in the body, the established PBPK model and PD model are used to reversely deduce the in-vitro release curve characteristics required to achieve the in-vivo behavior, thereby guiding the design and screening of new preparation formulas and realizing an intelligent and closed-loop long-acting injection development process; according to a pre-trained property prediction model, an in-vitro prediction release curve of at least one candidate preparation formula is obtained, i.e., the property prediction model can quickly predict the drug release curve of each candidate preparation formula, so as to be used for guiding the intelligent design and screening of the long-acting injection, greatly shortening the development cycle and improving the success rate; according to the in-vitro prediction release curve and the in-vitro target release curve of each candidate preparation formula, a target preparation formula is screened from at least one candidate preparation formula, thereby realizing efficient and accurate long-acting injection development and shortening the development cycle of the long-acting injection. Therefore, in order to solve the problems of long development cycle, poor predictability and insufficient in-vitro and in-vivo correlation of the long-acting injection, the application proposes an integrated PBPK model and PD model and machine learning model to construct a calculation-driven closed-loop development framework, which can quickly predict and optimize the in-vitro release behavior and in-vivo efficacy of the long-acting injection, i.e., taking calculation driving as the core, combining formula prediction driven by data and release mechanism interpretation driven by mechanism, thereby accelerating the formula design and optimization of the in-situ gel long-acting injection, and the triamcinolone acetonide in-situ gel preparation developed based on the method shows better drug exposure level than the commercial suspension in the rat pharmacokinetic study, thereby realizing efficient and accurate long-acting injection development and shortening the development cycle of the long-acting injection; and the method significantly shortens the formula development cycle from the traditional 8-10 years to 1-2 years, improves the formula screening efficiency, and is suitable for intelligent research and development of various long-acting injection forms.

[0040] Secondly, the application can greatly improve the in-situ gel preparation development efficiency: by integrating PBPK modeling, machine learning prediction and molecular dynamics mechanism analysis, the period from demand definition to formula determination is significantly shortened, saving more than 80% time and resources compared with the traditional trial-and-error method; and the in-vivo and in-vitro release behaviors are accurately linked, the in-vitro release target is reversely deduced by using the PBPK / PD model for the first time, ensuring that the screened formula can achieve the expected efficacy (such as analgesic duration) in the body, and the IVIVC correlation is strengthened.

[0041] In addition, the algorithm model with high precision and strong generalization ability constructed in the present application can be effectively used for rapid screening of the original gel formula, and the ionizable lipid molecules screened based on the model are verified to have the expected effect, with small error and high accuracy, thereby providing an advantageous means for the development of ionizable lipid molecules.

[0042] The in-situ gel formula provided in the present application has a suitable in-vitro release curve and in-vivo pharmacokinetic result, and has an equivalent or better drug delivery effect compared with the standard triamcinolone acetonide sustained-release dosage form known in the prior art, thereby providing more selectivity for the development of drug delivery carriers.

[0043] The molecular dynamics simulation method provided in the present application reveals the association mode, diffusion path and aggregation structure between the drug and the polymer through molecular dynamics simulation, thereby providing a microscopic explanation basis for the long-acting sustained-release mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0045] Figure 1 A flowchart of a formula determination method of a long-acting injection provided in an embodiment of the present application;

[0046] Figure 2 A flowchart of another formula determination method of a long-acting injection provided in an embodiment of the present application;

[0047] Figure 3A PBPK model fitting results for a triamcinolone acetonide intravenous administration dose of 2mg provided in an embodiment of the present application;

[0048] Figure 3B PBPK model fitting results for a triamcinolone acetonide intravenous administration dose of 400ug provided in an embodiment of the present application;

[0049] Figure 4 PBPK model fitting results for a triamcinolone acetonide suspension intra-articular administration dose of 40mg provided in an embodiment of the present application;

[0050] Figure 5A PBPK model fitting results for a triamcinolone acetonide microsphere intra-articular administration dose of 10mg provided in an embodiment of the present application;

[0051] Figure 5BPBPK model fitting results of the intra-articular administration dose of 40 mg of triamcinolone acetonide microspheres provided by the embodiment of the present application;

[0052] Figure 5C PBPK model fitting results of the intra-articular administration dose of 60 mg of triamcinolone acetonide microspheres provided by the embodiment of the present application;

[0053] Figure 6 Flowchart of the formula determination method of another long-acting injection provided by the embodiment of the present application;

[0054] Figure 7A PD model fitting results of the intra-articular administration dose of 20 mg of triamcinolone acetonide microspheres provided by the embodiment of the present application;

[0055] Figure 7B PD model fitting results of the intra-articular administration dose of 40 mg of triamcinolone acetonide microspheres provided by the embodiment of the present application;

[0056] Figure 8 Flowchart of the formula determination method of another long-acting injection provided by the embodiment of the present application;

[0057] Figure 9 In the "Me-too" and "Me-better" release kinetics, the plasma drug concentration and the average daily pain value change relative to the baseline provided by the embodiment of the present application are simulated;

[0058] Figure 10 Target in vitro release curve of the triamcinolone acetonide in-situ gel formula provided by the embodiment of the present application;

[0059] Figure 11 Flowchart of the formula determination method of another long-acting injection provided by the embodiment of the present application;

[0060] Figure 12 In vitro release curves and in vitro target release curve ranges of different triamcinolone acetonide in-situ gel formulas provided by the embodiment of the present application;

[0061] Figure 13 In vitro release curves and in vitro target release curve ranges of different triamcinolone acetonide in-situ gel formulas provided by the embodiment of the present application;

[0062] Figure 14 PK / PD simulation curve of the triamcinolone acetonide in-situ gel formula provided by the embodiment of the present application;

[0063] Figure 15 Pharmacokinetic parameters of intra-articular injection of triamcinolone acetonide in-situ gel and triamcinolone acetonide acetate suspension in rats provided by the embodiment of the present application;

[0064] Figure 16 The average plasma concentration / dose-time curve of triamcinolone acetonide after intra-articular injection of the triamcinolone acetonide in-situ gel (blue) and triamcinolone acetonide acetate suspension (pink) in rats provided by the embodiments of the present application;

[0065] Figure 17 The simulated plasma drug concentration (blue) of the triamcinolone acetonide in-situ gel and the change of the average daily pain value from the baseline (red) when the rats are administered with 40 mg provided by the embodiments of the present application;

[0066] Figure 18 The flowchart of the formula determination method of another long-acting injection provided by the embodiments of the present application;

[0067] Figure 19 The structural schematic diagram of the electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0068] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of description and illustration, and are not used to limit the protection scope of the present application. In addition, it should be understood that the schematic drawings are not drawn according to the actual proportions. The flowchart shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can not be implemented in sequence, and the steps without logical context relationship can be reversed in sequence or implemented simultaneously. In addition, one or more other operations can be added to the flowchart or one or more operations can be removed from the flowchart under the guidance of the content of the present application.

[0069] In addition, the described embodiments are only some of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0070] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0071] Firstly, the professional terms involved in the present application are explained.

