Protection liquid component optimization method for freeze-drying of exosome positive quality control product

Abstract

The invention provides a protection liquid component optimization method for freeze-drying of an exosome positive quality control product, and belongs to the technical field of exosomes, and the protection liquid component optimization method comprises the following steps: firstly, establishing a candidate component library containing a buffer liquid component, a protective agent component and an additive component, and collecting detailed physicochemical property data of each component; then, screening the candidate components based on parameters such as component compatibility, acid-base stability, osmotic pressure adaptability and the like to obtain a primary component data set; next, key characteristic indexes, including physical, chemical and biological characteristics, influencing the stability of the exosome are extracted by using a principal component analysis method; thirdly, constructing an exosome stability prediction model by adopting a double-layer machine learning algorithm, and optimizing the exosome stability prediction model through cross validation; and finally, setting an optimization objective function based on a result output by the prediction model, optimizing the type and proportion of the primarily selected components by adopting a genetic algorithm, preparing the exosome freeze-drying protection liquid, and obtaining a final efficient exosome freeze-drying protection liquid formula.

Description

Technical Field

[0001] The present invention belongs to the technical field of exosomes, and specifically relates to a method for optimizing the components of a protective solution for freeze-drying exosome positive control products. Background Art

[0002] As an important class of intercellular signal transduction carriers, exosomes have shown broad application prospects in the fields of cell communication, immune regulation, therapeutic targeted delivery, etc. However, the naturally low yield and easy degradation of exosomes severely limit their practical applications in the biomedical field. Therefore, how to improve the separation, concentration, and storage stability of exosomes has become a key technical problem to be solved urgently. Currently, freeze-drying technology, as an effective method for exosome preservation, has attracted much attention. By removing water, the degradation and inactivation of exosomes can be effectively inhibited. However, during the freeze-drying process, exosome particles are prone to aggregation and breakage, and their biological activity will also be significantly reduced. To solve this problem, adding freeze-drying protectants is a common strategy. Protectants can effectively maintain the integrity and biological functions of exosomes through various mechanisms, such as osmotic pressure regulation and membrane structure stabilization.

[0003] However, due to the very complex composition of exosomes, there are significant differences in the requirements for protectants among exosomes from different sources, and it is difficult to find a universal protectant formula. The optimization of the existing exosome freeze-drying protective solution formula mostly relies on empirical trial and error, with low efficiency and difficulty in systematically analyzing influencing factors.

[0004] Therefore, there is an urgent need to establish a systematic method for optimizing the exosome freeze-drying protective solution formula, which can comprehensively analyze the key factors affecting exosome stability and quickly screen out the optimal protectant ratio that meets the stability requirements in multiple aspects. Summary of the Invention

[0005] In view of this, the present invention provides a method for optimizing the components of a protective solution for freeze-drying exosome positive control products, which can solve the technical problems that the optimization of the existing exosome freeze-drying protective solution formula mostly relies on empirical trial and error, with low efficiency and difficulty in systematically analyzing influencing factors.

[0006] The present invention is implemented as follows:

[0007] The present invention provides a method for optimizing the composition of a cryoprotectant for freeze-drying exosome positive control products, including: establishing a candidate component library for the exosome freeze-drying cryoprotectant; collecting the physicochemical property data of all components in the candidate component library; screening the components in the candidate component library based on component compatibility parameters, acid-base stability parameters, and osmotic pressure adaptability parameters to obtain a preliminary selected component data set; extracting the characteristics of the components in the preliminary selected component data set, and screening out the key characteristic indicators affecting exosome stability by using the principal component analysis method; collecting exosome stability evaluation data to establish a training data set; constructing an exosome stability prediction model by using a double-layer machine learning algorithm; training and optimizing the parameters of the exosome stability prediction model by using the training data set; setting an optimization objective function and constraint conditions based on the prediction results of the exosome stability prediction model, and using a genetic algorithm to perform multi-generation iterative optimization on the component types and ratios in the preliminary selected component data set; preparing an exosome freeze-drying cryoprotectant according to the optimal component formula output by the genetic algorithm, and optimizing and updating the exosome stability prediction model through a freeze-drying protection performance verification experiment to obtain the final exosome freeze-drying cryoprotectant formula. Specifically, it includes the following steps:

[0008] S10. Establish a candidate component library for the exosome freeze-drying cryoprotectant, and the candidate component library includes buffer components, protective agent components, and additive components;

[0009] S20. Collect the physicochemical property data of all components in the candidate component library;

[0010] S30. Screen the components in the candidate component library based on component compatibility parameters, acid-base stability parameters, and osmotic pressure adaptability parameters to obtain a preliminary selected component data set;

[0011] S40. Extract the characteristics of the components in the preliminary selected component data set, and screen out the key characteristic indicators affecting exosome stability by using the principal component analysis method;

[0012] S50. Collect exosome stability evaluation data to establish a training data set, and the exosome stability evaluation data includes particle size distribution data, surface charge data, membrane integrity data, and biological activity data;

[0013] S60. Construct an exosome stability prediction model by using a double-layer machine learning algorithm. The exosome stability prediction model includes a convolutional neural network model and a recurrent neural network model. Input the key characteristic indicators into the convolutional neural network model and the recurrent neural network model, fuse the outputs of the two models through the fusion pooling layer, and obtain the exosome stability prediction result after processing by the fully connected layer;

[0014] S70. Use the training dataset to train the convolutional neural network model and the recurrent neural network model respectively, and adopt the cross-validation method to optimize the parameters of the exosome stability prediction model;

[0015] S80. Based on the prediction results of the exosome stability prediction model, set the exosome stability optimization objective function and the component ratio constraint conditions, and use the genetic algorithm to perform multi-generation iterative optimization on the component types and ratios in the primary selected component dataset;

[0016] S90. Prepare the exosome lyophilization protective solution according to the optimal component formula output by the genetic algorithm, and optimize and update the exosome stability prediction model through the lyophilization protection performance verification experiment to obtain the final exosome lyophilization protective solution formula.

