Cell culture medium mixed formula generation method and device and electronic equipment

Abstract

The invention provides a cell culture medium mixed formula generation method and device and electronic equipment. The method comprises the following steps: screening according to component difference entropy of each component in a historical prototype formula; taking the weight of the target prototype formula as a variable, and adopting Latin hypercube sampling to generate a plurality of mixed formulas in the high-dimensional weight space to form a candidate formula set; adaptive robust normalization is executed according to the median and quartile distance of the concentration of each component in the mixed formula, a standardized formula set is constructed, and a screened formula set is formed through screening of a greedy algorithm and a clustering algorithm; a preset evaluation index corresponding to the screening formula set is calculated, a relation model between the sample size and the coverage degree is constructed, the minimum sample size required by the target coverage degree is obtained through Gaussian process regression prediction, and a target mixed formula set corresponding to the minimum sample size is selected from the screening formula set. High-dimensional mixed space and high sample screening efficiency can be achieved, the intelligent level of cell culture medium formula design is improved, and the development cycle and cost are reduced.

Description

Technical Field

[0001] The present disclosure relates to the field of biopharmaceutical technology, and in particular to a method, device, and electronic device for generating a mixed formula for a cell culture medium. Background Art

[0002] Cell culture media are the basic substances that enable cells to grow, divide, and express target products in vitro. They are widely used in key areas such as biopharmaceuticals, vaccine preparation, stem cell therapy, and synthetic biology. The quality of the culture medium formulation directly affects the metabolic efficiency of cells, the level of product expression, and the economic efficiency of production. In practical applications, a complete culture medium formulation may contain up to hundreds of ingredients, such as amino acids, inorganic salts, vitamins, sugars, growth factors, and metal ions. The interactions between these ingredients exhibit significant nonlinear, synergistic, or antagonistic effects, making it difficult for traditional experience-driven design methods to cope with the optimal combination search in high-dimensional formula spaces.

[0003] Currently, the development of cell culture media is still mainly based on "trial and error experiments" or methods such as orthogonal design and response surface analysis. Traditional methods rely on expert experience or fixed templates to select prototype formulas for combination, making it difficult to systematically evaluate the differences and complementarities between prototypes, resulting in insufficient coverage of the mixing space and weak sample representativeness. In high-dimensional mixing ratio space, random sampling tends to cluster in local areas, and the sample size required for orthogonal design increases exponentially, resulting in high experimental costs. Existing methods are sensitive to outliers and order of magnitude differences when normalizing component concentrations, which can easily lead to sample distortion and affect the quality of subsequent screening. Summary of the Invention

[0004] The embodiments of the present disclosure provide at least one method, device, and electronic device for generating a cell culture medium mixing formula, which can achieve high-dimensional mixing space and high sample screening efficiency, improve the intelligence level of cell culture medium formula design, reduce development cycle and cost, and improve industrial application performance.

[0005] The present disclosure provides a method for generating a mixed formula of a cell culture medium, comprising:

[0006] Acquire multiple historical prototype formulas, calculate the component difference entropy of each component in all the historical prototype formulas, and select a target prototype formula whose component difference entropy is greater than a preset threshold;

[0007] Using the weight of the target prototype recipe as a variable, Latin hypercube sampling is used to generate multiple mixed recipes in a high-dimensional weight space to form a candidate recipe set;

[0008] For each of the mixed formulas in the candidate formula set, adaptive robust normalization is performed according to the median and interquartile range of the concentration of each component in the mixed formula to construct a standardized formula set and screen it through a greedy algorithm and a clustering algorithm to form a screened formula set;

[0009] Calculate the preset evaluation indicators corresponding to the screening formula set, construct a relationship model between sample size and coverage based on the preset evaluation indicators and the number of formulas, predict the minimum sample size required to achieve the target coverage through Gaussian process regression, and select the target mixed formula set corresponding to the minimum sample size from the screening formula set.

[0010] In an optional embodiment, calculating the component difference entropy of each component in all the historical prototype formulas, and screening the target prototype formula whose component difference entropy is greater than a preset threshold, specifically includes:

[0011] Determine a standardized value of the coefficient of variation corresponding to each ingredient in the historical prototype formula, and calculate the ingredient difference entropy corresponding to the ingredient based on the standardized value of the coefficient of variation, wherein the ingredient difference entropy is used to measure the degree of diversity between the historical prototype formulas;

[0012] The prototype formula having the component difference entropy greater than the preset threshold is screened, and the concentration range corresponding to each component in the target prototype formula is expanded by a preset ratio to generate the target prototype formula.

[0013] In an optional embodiment, Latin hypercube sampling is used to generate multiple mixed formulas in a high-dimensional weight space using the weight of the target prototype formula as a variable to form a candidate formula set, specifically including:

[0014] Dividing the weight range of each target prototype formula into a plurality of equal probability intervals whose number is equal to the sample size;

[0015] Randomly selecting a sample point for each of the equal probability intervals, and calculating a preset space filling index of the sample set consisting of the target prototype formula;

[0016] The sample positions are iteratively optimized according to the preset space filling index to generate a plurality of the mixed formulas, and the candidate formula set is formed by all the mixed formulas. Specifically, for the sample set with low difference, the Sobol sequence is used to generate the initial sample points, and the corresponding sample positions are adjusted by the simulated annealing algorithm.