[0072] 1. Long-acting injection, refers to the injection preparation that achieves long-term efficacy by controlling the drug release rate, i.e. the injection preparation that can release the drug in the body continuously after a single injection, thereby prolonging the duration of drug effect. Compared with ordinary injection, long-acting injection can significantly prolong the action time of the drug in the body, usually up to several days, several weeks, or even several months, thereby reducing the frequency of drug administration and improving patient compliance.

[0073] Among them, long-acting injections mainly include the following categories according to their release mechanism and dosage form:

[0074] 1) Microspheres

[0075] Principle: The drug is embedded in a degradable polymer (such as PLGA), and the drug release is controlled by polymer degradation; representative drugs include: Risperdal Consta, Zyprexa Sustenna;

[0076] 2) In situ forming gel

[0077] Principle: Phase change to form a gel matrix in the body environment after injection, and the drug is slowly diffused and released therein; representative solvents include: NMP (N-methyl pyrrolidone), benzyl alcohol, etc.; polymers: PLGA (poly(lactic-co-glycolic acid));

[0078] 3) Liposomes

[0079] Principle: The drug is wrapped in a lipid bilayer, and the release is controlled by the lipid membrane; representative drugs include: Doxil, amphotericin B liposomes;

[0080] 2. Reference drug preparation, refers to the known drug preparation that is used as "control" or "reference" in the present application, which is a long-acting injection that has been marketed and has a clear release behavior and drug effect.

[0081] Exemplarily, Zilretta triamcinolone acetonide sustained-release microspheres is a reference drug preparation, i.e. triamcinolone acetonide drug is wrapped in biodegradable polymer (such as PLGA), and the microspheres slowly degrade and release the drug in the joint cavity, achieving sustained anti-inflammatory and long-acting analgesia. Compared with traditional immediate-release triamcinolone acetonide injection, the drug effect duration of Zilretta triamcinolone acetonide sustained-release microspheres is significantly prolonged (from several weeks to more than 12 weeks).

[0082] 3. Physiologically Based Pharmacokinetic Model (PBPK) is a mathematical model based on physiology, anatomy, and the physicochemical properties of drugs. It is used to simulate the dynamic processes of drug absorption, distribution, metabolism, and excretion (ADME) in the body. Typically, the body of the test subject (e.g., rat) is divided into multiple compartments with clearly defined anatomical and physiological parameters (e.g., blood, liver, muscle, fat), and the transport of the drug between these compartments is simulated to predict the changes in blood drug concentration over time in each tissue.

[0083] 4. A pharmacodynamic model (PD model) describes the relationship between blood drug concentration and the intensity of pharmacological effects, i.e., how a drug produces therapeutic effects or side effects in the body. It typically uses blood drug concentration as the input variable and the output is the change in the intensity of the pharmacological effect (such as analgesia score, decrease in blood glucose, inflammatory factor levels, etc.). In other words, the output of the physiological pharmacokinetic model (i.e., the curve of blood drug concentration change in various tissues over time, also known as release behavior) can be used as the input of the pharmacodynamic model to evaluate the efficacy at different blood drug concentrations.

[0084] 5. The "Me-too" strategy refers to a new formulation whose release behavior and efficacy are comparable to the benchmark formulation.

[0085] 6. "Me-better" strategy: refers to a new formulation that is superior to the benchmark formulation in terms of release behavior or efficacy.

[0086] The following will provide a detailed description of the formulation determination method for the long-acting injectable provided in this application through several specific embodiments.

[0087] refer to Figure 1 The diagram shown illustrates a flowchart of a method for determining the formulation of a long-acting injectable drug according to this application. The method can be executed by an electronic device with data processing capabilities, such as a personal computer or laptop. It should be understood that in other embodiments, the order of some steps in the method for determining the formulation of the long-acting injectable drug can be interchanged according to actual needs, or some steps can be omitted or deleted. Figure 1 As shown, the method includes:

[0088] S101. Construct a physiological pharmacokinetic model of the benchmark drug formulation and a pharmacodynamic model of the benchmark drug formulation.

[0089] The physiological pharmacokinetic model is used to characterize the dynamic change process of drug absorption, distribution, metabolism and excretion in the body, and the pharmacodynamic model is used to characterize the relationship between different drug concentrations and pharmacological effects.

[0090] Exemplarily, the reference drug preparation is a known drug preparation used as a reference standard in the development process of long-acting injections, which is a drug preparation that has been marketed and has clear pharmacokinetic (PK) and pharmacodynamic (PD) characteristics, and is used for comparison with newly developed preparations to evaluate whether the performance of the new preparation meets the expected target.

[0091] In an implementable manner, after selecting the reference drug preparation, a mechanism-based mathematical model can be established based on the physicochemical properties of the reference drug preparation, the physiological parameters of the test subjects and the experimental data, and a PBPK model and a PD model of the reference drug preparation can be constructed by means of modeling software (such as PK-Sim, python) to simulate the release behavior of the drug in the test subjects and the change of the pharmacological effect, and finally to deduce the in vitro release target to guide the intelligent development of long-acting injections.

[0092] S102, according to the physiological pharmacokinetic model and the pharmacodynamic model, deducing the in vitro target release curve of the reference drug preparation.

[0093] The PBPK / PD model is not only used to simulate the pharmacokinetic process of the drug in the body, but also used to deduce the in vitro release behavior required to achieve the expected pharmacological effect.

[0094] Generally, the traditional preparation development adopts the trial-and-error method, that is, first do in vitro release, then test in vivo effect, and repeatedly adjust, which takes a long cycle of 8-10 years.

[0095] To solve the above problem, the present application proposes that the established PBPK model and PD model can be used to deduce the in vitro release curve characteristics required to achieve the in vivo pharmacokinetic (PK) and pharmacodynamic (PD) behavior, thereby guiding the design and screening of new preparation formulations and realizing intelligent and closed-loop long-acting injection development process.

[0096] S103, obtaining the in vitro predicted release curve of at least one alternative preparation formulation according to the pre-trained property prediction model.

[0097] The property prediction model can be trained based on any one of TabPFN, LightGBM, random forest, XGBoost, decision tree, multilayer perceptron (MLP), K-nearest neighbor algorithm (KNN), etc.

[0098] Exemplarily, the alternative formulation recipes can be in-situ gels, microspheres, sustained-release implants, wherein the formulation parameters of each alternative formulation recipe are different from each other, the formulation parameters referring to all the ingredients and their physical / chemical attributes that constitute the drug formulation, which together determine the drug release behavior (i.e. the speed and manner of drug release from the formulation).

[0099] In an implementable manner, the at least one alternative formulation recipe can be input into the property prediction model to obtain the in-vitro predicted release curves of each alternative formulation recipe output by the property prediction model. In this way, the property prediction model can be used to quickly predict the drug release curves of each alternative formulation recipe, so as to guide the intelligent design and screening of long-acting injections, greatly shorten the development cycle and improve the success rate.

[0100] S104. Screening a target formulation recipe from the at least one alternative formulation recipe according to the in-vitro predicted release curves of each alternative formulation recipe and the in-vitro target release curve.