[0017] On the basis of the above technical solutions, the method for optimizing the components of the protective solution for exosome positive control product lyophilization of the present invention can be further improved as follows:

[0018] Among them, the buffer component is composed of phosphate buffer, citrate buffer, and 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid buffer; the protective agent component is composed of sucrose, mannitol, trehalose, and dextran; the additive component is composed of amino acid substances, protein substances, and inorganic salt substances.

[0019] Further, the physicochemical property data includes molecular weight data, polarity data, solubility data, osmotic pressure data, pH value data, charge data, and affinity data.

[0020] Further, the key characteristic indexes include physical characteristic indexes, chemical characteristic indexes, and biological characteristic indexes.

[0021] Further, the stability prediction model adopts a two-layer model, including a first-layer model and a second-layer model;

[0022] The first-layer model adopts a convolutional neural network model, including an input layer, two convolutional layers, two activation layers, and a pooling layer; the second-layer model adopts a recurrent neural network structure, including an input layer, an embedded mathematical model, a long short-term memory unit layer, and a pooling layer.

[0023] Further, a fusion pooling layer is connected after the pooling layer of the first model and the pooling layer of the second model, which is used to fuse the outputs of the two pooling layers into a feature vector;

[0024] A fully connected layer and an output layer are connected after the fusion pooling layer. The fully connected layer is used for feature dimensionality reduction and non-linear mapping, and the output layer is used to generate an exosome stability prediction score;

[0025] The embedded mathematical model is a system of mathematical equations, including an osmotic pressure equation, a membrane stability equation, a particle size distribution equation, and an activity retention equation;

[0026] The first-layer model generates large-scale component ratio prediction results, stability evaluation values, and protection effect prediction values in a data-driven manner; the second-layer model generates small-scale osmotic pressure calculation results, membrane stability prediction values, and activity retention evaluation values using an accurate mathematical model.

[0027] Furthermore, the osmotic pressure equation is used to calculate the osmotic pressure of the protective solution. The inputs are component concentration and temperature parameters, and the output is the osmotic pressure value; specifically, it is expressed as follows:

[0028]

[0029] In the formula, π is the osmotic pressure (kPa); i is the degree of dissociation; M is the molar concentration of the solute (mol / L); R is the gas constant (8.314 J / (mol·K)); T is the temperature (K); v is the ionic strength; A, B, C are virial coefficients; V j is the molar volume (L / mol) of the jth non-electrolyte component; φ j is the volume fraction of the jth non-electrolyte component.

[0030] The first term iMRT of this equation is the basic term of the van't Hoff equation, which describes the osmotic pressure of an ideal dilute solution. Among them, i represents the degree of dissociation, considering the influence of electrolyte dissociation, and MRT reflects the basic contributions of temperature and concentration to the osmotic pressure; the second term (1 + Av 1 / 2 + Bv + Cv 3 / 2 ) is the virial series expansion term. Among them, v 1 / 2 term reflects the long-range ion-ion interaction; the v term represents the short-range interaction; the v 3 / 2 term describes the three-body interaction. This increasing power relationship can more accurately describe the behavior of non-ideal solutions. The third term represents the contribution of non-electrolytes and describes non-ideal behavior using a logarithmic relationship. Among them, 1 - φ j considers the excluded volume effect; this equation combines the comprehensive effects of electrolytes and non-electrolytes, considers multiple intermolecular interactions, and is applicable to high-concentration protective solution systems.

[0031] Furthermore, the membrane stability equation is used to evaluate the stability of the membrane structure. The inputs are ionic strength and pH value parameters, and the output is the membrane stability index; specifically, it is expressed as follows:

[0032]

[0033] In the formula, S mis the membrane stability index; k1, k2, k3 are rate constants; E a is the apparent activation energy (kJ / mol); c i is the concentration of the i-th component; β i is the component influence coefficient; γ is the surface tension (mN / m); t is the time (h); ζ is the real-time Zeta potential (mV); ζ0 is the initial Zeta potential (mV).

[0034] The first term of this equation Based on the Arrhenius equation, it describes the effect of temperature on membrane stability, and the exponential relationship reflects the activation energy barrier; the second term represents the synergistic effect of each component. Among them, the product form reflects the synergistic effect between components, and the power relationship describes the non-linear concentration effect; the third term represents interfacial kinetics. The derivative form describes the dynamic process, and the linear relationship assumes the direct effect of surface tension change; the fourth term k3|ζ - ζ0| 2 describes charge stability. Among them, the quadratic form reflects the penalty for deviating from the equilibrium state, and the absolute value ensures the equivalence of positive and negative deviations. This equation comprehensively considers thermodynamic, kinetic and electrical factors, introduces the dynamic change of surface tension, and considers the non-linear effect of potential deviation.

[0035] Furthermore, the particle size distribution equation is used to predict the trend of particle size change. The inputs are the molecular weight and concentration parameters of the protective agent, and the output is the particle size distribution curve; specifically expressed as follows:

[0036]

[0037] In the formula, f(d) is the particle size distribution density function; d is the particle size (nm); μ is the average particle size (nm); σ is the standard deviation; H k is the Hermite polynomial of order k; a k is the expansion coefficient; p is the number of expansion terms.

[0038] The first term of this equation is the Gaussian distribution term, which is used to describe the basic particle size distribution, considering the normal distribution assumption based on the central limit theorem; the second term is the Hermite polynomial correction term, which is used to describe the part deviating from the normal distribution.

[0039] Furthermore, the activity retention equation is used to calculate the biological activity. The inputs are the type of protective agent and the additive ratio parameters, and the output is the activity retention rate; specifically expressed as follows:

[0040]

[0041] In the formula, A is the real-time activity; A0 is the initial activity; k dis the inactivation rate constant; t is the time (h); θ is the surface coverage; K m is the protective agent binding constant; c p is the concentration of the protective agent; η is the synergistic coefficient; w l is the weight of the l-th additive; φ l is the protection efficiency of the l-th additive; q is the number of additives.

[0042] The first term of this equation, A0exp(-k d t) describes the natural inactivation process, where the exponential decay reflects first-order kinetics and the time dependence conforms to actual observations; the second term describes the action of the protective agent. Among them, the fractional form is similar to Langmuir adsorption, considering the saturation effect of the protective agent concentration; the third term describes the synergistic effect of additives. Among them, the linear summation reflects independent contributions, and the weight coefficient reflects the difference in importance.