[0101] In an implementable manner, a preset optimization algorithm such as Bayesian optimization, genetic algorithm, particle swarm optimization can be used, i.e. using the in-vitro target release curve as the quantitative design target, using the property prediction model to predict the in-vitro predicted release curves of each alternative formulation recipe, the optimization target being to minimize the deviation between the in-vitro predicted release curves and the in-vitro target release curve, and the in-situ gel formulation that maximizes the similarity factor between the in-vitro predicted release curves and the in-vitro target release curve is screened in the defined screening space, i.e. the best formulation that can achieve the target release behavior is automatically screened, so as to realize efficient and accurate development of long-acting injections, and shorten the development cycle of long-acting injections.

[0102] Optionally, in order to verify whether the screened target formulation formula can achieve equivalent (or better) efficacy with the benchmark drug formulation, it is also necessary to perform simulation verification, such as in vivo animal verification, on the target formulation formula to verify the in vivo drug exposure level of the target formulation formula. Among them, (1) the interaction mechanism between the drug and the polymer in the candidate formula can be simulated and verified by molecular dynamics simulation to obtain the micro interaction mechanism of the drug in the target formulation formula and its influence on the sustained release behavior of the drug, which provides an explanation and optimization basis at the molecular level for the screening of the target formulation formula; (2) pharmacokinetic experiments can be performed in animals (such as rats) to verify whether the target formulation formula achieves the expected efficacy (such as analgesic duration), supporting preclinical research and regulatory submission. Therefore, in the present application, molecular dynamics is used to simulate and verify the target formulation formula, which reveals the interaction mechanism between the drug and the target formulation formula at the molecular level; and in vivo pharmacokinetic experiments are performed on the target formulation formula to verify whether the target formulation formula can achieve the expected drug exposure level and efficacy in the real biological system, thereby realizing the whole-chain verification from the micro to the macro and from the simulation to the experiment, and providing a scientific basis for the research and development of the target formulation formula.

[0103] In summary, the embodiment of the present application provides a long-acting injection formula determination method, which comprises the following steps: constructing a PBPK model of a reference drug preparation and constructing a PD model of the reference drug preparation, that is, constructing a PBPK model and a PD model of a known drug preparation serving as a reference standard in the development process of the long-acting injection, to simulate the release behavior of the drug in the test subject and the change of the drug efficacy, and finally to be used to deduce the in-vitro release target, guide the intelligent development of the long-acting injection; then, starting from the pharmacokinetics (PK) and pharmacodynamics (PD) behavior of the drug in the body, the established PBPK model and PD model are used to reversely deduce the in-vitro release curve characteristics required to achieve the in-vivo behavior, so as to guide the design and screening of the new preparation formula, and realize the intelligent and closed-loop long-acting injection development process; according to the pre-trained property prediction model, the in-vitro prediction release curve of at least one candidate preparation formula is obtained, that is, the property prediction model is used to quickly predict the drug release curve of each candidate preparation formula, so as to be used to guide the intelligent design and screening of the long-acting injection, greatly shorten the development cycle and improve the success rate; according to the in-vitro prediction release curve of each candidate preparation formula and the in-vitro target release curve, a target preparation formula is screened from at least one candidate preparation formula, so as to realize efficient and accurate long-acting injection development and shorten the development cycle of the long-acting injection. Therefore, the present application solves the problems of long development cycle, poor predictability and insufficient in-vitro and in-vivo correlation of the long-acting injection, and proposes an intelligent closed-loop formula determination method integrating PBPK model and PD model and machine learning model simulation. The method takes calculation-driven as the core, combines the data-driven formula prediction with the mechanism-driven drug release mechanism explanation, so as to accelerate the formula design and optimization of the in-situ gel long-acting injection, and realize efficient and accurate long-acting injection development and shorten the development cycle of the long-acting injection.

[0104] The construction of the physiological pharmacokinetics model of the reference drug preparation will be specifically explained through the following examples.

[0105] Optionally, referring to Figure 2 As shown in the above step S101, comprising:

[0106] S201, obtaining the drug parameters of the reference drug preparation and the physiological parameters of the historical subjects.

[0107] Exemplarily, the selected reference drug preparation is triamcinolone acetonide microspheres. The drug parameters of the reference drug preparation include: physical and chemical properties (such as molecular weight, fat solubility, solubility, protein binding rate and pKa value), dose (mg / kg), injection site (such as joint cavity, vein), preparation type (such as sustained-release microspheres, in-situ gel), release behavior (such as burst release, diffusion, erosion stage).

[0108] The physiological parameters include: each tissue volume (such as liver, kidney, synovial membrane, etc.), blood flow rate (Q), tissue-plasma distribution coefficient (Kp), enzyme metabolism information (such as CYP enzyme activity, if not, use the system clearance Cl_sys instead), etc.

[0109] In an implementable manner, the intra-articular suspension or microsphere PK / PD research of triamcinolone acetonide in vivo can be collected from the drug approval database Clinical Pharmacology Biopharmaceutics Review and Medical Review in Drug Approval Package: ZILRETTA (triamcinolone acetonide) of the US Food and Drug Administration (FDA). The research on the intravenous pharmacokinetics of triamcinolone acetonide in PubMed is searched with "Triamcinolone acetonide" AND intravenous AND Pharmacokinetic" as the search terms and the screening condition of "adult" is applied. Finally, a total of 2 intravenous pharmacokinetic studies, 3 intra-articular suspension pharmacokinetic studies, 3 intra-articular microsphere pharmacokinetic studies, and 3 intra-articular microsphere PD studies are collected.

[0110] S202, a basic model of at least one chamber structure is established.

[0111] The basic model includes: a drug release model, and a disposition type of the drug in each chamber structure.

[0112] S203, the drug parameters of the reference drug preparation and the physiological parameters of the test subject are input into the basic model of each chamber structure to establish a physiological pharmacokinetic model of the reference drug preparation.

[0113] In an implementable manner, a PBPK / PD model and a construction process thereof are provided, which can be used to evaluate the pharmacokinetics of triamcinolone acetonide in situ gel. The specific construction process is as follows:

[0114] (1) Intravenous route PBPK model construction: the basic model of drug distribution and clearance in the whole body is established by using the collected human intravenous pharmacokinetic data in the literature. In the absence of specific enzyme metabolism information, the system clearance parameter is used instead of the enzyme-specific metabolism pathway. The tissue-plasma distribution coefficient (Kp value) is calculated by the rat tissue distribution experiment data, and the rest of the distribution and elimination parameters are derived from the literature and realized in the modeling in the software. Among them, the reference (2) Intra-articular suspension and microsphere route PBPK model construction: the basic model of drug distribution and clearance in the whole body is established by using the collected human intra-articular suspension and microsphere pharmacokinetic data in the literature. In the absence of specific enzyme metabolism information, the system clearance parameter is used instead of the enzyme-specific metabolism pathway. The tissue-plasma distribution coefficient (Kp value) is calculated by the rat tissue distribution experiment data, and the rest of the distribution and elimination parameters are derived from the literature and realized in the modeling in the software. Among them, the reference Figures 3A-3BPredicted and observed PK after single IV administration of 2 mg and 400 μg, respectively, solid line represents simulation results, dots represent raw observed data, model conditions are according to the original experimental design, results are expressed as concentration of triamcinolone acetonide.

[0115] (2) Joint cavity absorption model construction: For the microspheres or suspension formulations injected into the knee joint, a three-compartment absorption sub-model including "synovial cavity", "synovial membrane layer" and "sub-synovial tissue" three compartments was constructed, considering the release, diffusion and degradation process of the drug in different compartments. The model combines the Johnson equation to describe the dissolution equilibrium of the suspension, uses first-order kinetics to describe the degradation process of the drug in the synovial fluid, and uses the Noyes-Whitney diffusion model to describe the transport process of the drug from the synovial fluid to the synovial membrane. The release, diffusion and degradation of the drug between the compartments are jointly described by the following kinetic processes:

[0116] The dissolution process of the suspension drug is shown in the following formulas (1)-(2):

[0117]

[0118] The degradation rate of the drug in the synovial fluid is shown in the following formula (3):

[0119]

[0120] When the synovial fluid drug concentration tends to be saturated, the release at a constant rate is shown in the following formulas (4)-(5):

[0121]

[0122] tri = ifelse(t≤T tri ,1,0) (5)

[0123] The transport of the drug from the synovial fluid to the synovial membrane layer is shown in the following formulas (6)-(7):

[0124]

[0125] k subint,sys,in M sys (7)

[0126] In the above formulas (1)-(7), M syn , M int , M subint , M dis , M undis , M deg , M sys are the amounts of drug in the synovial fluid, intima, and extima, and in the dissolved, undissolved, degraded state, and in the systemic circulation, respectively. T triis the time point from Johnson to zero-order drug release. K trans，out and k subint,sys,out are the outflow rate constants between the synovial membrane circulation and the systemic circulation. In the joint model, K trans,in and k subint,sys,in are the inflow rate constants between the synovial membrane circulation and the systemic circulation. K deg is the degradation rate constant. SA int and SA subint are the surface areas of the intimal and the external membrane. h sol and h syn,free are the intimal and the external membrane thicknesses. f u,syn is the solubility of the drug in the synovial fluid, the free drug concentration, and the unbound fraction, respectively. p, h, r, s are the drug density, the diffusion layer thickness, the drug radius, and the shape factor, respectively.

[0127] (3) Coupling of the microsphere release function:

[0128] The simulation of the three-stage release behavior (burst, diffusion, and erosion) of the microspheres is shown in the following equation (8):

[0129]

[0130] The in vitro to in vivo scaling factor update for the microsphere release part is shown in the following equation (9):

[0131]

[0132] The in vivo drug release rate update for the synovial chamber is shown in the following equation (10):

[0133]

[0134] In the above equations (8)-(10), where Re% is the in vitro drug release percentage. X0 is the initial burst release percentage of the drug. Max is the maximum drug release percentage of the reservoir phase. b, a, c are the shape factor, the time scale factor, and the release rate of the reservoir phase, respectively. M r and M syn are the drug amount and the synovial fluid in the release state, respectively. X scal is the in vitro to in vivo release rate scaling factor. K trans is the rate constant in the joint model. SA int is the surface area of the intimal membrane. h int is the thickness of the intimal membrane.

[0135] Alternatively, as shown in Figure 4 , Figures 5A-5C Figure 4 ​Predicted and observed PK profiles for intra-articular administration of triamcinolone acetonide suspension Figure 5A Predicted and observed PK profiles for intra-articular administration of 10 mg triamcinolone acetonide microspheres, Figure 5B Predicted and observed PK profiles for intra-articular administration of 40 mg triamcinolone acetonide microspheres, and Figure 5C Predicted and observed PK profiles for intra-articular administration of 60 mg triamcinolone acetonide microspheres. The solid line represents the simulation results, and the dots represent the original observed data. The model conditions are based on the original experimental design, and the results are expressed as the concentration of triamcinolone acetonide.

[0136] Wherein, the predicted curve is the trend of drug concentration changing with time simulated by the PBPK model; the observed curve is the actual measured drug concentration changing with time by clinical experiment.

[0137] Therefore, the predicted curve and the observed curve shown in the above Figure 4 , can be compared to verify whether the constructed PBPK model can accurately simulate the absorption and distribution behavior of the suspension in the joint cavity. By comparing the predicted curve and the observed curve shown in the above Figures 5A-5C , it can be verified whether the constructed PBPK model can capture the slow-release behavior (such as burst release, diffusion, and erosion) stage of the microspheres.

[0138] Optionally, referring to Figure 6 , the pharmacodynamic model of the reference drug preparation in the above step S101 includes:

[0139] S301, obtaining a pre-constructed placebo effect model, and determining the change relationship between time and efficacy under the placebo effect model.

[0140] Wherein, the placebo effect model is used to characterize the pain relief effect caused by non-drugs in the simulation experiment.

[0141] S302, obtaining a pre-constructed drug effect model, and determining the change relationship between drug exposure and efficacy of the reference drug preparation under the drug effect model.

[0142] S303, obtaining the pharmacodynamic model of the reference drug preparation according to the change relationship between time and efficacy under the placebo effect model and the change relationship between drug exposure and efficacy of the reference drug preparation under the drug effect model.

[0143] In an implementable manner, the construction process of the PD model is as follows:

[0144] The efficacy part uses a transport-type PD model to evaluate the change of analgesic score (ΔADP total ) with time. The placebo effect model and the drug effect model are modeled as formulas (11)-(13) respectively:

[0145] ΔADP total =ΔADP placebo +ΔADP TCA (11)

[0146]

[0147] Among them, the placebo effect model is used to simulate the pain relief effect caused by non-drugs when no medication is taken; the drug effect model is used to simulate the actual analgesic effect caused by a benchmark drug formulation (such as triamcinolone).

[0148] In the above formulas (11)-(13), ΔADP total ΔADP placebo and ΔADP TCA The changes in ADP values ​​relative to baseline were due to treatment and placebo, respectively.

[0149] Therefore, the PD model of the benchmark drug formulation can be obtained according to the above formulas (11)-(12), as shown in formula (13):

[0150]

[0151] Wherein, ΔADP TCA I0 is the change in ADP value relative to baseline caused by the total effect of TCA, and I0 is the initial effect constant of TCA. max_p and I max_TCA These are the maximum effects of placebo and TCA, respectively, IC 50_p and IC 50_TCA These represent the time to reach the half-maximal response and the plasma drug concentration, respectively, where conc. is the plasma concentration of TCA. t_p and k t_TCA It is the rate constant of the effect of time on the placebo and TCA effects.

[0152] Optionally, refer to Figures 7A-7B The figures show the predicted and observed pain variability (PD) curves after intra-articular administration of triamcinolone acetonide microspheres at doses of 20 mg and 40 mg, respectively. The solid line represents the simulation results, and the dots represent the original observation data. Model conditions were based on the original experimental design. Results are expressed as daily average pain variation. Therefore, in Figures 7A-7B The accuracy of the PD model's prediction curve is determined by comparing the analgesia curve predicted by the PD model with the pain score measured in the experiment.

[0153] Optionally, refer to Figure 8 As shown, step S102 above includes:

[0154] S401, obtaining pre-set target pharmacodynamic response information.