[0043] The following is a detailed description of the methods for obtaining each parameter:

[0044] Determination of the degree of dissociation i: First, a series of electrolyte solutions with different concentrations need to be prepared, and the molar conductivity Λ of these solutions is measured using a conductivity meter under constant temperature conditions (usually 25 °C) m , and the molar conductivity at infinite dilution is obtained by extrapolation Finally, the degree of dissociation is calculated through the formula . The purity of the solution and the stability of the measurement environment need to be ensured throughout the process.

[0045] M is obtained by weighing and calculating, where m is the mass of the solute (g), M is the molar mass (g / mol), and V is the volume of the solution (L).

[0046] Determination of the ionic strength v: When using ion chromatography for determination, first prepare a series of standard solutions containing the target ions (at least 5 concentration gradients), analyze them using an ion chromatograph to obtain the standard curves of each ion, then dilute and filter the sample to be measured and inject it for analysis, qualitatively identify through the retention time of the chromatographic peak, quantitatively analyze through the integration of the peak area, and finally calculate the ionic strength based on the concentration of each ion where c i is the ion concentration, z i is the ion charge number.

[0047] Obtaining the virial coefficients A, B, C: By measuring the osmotic pressure of solutions with different concentrations, establish a system of equations Use the least squares method to perform multiple linear regression analysis on the experimental data to solve for the values of the virial coefficients A, B, C. The accuracy of the experimental data and the significance of the regression analysis need to be ensured during this process.

[0048] V j Determined by the density bottle method. Step 1: Measure the density of the pure component; Step 2: Calculate where M j is the molar mass, ρ j is the density; φ j Calculated by volumetric analysis, where c j is the concentration.

[0049] Obtaining the rate constants k1, k2, k3: Conduct membrane stability kinetic experiments under different temperature conditions (at least 3 temperature points), monitor the change curve of the membrane stability index over time. After collecting sufficient kinetic data points, use the non-linear least squares method to fit the experimental data and solve for the rate constant values at different temperatures. The entire process requires controlling the consistency of experimental conditions and the repeatability of data.

[0050] E a Obtained by calculating through the Arrhenius equation. Component influence coefficient β i Obtaining: Design an orthogonal experimental scheme, select appropriate factor levels, conduct multi-factor orthogonal experiments, measure the membrane stability index under different component ratios, calculate the influence degree of each component on membrane stability through variance analysis method, and finally determine the influence coefficient β i value of each component. The experimental design needs to consider the independence and interaction of factors. γ is determined by a surface tensiometer, and ζ is determined by a Zeta potentiometer. μ, σ are determined by dynamic light scattering method.

[0051] H k Obtained by numerical calculation,

[0052] Obtaining the Langmuir binding constant K related to activity retention m : First, determine the binding enthalpy change between the protectant and exosomes by isothermal titration calorimetry to obtain thermodynamic parameters; at the same time, use fluorescence quenching method to study the binding kinetics and measure the change of fluorescence intensity in the presence of different concentrations of protectant; finally, obtain the binding constant K through non-linear fitting m , and the temperature and pH conditions need to be strictly controlled during the experimental process. A0 is obtained by enzyme activity determination, and k d is calculated by kinetic experiments, where t is time; θ is determined by surface adsorption experiment. Step 1: Prepare solutions with different concentrations; Step 2: Measure the adsorption amount; Step 3: Fit the Langmuir isotherm.

[0053] Obtaining the synergy coefficient η: Design single-factor experiments to investigate the individual effects of each additive, then conduct two-factor experiments to study the interactions between additives. Calculate the synergy index based on the activity measurement results, and finally obtain the synergy coefficient η through mathematical model fitting. The experimental design needs to consider the possible antagonistic effects between additives.

[0054] Additive weight w l and protection efficiency φ l Obtaining: By designing a multi-level orthogonal experimental scheme, systematically investigate the effects of different additive combinations on exosome activity, conduct activity measurement experiments, and use multivariate statistical analysis methods to process experimental data. Calculate the weight coefficients and protection efficiency of each additive. The experimental scheme needs to consider the interaction effects and ratio optimization between additives.

[0055] Compared with the prior art, the beneficial effects of a method for optimizing the components of a protective solution for freeze-drying exosome positive control products provided by the present invention are as follows:

[0056] 1. A comprehensive candidate component library for exosome freeze-drying protective solutions is established, including buffer components, protective agent components, and additive components, providing a wide range of choices for subsequent screening and optimization.

[0057] 2. Detailed physicochemical property data of the candidate components are collected, including parameters such as molecular weight, polarity, solubility, osmotic pressure, pH value, charge, and affinity, providing the necessary input for component screening.

[0058] 3. A screening strategy combining component compatibility, acid-base stability, and osmotic pressure adaptability is proposed to effectively screen out the primary components closely related to exosome stability.

[0059] 4. The principal component analysis method is used to identify the key physical, chemical, and biological characteristic indicators affecting exosome stability, laying a foundation for the construction of subsequent prediction models.

[0060] 5. A two-layer machine learning model integrating data-driven and physical modeling is constructed, which can comprehensively predict various performance indicators of exosomes during the freeze-drying process, such as particle size change, membrane structure stability, and biological activity retention.

[0061] 6. Based on the stability prediction results, the genetic algorithm is used to realize the intelligent optimization of the protective agent ratio, and the best protective solution formula meeting the requirements of multiple stability indicators is obtained.