[0155] S402, simulating blood concentration changes and therapeutic response information under at least one release curve according to a physiological pharmacokinetic model and a pharmacodynamic model.

[0156] Different release curves will result in different pharmacodynamics, such as the time required to achieve maximum analgesic effect (Tmax), blood concentration duration (AUC), and analgesic duration (ΔADPtotal).

[0157] Alternatively, different in vitro release curves can be input into the PBPK model and the PD model, the PBPK model is used to simulate the blood concentration curves (PK) generated in vivo by these release curves, and the PD model is used to simulate the therapeutic effect curves (PD) caused by these blood concentrations, i.e. to obtain blood concentration changes and therapeutic response information under multiple different release curves, so as to determine which release curves can achieve “equivalent to ” or “better than ” therapeutic effect, and further define the “release window” standard.

[0158] S403, identifying a release curve window that meets the target pharmacodynamic response information from at least one release curve.

[0159] S404, obtaining pre-defined key release parameters of the in vitro target release curve.

[0160] Exemplarily, the key release parameters of the in vitro target release curve include: initial burst ratio of 10-20%, release curve shape factor of 1.2-1.8, zero-order release rate of 0.04-0.07% / h, etc.

[0161] In the present application, by defining the key release parameters of the in vitro target release curve, it can be ensured that the screened formula release curve meets the therapeutic effect target, and the purpose of improving the development efficiency is achieved.

[0162] S405, obtaining the in vitro target release curve of the reference drug preparation according to the key release parameters of the in vitro target release curve and the release curve window.

[0163] In an implementable manner, in order to obtain the in vitro target release curve of the reference drug preparation, it is proposed that the key parameter range of the “ideal in vitro release behavior” (i.e. the release curve window) can be deduced and defined by using the PBPK model and the PD model from the pre-set target pharmacodynamic response information, and the in vitro target release curve of the reference drug preparation is constructed to perform formula optimization and experimental verification, forming a closed-loop development path from “in vivo therapeutic effect” to “in vitro design”. Specifically, the following steps are included:

[0164] (1) Simulation of in vivo exposure and therapeutic response under different release curves: The changes in blood drug concentration and ΔADP_total response corresponding to different release curves were evaluated through system simulation to determine whether the drug could achieve the effect of "applying to the FDA in 2017". "Me-too" and "superior" to triamcinolone acetonide sustained-release microspheres The release window standard is “Me-better”. Of course, those skilled in the art can also choose other well-known long-acting sustained-release formulations as standard formulations for comparison. The selection of a standard formulation will not change the compounds obtained from the screening, but is only used as a baseline.

[0165] (2) Define key parameters for the in vitro release target curve: Based on the simulation results of the PBPK and PD models, a quantitative in vitro release target range is proposed as a design guide for subsequent formulation screening and optimization. This includes, but is not limited to: an initial burst release ratio of 10–20%, balancing onset speed and subsequent maintenance; a release curve shape factor between 1.2 and 1.8 to ensure a smooth transition from the burst release phase to the maintenance release phase; a zero-order release rate controlled at 0.04–0.07% / h to ensure sustained achievement of therapeutic blood drug concentrations; and other low-sensitivity parameters (such as maximum burst release duration ≤55h, and release plateau phase not exceeding 10%) are fixed based on existing experimental data.

[0166] (3) Output the recommended in vitro target release curve: The above reverse process is based on the release function (including initial burst release, curvature factor and zero-order rate) defined by mathematical expression, and uses it as the target input for machine learning prediction, optimization algorithm search and experimental verification.

[0167] Optionally, the mean average error of deviation (AAFE) and average error of deviation (AFE) can be used to evaluate the fitting accuracy of the PBPK / PD model for different dosage forms. PK / PD reference standards are set according to different efficacy targets (such as duration of analgesia), and the standard triamcinolone acetonide long-acting sustained-release formulation is the one approved by the FDA in 2017. Triamcinolone acetonide sustained-release microsphere formulation. Further, based on the "Me-too" and "Me-better" clinical goals, the parameter ranges for the in vitro release curve were defined, including an initial burst release ratio of 10–20%, a release shape factor of 1.2–1.8, and a zero-order release rate of 0.04–0.07% / h, providing quantitative targets for subsequent formulation prediction and optimization.

[0168] Figure 9 The release kinetics simulation curves show changes in plasma drug concentration and ADP value from baseline for "Me-too" and "Me-better" drugs. Figure 10 The range of in vitro release profiles for the recommended triamcinolone acetonide in situ gel formulation is shown.

[0169] Optionally, referring to Figure 11 The step S103 includes:

[0170] S501, obtaining the formulation parameters of each alternative formulation.

[0171] S502, inputting the formulation parameters of each alternative formulation into the property prediction model respectively to obtain the in-vitro predicted release curve of each alternative formulation.

[0172] Generally, in the conventional drug formulation development process, the relationship between the formulation parameters of each alternative formulation and the drug release behavior is obtained by relying on the experience trial-and-error method, in-vitro release experiment or in-vivo animal experiment. However, the traditional method has the problem of long cycle. In order to improve the development efficiency of the formulation, the present application proposes that the artificial intelligence method can be used, that is, the formulation parameters of each alternative formulation are input into the property prediction model, and the in-vitro predicted release curve of each alternative formulation is predicted by the property prediction model.

[0173] Optionally, the step S104 includes:

[0174] The similarity factor between the in-vitro predicted release curve of each alternative formulation and the in-vitro target release curve is determined respectively using the pre-set closed-loop optimization strategy, and the alternative formulation corresponding to the maximum similarity factor is taken as the target formulation.

[0175] In an implementable manner, the present application provides an embodiment of screening an in-situ gel formulation, using a pre-set closed-loop optimization strategy to realize the screening of the target formulation, and the specific steps include:

[0176] (1) The optimization target is to minimize the deviation between the in-vitro predicted release curve of each alternative formulation and the in-vitro target release curve, and to maximize the similarity factor (f) with the in-vitro target release curve. In the optimization process, the active ingredient and the organic solvent are fixed, and the change variable is the polymer attribute and the concentration of each component. The search range is reasonably set based on literature and experimental experience to ensure the design feasibility.

[0177] In the present embodiment, the active ingredient is triamcinolone acetonide, the organic solvent is NMP, and the PLGA types are three: a. LA / GA molar ratio 50 / 50, inherent viscosity 0.46 dL / g, end group: acid end; b. LA / GA molar ratio 85 / 15, inherent viscosity 0.27 dL / g, end group: ester end; LA / GA molar ratio 75 / 25, inherent viscosity 0.52 dL / g, end group: ester end.

[0178] The similarity factor score between the predicted release curve and the in-vitro target release curve can be calculated according to the following formula (14) as follows:

[0179]

[0180] In formula (14), f2 is the similarity factor; n is the number of time points for calculation (usually 3-15 are selected); R t is the release percentage of the control preparation at the tth time point; T t is the release percentage of the test preparation at the tth time point; t represents different sampling time points. When f2≥50, it is considered that the two release curves have similarity; the maximum value of f2 is 100, indicating that the two curves are completely coincident.

[0181] (2) Formulation screening is performed using the Bayesian optimization algorithm, a target function is constructed, the input is the formulation variable, and the output is the release curve predicted by TabPFN. The Bayesian algorithm is used for iterative optimization to screen the candidate formulations whose release behavior falls within the target window.

[0182] In this embodiment, the following Table 1 is three candidate formulations F1-F3 generated by the first round of optimization, wherein the specific preparation method is: the PLGA polymer is dissolved in NMP solvent, and a blank polymer solution is prepared after ultrasonic homogenization, and triamcinolone acetonide is added after swelling at room temperature for 4-5 hours. All formulations are stored at 4°C.

[0183] Table 1 Formulation information of the first round of optimization

[0184]

[0185] The in-vitro release experiment is performed using a Distek eight-paddle device (USP Paddle II), the dissolution medium is 1000 mL of phosphate buffer (PBS, pH 7.2) containing 0.3% sodium dodecyl sulfate (SDS), the temperature is maintained at 35±0.5°C, and the stirring speed is 75 rpm. The sample containing 40 mg of TCA is injected into the dissolution medium through an 18G injection needle, and the sample is taken at regular intervals (a total of 23 time points), filtered through a 0.45 μm filter membrane, and then analyzed by UV spectrophotometry (λ_max=239 nm). The cumulative release amount is converted by the regression equation A=0.0331C+0.0069.

[0186] After verification by in-vitro experiments, it is found that the release rate is generally low and does not meet the burst release requirement, indicating that the model needs to improve the burst release prediction ability, Figure 12 are the in-vitro release curves of three different triamcinolone acetonide in-situ gel formulations, and the in-vitro target release curve range.

[0187] (3) Introduce experimental feedback to retrain the property prediction model, supplement the first round of experimental data (F1-F3) into the original data set, update the property prediction model, and enhance the ability to capture the early release stage.

[0188] In this embodiment, the second round of optimization is carried out based on the updated property prediction model, and the F4 candidate formula is obtained, wherein the specific preparation method is: dissolving PLGA polymer in NMP solvent, preparing blank polymer solution after ultrasonic homogenization, adding triamcinolone acetonide after swelling at room temperature for 4-5 hours. All formulations are stored at 4℃.

[0189] Second round of optimization formula information

[0190]

[0191] In vitro release experiment was carried out using Distek eight-paddle method device (USP Paddle II), the dissolution medium was 1000 mL of phosphate buffer (PBS, pH 7.2) containing 0.3% sodium dodecyl sulfate (SDS), the temperature was maintained at 35±0.5℃, and the stirring speed was 75 rpm. The sample containing 40 mg TCA was injected into the dissolution medium through an 18G injection needle, and the sample was taken at regular time intervals (a total of 23 time points), filtered through a 0.45 μm filter membrane, and analyzed by UV spectrophotometry (λ_max=239 nm), and the cumulative release amount was converted by the regression equation A=0.0331C+0.0069.

[0192] Figure 13 The in vitro release curves of three different triamcinolone acetonide in situ gel formulations, and the in vitro target release curve range.

[0193] In this embodiment, F4 has better in vitro release level than the "Me-too" effect, and in vivo pharmacokinetic evaluation experiment is carried out, and the specific experimental scheme is as follows:

[0194] Male Sprague-Dawley (SD) rats (body weight 250-350 g) were used in this study. The experimental animals were divided into two groups, 6 rats in each group. All animals only received intra-articular injection of knee joint once. The experimental group was injected with test triamcinolone acetonide in situ gel formula (1 unit of gel was injected per rat), the weight of the gel was 3.8±2.2 mg, the concentration of TCA was 11.59% (w / w), and the amount of TCA was about 0.42±0.24 mg; the control group was injected with commercially available triamcinolone acetonide injection suspension (batch number KD2217); the target TCA dose of the two groups was about 2.3 mg / kg.

[0195] Prior to injection, all animals were anesthetized with isoflurane and the knee joint area was shaved and disinfected with iodophor and 70% ethanol. The injection was performed under aseptic conditions using a calibrated microsyringe. After the injection was completed, all animals were fed under standard experimental conditions and monitored for potential adverse reactions.

[0196] Blood sample collection and pharmacokinetic parameter calculation: At multiple preset time points (1 h, 2 h, 4 h, 6 h, 8 h, 10 h, 12 h, 1 d, 2 d, 3 d, 4 d, 7 d, 14 d, 21 d, 28 d, 35 d, 42 d, and 49 d) after injection, blood samples were collected through the jugular vein and placed in tubes containing EDTA-K2 anticoagulant. Plasma was separated by centrifugation at 4000 rpm and 4°C for 15 minutes and stored at -80°C until analysis. Pharmacokinetic parameters were analyzed using DAS (V3.2.8) software for non-compartmental models, and the calculation indicators included: maximum plasma concentration (Cmax); peak time (Tmax); area under the curve (AUC 0- ); terminal elimination half-life (t / ). To eliminate the influence of differences during intra-articular injection, all PK parameters were normalized according to the actual injection dose. No deaths or treatment-related adverse reactions were observed during the study, indicating that the triamcinolone acetonide in situ gel formulation has good in vivo tolerability.

[0197] Blood concentration analysis method: The concentration of TCA in plasma was determined by ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS / MS, Waters UPLCI Class-TQ / XS). Chromatographic separation used a C18 analysis column under gradient elution conditions. The mass spectrometer used positive ion electrospray (ESI) mode, and the TCA and internal standard were detected using multiple reaction monitoring (MRM) mode; the internal standard was aripiprazole-D8 (concentration 100 μg / mL, Sigma company); the calibration standard was prepared by adding TCA of known concentration (range: 1000 ng / mL-100 μg / mL) to blank rat plasma; the calibration curve was established by plotting the peak area ratio of TCA to internal standard against the nominal concentration; each batch of analysis included quality control samples to ensure the accuracy and reproducibility of the results; the lower limit of quantification (LLOQ) was 0.5 ng / mL, ensuring reliable detection of low concentrations of TCA throughout the experimental period.

[0198] wherein, Figure 14 are the pharmacokinetic parameters after intra-articular injection of in situ gel and suspension in rats, wherein the blue curve represents the simulated plasma drug concentration for 24 weeks, and the red curve represents the corresponding pharmacodynamic response. The red shaded area represents the time during which the therapeutic effect is maintained.

[0199] Figure 15 are the plasma concentration / dose-time curves for intra-articular injection of in situ gel and suspension in rats.

[0200] Figures 16-17 The changes in simulated plasma drug concentration and ADP values from baseline for intra-articular in situ gel administration and suspension administration, respectively. The shaded area indicates the time required to maintain a reduced ADP level at at least three scaled points.

[0201] The experimental results show that the in vivo exposure level of the F4 optimized formula is better than that of the commercially available triamcinolone acetonide suspension, achieving the "Me-better" effect, and the analgesic duration is increased to 14.7 weeks.