[0062] Through the above innovative method, the present invention can systematically analyze the key factors affecting the freeze-drying stability of exosomes and quickly screen out the optimal cryoprotectant formulation that meets the performance requirements in multiple aspects. Compared with the prior art, this method significantly improves the freeze-drying protection performance of exosomes, solves the technical problems that the formulation optimization of existing exosome freeze-drying cryoprotectants mostly relies on empirical trial and error, has low efficiency, and is difficult to systematically analyze the influencing factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a flowchart of the method provided by the present invention;

[0064] Figure 2 is the principal component analysis result of the exosome cryoprotectant formulation in the embodiment;

[0065] Figure 3 is a comparison chart of particle size distributions in the embodiment;

[0066] Figure 4 is a radar chart of stability indicators in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0068] As Figure 1 shown, it is a flowchart of a method for optimizing the components of a cryoprotectant for freeze-drying exosome positive control products provided by the present invention. This method includes the following steps:

[0069] S10. Establish a candidate component library for the exosome freeze-drying cryoprotectant, where the candidate component library includes buffer components, cryoprotectant components, and additive components;

[0070] S20. Collect the physical and chemical property data of all components in the candidate component library, where the physical and chemical property data includes molecular weight data, polarity data, solubility data, osmotic pressure data, pH value data, charge data, and affinity data;

[0071] S30. Screen the components in the candidate component library based on component compatibility parameters, acid-base stability parameters, and osmotic pressure adaptability parameters to obtain a primary selected component data set;

[0072] S40. Extract the characteristics of the components in the primary selected component data set, and use the principal component analysis method to screen out the key characteristic indicators affecting the stability of exosomes. The key characteristic indicators include physical characteristic indicators, chemical characteristic indicators, and biological characteristic indicators;

[0073] S50. Collect the exosome stability evaluation data and establish a training data set. The exosome stability evaluation data includes particle size distribution data, surface charge data, membrane integrity data, and biological activity data.

[0074] S60. Construct an exosome stability prediction model using a double-layer machine learning algorithm. The exosome stability prediction model includes a convolutional neural network model and a recurrent neural network model. Input the key feature indicators into the convolutional neural network model and the recurrent neural network model, fuse the outputs of the two models through the fusion pooling layer, and obtain the exosome stability prediction result after processing by the fully connected layer.

[0075] S70. Use the training data set to train the convolutional neural network model and the recurrent neural network model respectively, and adopt the cross-validation method to optimize the parameters of the exosome stability prediction model.

[0076] S80. Set the exosome stability optimization objective function and the component ratio constraint conditions based on the prediction result of the exosome stability prediction model, and use the genetic algorithm to perform multi-generation iterative optimization on the component types and ratios in the preliminary selected component data set.

[0077] S90. Prepare the exosome lyophilization protective solution according to the optimal component formulation output by the genetic algorithm, optimize and update the exosome stability prediction model through the lyophilization protection performance verification experiment, and obtain the final exosome lyophilization protective solution formulation.

[0078] Among them, the buffer component is composed of phosphate buffer, citrate buffer, and 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid buffer. The protective agent component is composed of sucrose, mannitol, trehalose, and dextran. The additive component is composed of amino acid substances, protein substances, and inorganic salt substances.

[0079] The following is a detailed description of the specific implementation manners of the above steps:

[0080] The specific implementation manner of step S10 is as follows: First, establish a candidate component library for the exosome lyophilization protective solution. This candidate component library contains three major types of components, namely buffer components, protective agent components, and additive components. The buffer component is composed of 3 common buffers, including phosphate buffer, citrate buffer, and 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid buffer. The protective agent component consists of 4 representative protective agents, including sucrose, mannitol, trehalose, and dextran. The additive component is composed of amino acid substances, protein substances, and inorganic salt substances. By establishing such a comprehensive candidate component library, it provides rich choices for subsequent screening and optimization.

[0081] The specific implementation of step S20 is as follows: Collect the physicochemical property data of all components in the candidate component library. These physicochemical property data include parameters such as molecular weight, polarity, solubility, osmotic pressure, pH value, charge, and affinity. Through the collection and collation of these basic data, it provides the necessary input data for subsequent component screening and the establishment of a stability prediction model.

[0082] The specific implementation of step S30 is as follows: Screen the components in the candidate component library according to the component compatibility parameter, acid-base stability parameter, and osmotic pressure adaptability parameter, so as to obtain a preliminary selected component data set. The component compatibility parameter is used to evaluate the chemical interaction between components, the acid-base stability parameter evaluates the stability of components under different pH conditions, and the osmotic pressure adaptability parameter reflects the tolerance ability of components under different osmotic pressures. By comprehensively considering these three types of parameters, the preliminary selected components that are most critical to exosome stability can be screened out.

[0083] The specific implementation of step S40 is as follows: Extract the features of the components in the preliminary selected component data set, and use the principal component analysis method to screen out the key feature indicators that affect exosome stability. These key feature indicators include physical feature indicators, chemical feature indicators, and biological feature indicators. The physical feature indicators mainly reflect the physical properties of components such as molecular weight, polarity, and solubility; the chemical feature indicators reflect the chemical properties of components such as acidity and alkalinity, and charge characteristics; the biological feature indicators describe the impact of components on biological activity. Through the feature dimensionality reduction and screening of the principal component analysis method, the key feature indicators that affect exosome stability can be determined.

[0084] The specific implementation of step S50 is as follows: Collect exosome stability evaluation data and establish a training data set. These stability evaluation data include particle size distribution data, surface charge data, membrane integrity data, and biological activity data. Through the experimental determination of these stability-related indicators, the actual evaluation data of exosome stability can be obtained, providing the necessary sample data for subsequent prediction model training.

[0085] The specific implementation of step S60 is as follows: Use a two-layer machine learning algorithm to construct an exosome stability prediction model. This prediction model includes two levels: a convolutional neural network model and a recurrent neural network model. First, input the aforementioned key feature indicators into the convolutional neural network model and the recurrent neural network model for feature extraction and classification. Then, fuse the outputs of the two models through a fusion pooling layer, and after processing by the fully connected layer, the prediction result of exosome stability can be obtained. This two-layer model structure can give full play to the respective advantages of data-driven models and physical models, improving the accuracy and interpretability of prediction.

[0086] The specific implementation of step S70 is as follows: The convolutional neural network model and the recurrent neural network model are trained respectively using the training dataset, and the cross-validation method is adopted to optimize the parameters of the entire exosome stability prediction model. The cross-validation method can effectively avoid the occurrence of overfitting phenomena and ensure that the model also has good prediction performance on new data. By iteratively optimizing the parameters, the fitting degree and generalization ability of the prediction model can be continuously improved.