[0202] Optionally, in some embodiments of the present application, the in situ gel formula comprises:

[0203] (1) LA / GA molar ratio 50 / 50, inherent viscosity 0.46 dL / g, end group: acid end. Drug loading ratio 12.85% w / w), PLGA percentage (24.15% w / w);

[0204] (2) LA / GA molar ratio 85 / 15, inherent viscosity 0.27 dL / g, end group: ester end. Drug loading ratio 12.53% w / w), PLGA percentage (22.17% w / w);

[0205] (3) LA / GA molar ratio 75 / 15, inherent viscosity 0.52 dL / g, end group: ester end. Drug loading ratio 12.80% w / w), PLGA percentage (24.10% w / w);

[0206] Optionally, the in situ gel formula can be reconstructed by researchers in the field based on conventional operations in the field after self-defined screening space and training data.

[0207] Optionally, referring to FIG. 1, the training process of the property prediction model comprises: Figure 18

[0208] S601, obtaining a training data set.

[0209] The training data set comprises a plurality of historical formulation formulas and an in vitro release curve corresponding to each historical formulation formula.

[0210] In order to improve the accuracy of the finally trained property prediction model, it is proposed that the formulation information of a plurality of candidate formulation formulas and the actual in vitro release curve of each candidate formulation formula can be added to the training data set to iteratively optimize the property prediction model.

[0211] S602, preprocessing the training data set to obtain a preprocessed training data set.

[0212] The preprocessing comprises data cleaning, molecular structure conversion, and feature encoding. ​

[0213] Optionally, the role of the pretreatment is to keep as much data as possible while improving the uniformity of the in-situ gel formulation data. Specifically, improving the uniformity of the in-situ gel formulation data includes removing data of combination drugs, removing data of multi-polymer or multi-solvent systems, and removing data of PLGA-PEG copolymers or carrier-gel combined systems.

[0214] The standard in-situ gel formulation is an in-situ gel formulation of a pharmaceutical active ingredient, a polymer PLGA, and a solvent.

[0215] Optionally, in some embodiments, the encoding method includes using simplified molecular linear input specification (SMILES) and molecular physical and chemical property descriptors for encoding, and using one-hot encoding for classification variables.

[0216] S603, iteratively train the initial prediction model using the pretreated training data set until the new prediction model obtained by the current training meets the preset condition, and then use the new prediction model as the property prediction model.

[0217] In this embodiment, the training process of the property prediction model specifically includes the following steps:

[0218] (1) Collect and organize existing in-situ gel formulation experimental data from databases and / or literature to construct a training data set. The information contained in the obtained experimental data needs to include the characteristics of the pharmaceutical active ingredient: molecular weight, hydrogen bond acceptor number, melting point, logP, logS, etc.; the characteristics of the polymer PLGA: LA / GA molar ratio, molecular weight, end group, intrinsic viscosity, concentration; the characteristics of the organic solvent: viscosity, dielectric constant, polar surface area, melting / boiling point, vapor pressure; dissolution conditions: pH, temperature, stirring speed, medium volume, surfactant concentration, etc.; and other characteristics: first-day release amount, used to capture the burst release behavior. Among them, the data of the pharmaceutical active ingredient, the polymer, the organic solvent, and the dissolution conditions can be represented by classification variables and chemical descriptors, and other data can be represented by conventional representation methods.

[0219] After the collection of the training data set is completed, subsequent data processing work is also carried out to improve the uniformity of the data while maintaining as much data as possible. The data processing standard is that all samples exclude combination drugs, multi-polymer or multi-solvent systems, and the use of PLGA-PEG copolymers or carrier-gel combined systems. Finally, the collected data set contains a total of 197 in-situ gel formulations, including 2500 cumulative release time points.

[0220] (2) Extract all the molecular structures of the drug active ingredients from the compounds in the above training dataset, and convert these molecular structures into their simplified molecular linear input specification (SMILES).

[0221] In this embodiment, the extracted information is specifically: drug active ingredient characteristics (molecular weight, hydrogen bond acceptor number, melting point, logP, logS, etc.); polymer PLGA characteristics (LA / GA molar ratio, molecular weight, end group, intrinsic viscosity, concentration); solvent characteristics (viscosity, electric constant, polar surface area, melting / boiling point, vapor pressure); dissolution conditions (pH, temperature, stirring speed, medium volume, surfactant concentration, etc.); other characteristics: release_day1, used to capture the burst behavior.

[0222] In this embodiment, all molecular characteristics are calculated by SMILES via RDKit and public databases, and categorical variables such as PLGA end groups and dissolution methods are processed using one-hot encoding.

[0223] (3) According to the input information in step (2), a model is established to correlate the input information with the possible in-situ gel formulation drug cumulative release percentage. The construction algorithm for establishing the model can be TabPFN, LightGBM, random forest, XGBoost, decision tree, multilayer perceptron, etc. In this embodiment, the construction algorithm for establishing the model is TabPFN.

[0224] Each release curve is taken as a training unit, and the cumulative release value at different time points is taken as the prediction target. GroupKFold is used for cross-validation (5x5 nested strategy), and the inner loop is used for hyperparameter search, and the outer loop is used for generalization performance evaluation. The evaluation indicators include MAE, RMSE and determination coefficient R 2 , ensuring that the training-validation process is strictly separated to avoid data leakage. These indicators are calculated by the following formulas (15)-(17).

[0225]

[0226] where RMSE is only a measure of the size of the error between the predicted value and the actual value. MAE is another measure for evaluating the prediction error. MAE is similar to RMSE, which measures the average absolute difference between the predicted value and the actual value. But different is that MAE pays more attention to the absolute error rather than their square values. R 2 is a statistical quantity indicating the fitting degree of the model and the data, and the closer the score is to 1, the better the fitting degree.

[0227] Table 2 below is a performance test of the model constructed using the above training dataset, and the results are as follows:

[0228] Table 2 Performance of TabPFN model predicting cumulative percentage drug release of in situ gel formulation on the collected dataset

[0229]

[0230] From the above Table 2, it can be obtained that the TabPFN model predicting cumulative percentage drug release of in situ gel formulation has good performance on the dataset containing the first-day release percentage feature or not, R 2 are all above 0.8.

[0231] (4) In this embodiment, the feature importance of input information (parameters) on the model output is calculated by SHAP algorithm after the model training.

[0232] In this embodiment, evaluating the in situ gel formulation information helps to provide interpretability for the machine learning model and which formulation or molecular structure feature is beneficial or detrimental to drug release.

[0233] Optionally, the method further comprises:

[0234] The interaction mechanism between the drug and the polymer in the target formulation is simulated and verified by using a preset molecular dynamics model, and a verification result of the target formulation is obtained.

[0235] In this embodiment, a molecular dynamics simulation method is provided to reveal the micro-interaction mechanism of drug molecules in the in situ gel system and its influence on the sustained release behavior of the drug, which specifically includes the following steps:

[0236] (1) Constructing initial model and parameter file: using GROMACS2023 software as the simulation platform, the force field adopts GAFF (General Amber Force Field). Among them: the structure of small molecule drug triamcinolone acetonide and PLGA monomer is optimized in Gaussian 16 at B3LYP(D3) / 6-311G** level; the topology structure of TCA is generated by Sobtop, and the PLGA chain is constructed by ztop tool according to the 40:40 ratio of lactic acid and glycolic acid monomers, with a molecular weight of about 5218, close to the experimental range; all molecules are calculated by electrostatic potential constraint method (RESP), and the parameters are derived from Multiwfn program.