[0087] The specific implementation of step S80 is as follows: Based on the prediction results of the exosome stability prediction model, an optimization objective function and component ratio constraint conditions are set, and the genetic algorithm is used to perform multi-generation iterative optimization on the component types and ratios in the preliminary selected component dataset. The optimization objective function includes a weighted combination of indicators such as particle size stability score, membrane stability score, activity retention score, and osmotic pressure adaptability score. The component ratio constraint conditions ensure that the mass fractions of each component are within a reasonable range. Through the iterative optimization of the genetic algorithm, the optimal component ratio that meets the stability requirements can be found.

[0088] The specific implementation of step S90 is as follows: An exosome lyophilization protective solution is prepared according to the optimal component formulation output by the genetic algorithm, and the exosome stability prediction model is optimized and updated through a lyophilization protection performance verification experiment. Through experimental verification, the accuracy of the prediction model can be evaluated, and the model parameters can be corrected and optimized according to the experimental results, and finally a reliable exosome lyophilization protective solution formulation can be obtained.

[0089] In the whole method, the specific implementation of the double-layer machine learning model is as follows:

[0090] The first-layer model adopts a convolutional neural network structure, including an input layer, two convolutional layers, two activation layers, and a pooling layer. This model mainly extracts large-scale prediction results such as component ratios, stability indicators, and protection effects in a data-driven manner.

[0091] The second-layer model adopts a recurrent neural network structure, including an input layer, an embedded mathematical model, a long short-term memory unit layer, and a pooling layer. This model uses an accurate mathematical model to generate small-scale high-precision outputs such as osmotic pressure calculation results, membrane stability prediction values, and activity retention evaluation values.

[0092] The outputs of the pooling layers of the two models are subjected to feature fusion through a fusion pooling layer, and then connected to the fully connected layer and the output layer to complete the comprehensive prediction of exosome stability.

[0093] The embedded mathematical model contains 4 key equations:

[0094] is the osmotic pressure equation, used to calculate the osmotic pressure of the cryoprotectant solution. Where, i is the degree of dissociation, M is the solute concentration, v is the ionic strength, A, B, and C are virial coefficients, V j is the molar volume of the non-electrolyte component, φ j is the volume fraction of the non-electrolyte component.

[0095] is the membrane stability equation, used to evaluate the stability of the membrane structure. Where, k1, k2, and k3 are rate constants, E a is the apparent activation energy, c i is the concentration of the i-th component, β i is the component influence coefficient, γ is the surface tension, and ζ is the Zeta potential.

[0096] is the particle size distribution equation, used to predict the change trend of particle size. Where, d is the particle size, μ is the average particle size, σ is the standard deviation, H k is the Hermite polynomial, a k is the expansion coefficient.

[0097] is the bioactivity retention equation, used to calculate the retention of biological activity. Where, A is the real-time activity, A0 is the initial activity, k d is the inactivation rate constant, θ is the surface coverage, K m is the cryoprotectant binding constant, c p is the cryoprotectant concentration, η is the synergistic coefficient, w l and φ l are the weight and protection efficiency of the additive, respectively.

[0098] The parameters of these mathematical models are obtained through different experimental measurement methods, such as the conductivity method to measure the degree of dissociation, the pycnometer method to measure the molar volume, and kinetic experiments to measure the rate constants, etc. Through these refined mathematical models, the influence of various physical and chemical factors on the stability of exosomes can be fully considered.

[0099] Combining the above steps, this optimization method for exosome lyophilization cryoprotectant makes full use of the idea of combining data-driven and physical models. By establishing a comprehensive component library, extracting key characteristic indicators, constructing a stability prediction model, and performing genetic algorithm optimization, etc., the optimized exosome lyophilization cryoprotectant formula is finally obtained. This method can not only quickly screen out suitable cryoprotectant components, but also deeply analyze the influencing factors, providing scientific theoretical support for the lyophilization stability of exosomes.

[0100] Specifically, the principle of the present invention is as follows: It combines two technical ideas of mathematical modeling and machine learning. On the one hand, a series of refined mathematical models are established, such as the osmotic pressure equation, membrane stability equation, particle size distribution equation, and activity retention equation, which comprehensively describe the key physicochemical factors affecting the stability of exosomes. On the other hand, a two-layer machine learning algorithm is adopted, including a data-driven convolutional neural network model and a recurrent neural network model embedded with physical models, which can make full use of experimental data and theoretical knowledge to obtain accurate prediction results of exosome stability. Through the organic combination of this mathematical modeling and machine learning, a comprehensive analysis and accurate prediction of the factors affecting exosome stability are achieved.

[0101] Specifically, first, a rich candidate component library containing buffers, cryoprotectants, and additives was established, and detailed physicochemical property data of these components were collected. Then, a screening strategy based on component compatibility, acid-base stability, and osmotic pressure adaptability was proposed to effectively screen out the primary components closely related to exosome stability. Next, the principal component analysis method was used to identify the key characteristic indicators affecting exosome stability, including physical, chemical, and biological characteristics.

[0102] Based on the above analysis results, a two-layer machine learning model integrating data-driven and physical modeling was constructed. The first-layer convolutional neural network model uses the key characteristic indicators as inputs and outputs large-scale component ratio prediction results, stability evaluation values, and protection effect prediction values. The second-layer recurrent neural network model embeds refined mathematical models, such as the osmotic pressure equation, membrane stability equation, etc., to generate small-scale high-precision outputs, such as osmotic pressure calculation results, membrane stability prediction values, and activity retention evaluation values. The outputs of the two models are fused through a fusion pooling layer, and finally, a prediction score of the comprehensive stability of exosomes is obtained.

[0103] With this stability prediction model as the basis, the genetic algorithm is further used to intelligently optimize the primary components. First, an objective function for the comprehensive stability of exosomes is set, covering multiple indicators such as particle size stability, membrane stability, activity retention, and osmotic pressure adaptability. At the same time, reasonable component ratio constraint conditions are also set. On this basis, the genetic algorithm searches for the optimal cryoprotectant ratio that meets the stability requirements through multiple generations of iteration.

[0104] In summary, the key to the method for optimizing the components of the exosome freeze-drying protective solution of the present invention lies in fully integrating the technical advantages of mathematical modeling and machine learning. On the one hand, the refined physicochemical model can deeply analyze the key factors affecting exosome stability; on the other hand, the powerful machine learning algorithm can effectively integrate experimental data and theoretical knowledge to obtain accurate stability predictions. Based on this, the intelligent optimization of the cryoprotectant ratio is realized, and the comprehensive stability of exosomes during the freeze-drying process is greatly improved.

[0105] A specific Embodiment 1 of the present invention is provided below. The specific implementation of each step in this Embodiment 1 is described in detail as follows: In step S10, a candidate component library of exosome lyophilization protective solution containing buffer components, cryoprotectant components, and additive components was first established. Among them, the buffer components consist of 3 common buffers {phosphate buffer, citrate buffer, 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid buffer}; the cryoprotectant components are composed of 4 representative cryoprotectants {sucrose, mannitol, trehalose, dextran}; and the additive components are composed of amino acid substances, protein substances, and inorganic salt substances. By establishing such a comprehensive candidate component library, a wide range of choices are provided for subsequent screening and optimization.

[0106] In step S20, the physicochemical property data of all components in the candidate component library were collected, including molecular weight M, polarity P, solubility S, osmotic pressure π, pH value pH, charge q, and affinity K. d . These data can be expressed by the following formulas:

[0107]

[0108]

[0109] pH = -log[H + ;

[0110]

[0111] where m is the solute mass, V is the solution volume, q i is the charge of the i-th atom, r i is the position vector of the i-th atom, m s is the solute mass, V s is the solvent volume, i is the degree of dissociation, V is the ionic strength, A, B, C are virial coefficients, V j is the molar volume of the j-th nonelectrolyte component, φ j is the volume fraction of the j-th nonelectrolyte component, [L] is the ligand concentration, [R] is the receptor concentration, and [LR] is the ligand-receptor complex concentration. These physicochemical property data provide the necessary inputs for subsequent component screening and the establishment of stability prediction models.

[0112] In step S30, the candidate components were screened according to the component compatibility parameter P c , acid-base stability parameter S a , and osmotic pressure adaptability parameter Π a to obtain a preliminary selected component data set. Among them:

[0113]

[0114] In the above formula, P i and P j are the polarities of the i-th and j-th components respectively, c i is the concentration of the i-th component, k1, E a and β i are the rate constant, activation energy and influence coefficient, π is the osmotic pressure of the protective liquid, and C0 is the reference concentration. By comprehensively considering these three types of parameters, the primary selected components that are most critical to the stability of exosomes can be screened out.

[0115] In step S40, feature extraction and screening are performed on the components in the primary selected component dataset. First, the physical feature index F p , chemical feature index F c and biological feature index F b are extracted, where:

[0116] F p ={M, P, S};

[0117] F c ={pH, q};

[0118] F b ={K d};

[0119] Then, the principal component analysis method is used to reduce the dimension of these feature indexes to obtain the key feature index F k for the stability of exosomes:

[0120] F k ={F1, F2,..., F l};

[0121] Among them, F1, F2,..., F l are the principal component scores. Through this step, the most critical stability influencing factors are determined.

[0122] In step S50, the particle size distribution f(d), surface charge ζ, membrane integrity S m and biological activity A and other stability evaluation data of exosomes are collected to establish a training dataset. Among them:

[0123]

[0124] The experimental measurement data of these stability-related indexes provide necessary samples for the training of the subsequent prediction model.

[0125] In step S60, a two-layer machine learning algorithm including a convolutional neural network model and a recurrent neural network model is constructed. The first-layer convolutional neural network model M cLarge-scale prediction results such as extraction group allocation ratios, stability indicators, and protection effects can be expressed as:

[0126]

[0127] The second-layer recurrent neural network model M r Using an accurate mathematical model to generate small-scale, high-precision outputs such as osmotic pressure, membrane stability, and activity retention can be expressed as:

[0128]

[0129] The outputs of the two models are fused through the fusion pooling layer F to obtain the comprehensive prediction result S of exosome stability total :

[0130] S total = Φ(F(M c ,M r ));

[0131] This double-layer structure gives full play to the respective advantages of data-driven models and physical models.

[0132] In step S70, the convolutional neural network model and the recurrent neural network model are trained using the training dataset D train and the parameters of the entire prediction model are optimized using the cross-validation method. Among them, the objective function of cross-validation is:

[0133]

[0134] In the formula, L i is the loss function of the i-th fold validation set, is the i-th fold validation set. By minimizing L, it can be ensured that the model also has good prediction performance on new data.

[0135] In step S80, first, the optimization objective function S of exosome stability is set total :

[0136] S total = α1S size + α2S membrane + α3S activity + α4S osmotic ;

[0137] Among them, S size is the particle size stability score, S membrane is the membrane stability score, S activity is the activity retention score, S osmoticis the osmotic pressure adaptability score, and α1, α2, α3, and α4 are weight coefficients, satisfying α1 + α2 + α3 + α4 = 1. At the same time, the constraint conditions for the component ratio are set:

[0138]

[0139] l i ≤ x i ≤ u i , i = 1, 2,..., n;

[0140] In the formula, x i is the mass fraction of the i-th component, l i and u i are its minimum and maximum mass fractions respectively, and n is the total number of components. On this basis, the genetic algorithm is used to perform multi-generation iterative optimization on the initially selected components to find the optimal ratio that meets the stability requirements; among them, the chromosome encoding in the genetic algorithm:

[0141] C k = [g1, g2,..., g m ;

[0142] In the formula, C k is the k-th chromosome; g j is the j-th gene position, representing the component ratio; m is the number of genes.

[0143] The fitness function in the genetic algorithm:

[0144]

[0145] In the formula, F(C k ) is the fitness value of chromosome C k ; λ is the penalty factor; P i (x i ) is the penalty function of the i-th component.

[0146] In step S90, the exosome freeze-drying protective solution was prepared according to the output of the genetic algorithm, and the prediction model was optimized and updated through the freeze-drying protection performance verification experiment. Specifically, indexes such as particle size distribution, surface charge, membrane integrity, and biological activity of the protective solution during the actual freeze-drying process were measured, and the experimental data were compared with the prediction results, and finally a reliable exosome freeze-drying protective solution formula was obtained.