[0237] (2) Simulating the formation process of in situ gel system: according to the experimental ratio, first shrink the PLGA chain alone in the vacuum box to improve the compactness of the system, and then put it and TCA together in 12x12x16 nm 3The mixture was placed in a container and NMP solvent was added. Non-physical contact was eliminated by minimizing energy. A 110 ns simulation was performed using a four-stage heating and cooling annealing process (298 K, 323 K, 323 K, 298 K) to simulate the ultrasonic homogenization process in the experiment, ensuring uniform mixing of the system; a final stable gel structure with a size of approximately 7 × 7 × 10 nm was formed. 3 This is for subsequent dissolution simulation.

[0238] (3) Simulate the dissolution process: Place the gel model at a 14×14×20nm depth. 3 In aquatic environments, add SDS and Na + Cl - Ions and the SPC water model. Simulation conditions are as follows: isothermal and isobaric ensemble (NPT), temperature 298K (velocity rescaling temperature controller), pressure 1 bar (stochastic scale rescaling pressure controller); time step 0.002 ps, total simulation time 500 ns; PME algorithm is used to calculate long-range Coulomb interactions, and the short-range cutoff distance is set to 1.2 nm.

[0239] (4) Analysis of drug-polymer interaction behavior: Trajectories were collected at time points of 50 ns, 200 ns, 350 ns and 500 ns for the following microscopic analysis: the number of hydrogen bonds and contacts between TCA and PLGA were counted; the minimum distance between each TCA molecule and the nearest PLGA chain was measured (usually distributed in 0.14–0.18 nm); after removing NMP and SDS, the aggregation behavior of PLGA and the embedding position of TCA were observed, and it was found that TCA was mostly located on the polymer surface or in the voids; the diffusion coefficient (D) of each TCA molecule was calculated, and its distribution was wide and uniform, indicating that the system has the potential for sustained release.

[0240] Optionally, refer to Figure 19 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. The electronic device can be a personal computer, laptop computer, or other device with data processing capabilities.

[0241] The electronic device includes: processor 2201 and memory 2202.

[0242] The memory 2202 is used to store programs, and the processor 2201 calls the programs stored in the memory 2202 to execute the above method embodiments. The specific implementation and technical effects are similar, and will not be described again here.

[0243] Optionally, the present invention also provides a program product, such as a computer-readable storage medium, including a program that, when executed by a processor, is used to perform the above-described method embodiments.

[0244] In several embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the described apparatus embodiments are merely schematic. The division of the units is merely a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0245] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0246] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can be physically present separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0247] The integrated unit implemented in the form of software functional units can be stored in a computer readable storage medium. The software functional unit stored in the storage medium includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to execute part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (English: Read-Only Memory, abbreviated as: ROM), a random access memory (English: Random Access Memory, abbreviated as: RAM), a magnetic disk or an optical disk, and various program code storage media.

Claims

1. A method for determining a formulation of a long-acting injectable, characterized by, The method comprises: constructing a physiological pharmacokinetic model of a benchmark drug preparation, and constructing a pharmacodynamic model of the benchmark drug preparation, wherein the physiological pharmacokinetic model is used to characterize the dynamic change process of drug release, distribution, metabolism and excretion in vivo, and the pharmacodynamic model is used to characterize the relationship between different drug concentrations and drug effects; backing out an in vitro target release curve of the benchmark drug preparation according to the physiological pharmacokinetic model and the pharmacodynamic model; obtaining an in vitro predicted release curve of at least one alternative preparation formula according to a pre-trained property prediction model, wherein the property prediction model is used to predict the in vitro drug release curve of the long-acting injection formula; screening a target preparation formula from the at least one alternative preparation formula according to the in vitro predicted release curve of each alternative preparation formula and the in vitro target release curve.

2. The method of claim 1, wherein, The method comprises: obtaining drug parameters of the benchmark drug preparation and physiological parameters of a test subject; establishing at least one basic model of a house chamber structure, wherein the basic model comprises a drug release model and a drug disposition model in each house chamber structure; inputting the drug parameters of the benchmark drug preparation and the physiological parameters of the test subject into the basic model of each house chamber structure to establish a physiological pharmacokinetic model of the benchmark drug preparation.

3. The method of claim 1, wherein, The method comprises: obtaining a pre-constructed placebo effect model and determining the change relationship between time and drug effect under the placebo effect model, wherein the placebo effect model is used to characterize the pain relief effect caused by non-drugs in a simulation experiment; obtaining a pre-constructed drug effect model and determining the change relationship between drug exposure and drug effect of the benchmark drug preparation under the drug effect model; obtaining a pharmacodynamic model of the benchmark drug preparation according to the change relationship between time and drug effect under the placebo effect model and the change relationship between drug exposure and drug effect of the benchmark drug preparation under the drug effect model.

4. The method of claim 1, wherein, The method comprises: obtaining pre-set target drug effect response information; simulating blood drug concentration changes and drug effect response information under at least one release curve according to the physiological pharmacokinetic model and the pharmacodynamic model; identifying a release curve window meeting the target drug effect response information from the at least one release curve; obtaining pre-defined key release parameters of the in vitro target release curve; obtaining an in vitro target release curve of the benchmark drug preparation according to the key release parameters of the in vitro target release curve and the release curve window.

5. The method of claim 1, wherein, The method comprises: obtaining formula parameters of each alternative preparation formula; inputting the formula parameters of each alternative preparation formula into the property prediction model respectively to obtain an in vitro predicted release curve of each alternative preparation formula.

6. The method of claim 1, wherein, The in-vitro predicted release curve of each of the alternative formulation recipes is screened from the at least one alternative formulation recipe to obtain a target formulation recipe, including: A pre-set closed-loop optimization strategy is used to determine a similarity factor between the in-vitro predicted release curve of each of the alternative formulation recipes and the in-vitro target release curve, and the alternative formulation recipe corresponding to the maximum similarity factor is taken as the target formulation recipe.

7. The method of claim 1, wherein, The training process of the property prediction model includes: Obtaining a training data set, the training data set including a plurality of historical formulation recipes and an in-vitro release curve corresponding to each historical formulation recipe, and the training data set further including a plurality of candidate formulation recipes and an actual in-vitro release curve of each candidate formulation recipe; Preprocessing the training data set to obtain a preprocessed training data set; Using the preprocessed training data set to iteratively train an initial prediction model until a new prediction model obtained by current training meets a pre-set condition, and then taking the new prediction model as the property prediction model.

8. The method according to any one of claims 1 to 7, characterized in that, Further including: Using a pre-set molecular dynamics model to simulate and verify the interaction mechanism between the drug and the polymer in the target formulation recipe to obtain a verification result of the target formulation recipe.

9. An electronic device, comprising: Including: A processor, a storage medium and a bus, the storage medium storing machine-readable instructions executable by the processor, when the electronic device is running, the processor and the storage medium communicate through the bus, the processor executes the machine-readable instructions to perform the steps of the method of any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is executed by the processor to perform the steps of the method of any one of claims 1-8.

Images