[0147] To further better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: A certain biomedical enterprise developed an exosome positive control product for drug efficacy evaluation and quality detection. To ensure the stability of the exosome positive control product during freeze-drying preservation, the enterprise decided to develop a protective liquid formulation using the exosome freeze-drying protective liquid component optimization method proposed by the present invention.

[0148] First, the enterprise established a candidate component library for the exosome freeze-drying protective liquid. Based on past experience and relevant literature, they selected 3 buffer components (phosphate buffer, citrate buffer, and 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid buffer), 4 protective agent components (sucrose, mannitol, trehalose, and dextran), and 3 additive components (amino acids, proteins, and inorganic salts). Through the combination of different components, a protective liquid library containing 100 candidate formulations was initially constructed.

[0149] Subsequently, the enterprise collected detailed physicochemical property data of each component in these 100 candidate formulations. Specifically included:

[0150] 1. Molecular weight M: Obtained by weighing and calculation. Taking sucrose as an example, M = 342.30 g / mol.

[0151] 2. Polarity P: Measured by the polar surface area method. Taking sucrose as an example, P = 397.85 Ų 2 .

[0152] 3. Solubility S: Measured by the saturation solubility method. Taking sucrose as an example, S = 1.67 mol / L.

[0153] 4. Osmotic pressure π: Measured by an osmometer. Taking sucrose as an example, π = 2.42 MPa.

[0154] 5. Acid-base value pH: Measured by a pH meter. Taking phosphate buffer as an example, pH = 7.40.

[0155] 6. Charge q: Measured by a Zeta potential analyzer. Taking sucrose as an example, q = -3.21 mC / m 2 .

[0156] 7. Affinity K d : Measured by isothermal titration calorimetry. Taking the sucrose-exosome interaction as an example, K d = 1.85×10 -6 mol / L.

[0157] With these detailed physicochemical property data, the enterprise preliminarily screened the candidate components according to the screening strategy in step S30 of the present invention. First, according to the component compatibility parameter P c, formulations showing strong chemical interactions were excluded. Secondly, based on the acid-base stability parameter S a , formulations unstable under specific pH conditions were excluded. Finally, based on the osmotic pressure adaptability parameter Π a , formulations with unsuitable osmotic pressure were removed. After this series of screenings, the enterprise finally obtained 20 initially selected preservation fluid formulations.

[0158] In step S40, the enterprise performed feature extraction and principal component analysis on these 20 initially selected formulations. According to the method of the present invention, they extracted the physical feature indicators F p = {M, P, S}, chemical feature indicators F c = {pH, q}, and biological feature indicators F b = {K d}. Through principal component analysis, the enterprise determined 3 key feature indicators F k = {F1, F2, F3} that affect the stability of exosomes, which explained 65%, 21%, and 9% of the total variance respectively. As Figure 2 shown, it presents the distribution of the 20 initially selected formulations in the three principal component spaces. The color in the scatter plot represents the value of the third principal component, and each point is labeled with the formulation number. This figure intuitively shows the similarities and differences between different formulations.

[0159] It is recommended to add a description of this figure after describing the results of the principal component analysis in step S40 of the embodiment.

[0160] With the initially selected formulations and key feature indicators, the enterprise began to establish a prediction model for exosome stability. They adopted the double-layer machine learning algorithm proposed by the present invention to construct a prediction framework containing a convolutional neural network model and a recurrent neural network model. Specifically, the first-layer convolutional neural network model took the key feature indicators F k as inputs and output the following 3 prediction results:

[0161] 1. Prediction result of particle size distribution

[0162] 2. Prediction result of membrane stability

[0163] 3. Prediction result of activity retention

[0164] The second-layer recurrent neural network model then embedded a refined mathematical model and output the following 3 high-precision predicted values:

[0165] 1. Calculation result of osmotic pressure

[0166] 2. Membrane stability index

[0167] 3. Activity retention rate

[0168] The outputs of the two models are integrated into the comprehensive stability prediction score of exosomes through a fusion pooling layer

[0169] To train this two - layer prediction model, the enterprise collected the stability evaluation data of 20 primary - selected formulations in actual freeze - drying experiments, including:

[0170] 1. Particle size distribution data {d i , f i};

[0171] 2. Surface charge data {ζ j};

[0172] 3. Membrane integrity data {S m,k};

[0173] 4. Biological activity data {A l}.

[0174] The cross - validation method was used to optimize the training of the model parameters, and finally an exosome stability prediction model with a prediction accuracy of 85% was obtained

[0175] Based on this prediction model, the enterprise further used the genetic algorithm to intelligently optimize the cryoprotectant ratio. First, they set the objective function S of the comprehensive stability of exosomes total , covering four indicators: particle size stability S size , membrane stability S membrane , activity retention S activity and osmotic pressure adaptability S osmotic :

[0176] S total = 0.4S size + 0.3S membrane + 0.2S activity + 0.1S osmotic ;

[0177] At the same time, the enterprise also set the constraint conditions for the component ratio to ensure that the mass fractions of each component are within a reasonable range:

[0178]

[0179] 0.05 ≤ x1, x2, x3 ≤ 0.30;

[0180] 0.10 ≤ x4, x5, x6 ≤ 0.40;

[0181] 0.01 ≤ x7, x8, x9, x 10 , x 11 ≤ 0.15;

[0182] Among them, x1 to x3 respectively represent the mass fractions of 3 buffer components, x4 to x6 respectively represent the mass fractions of 4 protective agent components, and x7 to x 11 respectively represent the mass fractions of 5 additive components.

[0183] On this basis, the enterprise optimized and iterated the initially selected formula 20 times using the genetic algorithm. By continuously updating the population, the genetic algorithm finally output an optimal formula with a comprehensive stability score S total = 0.92, and the mass fractions of each component are shown in Table 1:

[0184] Table 1 Optimal cryoprotectant formula

[0185]

[0186]

[0187] With this optimal formula, the enterprise prepared an exosome lyophilization cryoprotectant and verified its actual lyophilization protection performance. The experimental results show that during the lyophilization process, the cryoprotectant can effectively maintain the stable particle size distribution of exosomes, with an average particle size of 110 ± 15 nm and a unimodal particle size distribution curve. At the same time, the membrane integrity of exosomes is also well protected, and the membrane stability index S m is maintained above 0.85. In addition, the biological activity retention rate reaches more than 90%, indicating that the functional characteristics of exosomes are not significantly affected.

[0188] By comparing with existing lyophilization protection technologies, the cryoprotectant prepared by the method of the present invention shows significant advantages in the lyophilization stability of exosomes. For example, under the same conditions, an ordinary sucrose cryoprotectant can only maintain a particle size of 150 ± 30 nm, a membrane stability index of 0.65, and a biological activity retention rate of about 80%. It can be seen that the method of the present invention significantly improves the comprehensive stability during the lyophilization of exosomes by systematically analyzing the influencing factors and intelligently optimizing the formula.

[0189] As Figure 3 shown, the effects of the optimized cryoprotectant formula and the ordinary sucrose cryoprotectant on the particle size distribution of exosomes are compared. The figure clearly shows that the optimized formula has a narrower particle size distribution range and a more ideal average particle size.

[0190] It is recommended to add a description of this figure after mentioning the particle size distribution data when describing the verification results in the last part of the embodiment.

[0191] As Figure 4As shown, a radar chart is used to display the performance of the optimized protective liquid formulation and the ordinary sucrose protective liquid in four key stability indicators. The chart includes the indicator names with superscripts, visually showing the comprehensive advantages of the optimized formulation.

[0192] It is recommended to add a description of this figure when describing the objective function of the formulation optimization in the embodiments.

[0193] Generally speaking, this embodiment fully embodies the innovation and practicality of the method for optimizing the components of the exosome freeze-drying protective liquid of the present invention. The enterprise has not only established a comprehensive candidate component library and collected detailed physical and chemical property data, but also adopted a combination of mathematical modeling and machine learning to deeply analyze the key factors affecting exosome stability, and finally obtained an excellent protective liquid formulation through intelligent optimization.

[0194] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. An optimization method for the components of a protective solution for freeze-drying exosome positive control products, characterized in that, Including: Establish a library of candidate components for exosome lyophilization protectant; Collect the physicochemical property data of all components in the candidate component library; Screen the components in the candidate component library based on component compatibility parameters, acid-base stability parameters, and osmotic pressure adaptability parameters to obtain a preliminary selected component dataset; Extract the features of the components in the preliminary selected component dataset, and use the principal component analysis method to screen out the key feature indicators affecting exosome stability; Collect exosome stability evaluation data to establish a training dataset; use a two-layer machine learning algorithm to construct an exosome stability prediction model; use the training dataset to train and optimize the parameters of the exosome stability prediction model; set an optimization objective function and constraint conditions based on the prediction results of the exosome stability prediction model, and use the genetic algorithm to perform multi-generation iterative optimization on the component types and ratios in the preliminary selected component dataset; prepare an exosome lyophilization protectant according to the optimal component formula output by the genetic algorithm, and optimize and update the exosome stability prediction model through a freeze-drying protection performance verification experiment to obtain the final exosome lyophilization protectant formula.

2. The optimization method of the protective liquid components for freeze-drying of exosome positive control products according to claim 1, wherein, The candidate component library includes buffer components, protectant components, and additive components; the buffer components are composed of phosphate buffer, citrate buffer, and 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid buffer, the protectant components are composed of sucrose, mannitol, trehalose, and dextran, and the additive components are composed of amino acid substances, protein substances, and inorganic salt substances.

3. The method for optimizing the protective liquid components for freeze-drying of exosome positive control products according to claim 2, characterized in that, The physicochemical property data includes molecular weight data, polarity data, solubility data, osmotic pressure data, pH value data, charge data, and affinity data.

4. A method for optimizing the composition of a protective solution for freeze-drying of exosome positive control products according to claim 3, characterized in that The key feature indicators include physical feature indicators, chemical feature indicators, and biological feature indicators.

5. The optimization method of the protective liquid components for freeze-drying of exosome positive control products according to claim 4, characterized in that The exosome stability evaluation data includes particle size distribution data, surface charge data, membrane integrity data, and biological activity data.

6. The optimization method of the protective liquid components for freeze-drying of exosome positive control products according to claim 5, characterized in that, The exosome stability prediction model includes a first-layer model and a second-layer model; the first-layer model uses a convolutional neural network model, including an input layer, two convolutional layers, two activation layers, and a pooling layer; the second-layer model uses a recurrent neural network structure, including an input layer, an embedded mathematical model, a long short-term memory unit layer, and a pooling layer.

7. A method for optimizing the composition of a protective solution for freeze-drying of exosome positive control products according to claim 6, characterized in that, A fusion pooling layer is connected after the pooling layer of the first-layer model and the pooling layer of the second-layer model to fuse the outputs of the two pooling layers into a feature vector; a fully connected layer and an output layer are connected after the fusion pooling layer, the fully connected layer is used for feature dimensionality reduction and non-linear mapping, and the output layer is used to generate an exosome stability prediction score.

8. A method for optimizing the composition of a protective solution for freeze-drying of exosome positive control products according to claim 7, characterized in that, The embedded mathematical model is a system of mathematical equations, including an osmotic pressure equation, a membrane stability equation, a particle size distribution equation, and an activity retention equation; the first-layer model uses a data-driven method to generate large-scale component ratio prediction results, stability evaluation values, and protection effect prediction values; the second-layer model uses an accurate mathematical model to generate small-scale osmotic pressure calculation results, membrane stability prediction values, and activity retention evaluation values.

9. The method for optimizing the composition of the protective solution for freeze-drying of exosome positive control products according to claim 8, wherein The osmotic pressure equation is used to calculate the osmotic pressure of the protective solution. The inputs are component concentration and temperature parameters, and the output is the osmotic pressure value. The membrane stability equation is used to evaluate the stability of the membrane structure. The inputs are ionic strength and pH value parameters, and the output is the membrane stability index.

10. The method for optimizing the protective liquid components for freeze-drying of exosome positive control products according to claim 9, characterized in that, The particle size distribution equation is used to predict the trend of particle size change. The inputs are the molecular weight and concentration parameters of the protective agent, and the output is the particle size distribution curve. The activity retention equation is used to calculate the biological activity. The inputs are the type of protective agent and the additive ratio parameters, and the output is the activity retention rate.

Images