[0017] In an optional embodiment, the standardized formula set is filtered by a greedy algorithm and a clustering algorithm to form the filtered formula set, which specifically includes:

[0018] Randomly selecting an initial standardized formula from the standardized formula set;

[0019] Starting from the initial standardized recipe, iteratively selecting the standardized recipe with the largest minimum Euclidean distance from the selected recipes in the standardized recipe set to form a selected recipe set, until the number of the standardized recipes in the selected recipe set reaches a preset sample number;

[0020] The local density and relative distance are calculated for the standardized recipes in the selected recipe set, the cluster center is determined according to a preset decision diagram, and the filtered recipes whose local density is greater than a preset density threshold and whose relative distance is greater than a preset distance threshold are screened to form the filtered recipe set.

[0021] In an optional embodiment, the preset evaluation index includes at least a diversity dimension and a component dimension;

[0022] For the diversity dimension, the preset evaluation indicators include at least the average pairwise Euclidean distance, the minimum pairwise Euclidean distance, the candidate pool coverage distance, and the spatial distribution width;

[0023] For the component dimension, the preset evaluation indicators include at least minimum concentration, maximum concentration, concentration fluctuation range and concentration fluctuation multiple.

[0024] In an optional embodiment, a relationship model between sample size and coverage is constructed based on the preset evaluation index and the number of formulations, and the minimum sample size required to achieve the target coverage is predicted by Gaussian process regression, specifically including:

[0025] Taking the preset evaluation index and the number of formulations corresponding to the screening formulations as input, constructing the relationship model between sample size and coverage;

[0026] The number of cost rows is defined according to the preset experimental resource consumption, and the minimum sample size required to achieve the target coverage is predicted through Gaussian process regression based on the relationship model. Experimental plans under different sample numbers are generated, and the Pareto optimal number of experimental groups that balances experimental resource consumption and sample size is output.

[0027] In an optional embodiment, after selecting the target mixed formula set corresponding to the minimum sample size from the screening formula set, the method further includes:

[0028] For the candidate recipes that are not selected in the candidate recipe set, calculating the corresponding outliers and performing uncertainty sampling to generate unsampled areas;

[0029] Constructing a random forest prediction model using the screened formula set as a training set, determining the predicted results of the component distribution diversity in the unsampled area, and generating a supplementary formula;

[0030] Adding the supplementary formula to the candidate formula set, and iterating the screening process based on the greedy algorithm and the clustering algorithm and the calculation process of the preset evaluation index;

[0031] If the average pairwise distance corresponding to the candidate formula increases by less than a preset first amplitude threshold in a consecutive preset number of iterations, and the coverage distance of the candidate formula set decreases by less than a preset second amplitude threshold, the iteration is terminated and the final target mixed formula set is output.

[0032] The present disclosure also provides a device for generating a cell culture medium mixed formula, comprising:

[0033] A prototype formula generation module is used to obtain multiple historical prototype formulas, calculate the component difference entropy of each ingredient in all the historical prototype formulas, and select a target prototype formula whose component difference entropy is greater than a preset threshold;

[0034] A candidate recipe set sampling module is used to generate multiple mixed recipes in a high-dimensional weight space using the weight of the target prototype recipe as a variable using Latin hypercube sampling to form a candidate recipe set;

[0035] A recipe screening module is used to perform adaptive robust normalization on each of the mixed recipes in the candidate recipe set according to the median and interquartile range of the concentration of each component in the mixed recipe, construct a standardized recipe set, and screen it using a greedy algorithm and a clustering algorithm to form a screened recipe set;

[0036] The experimental formula output module is used to calculate the preset evaluation indicators corresponding to the screening formula set, construct a relationship model between sample size and coverage based on the preset evaluation indicators and the number of formulas, predict the minimum sample size required to achieve the target coverage through Gaussian process regression, and select the target mixed formula set corresponding to the minimum sample size from the screening formula set.

[0037] An embodiment of the present disclosure further provides an electronic device, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the above-mentioned method for generating a mixed formula of a cell culture medium or steps in any possible implementation of the above-mentioned method for generating a mixed formula of a cell culture medium are performed.

[0038] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes the above-mentioned method for generating a mixed formula of a cell culture medium, or the steps of any possible embodiment of the above-mentioned method for generating a mixed formula of a cell culture medium.

[0039] The present disclosure also provides a computer program product, including a computer program / instructions. When the computer program or instructions are executed by a processor, the computer program or instructions implement the above-mentioned method for generating a mixed formula of a cell culture medium, or the steps of any possible embodiment of the above-mentioned method for generating a mixed formula of a cell culture medium.

[0040] The disclosed embodiments provide a method, device, and electronic device for generating a cell culture medium mixed formula. The method comprises obtaining multiple historical prototype formulas, calculating the component differential entropy of each component in all the historical prototype formulas, and screening a target prototype formula whose component differential entropy exceeds a preset threshold. Using the weights of the target prototype formulas as variables, Latin hypercube sampling is used to generate multiple mixed formulas in a high-dimensional weight space to form a candidate formula set. For each mixed formula in the candidate formula set, adaptive robust normalization is performed based on the median and interquartile range of the concentrations of each component in the mixed formula to construct a standardized formula set, which is then screened using a greedy algorithm and a clustering algorithm to form a screened formula set. Preset evaluation indicators corresponding to the screened formula set are calculated, and a relationship model between sample size and coverage is constructed based on the preset evaluation indicators and the number of formulas. The minimum sample size required to achieve target coverage is predicted using Gaussian process regression, and a target mixed formula set corresponding to the minimum sample size is selected from the screened formula set. This method can achieve a high-dimensional mixed space and high sample screening efficiency, improve the intelligent level of cell culture medium formula design, reduce development cycle and cost, and enhance industrial application performance.

[0041] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments. The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without inventive effort.

[0043] Figure 1 A flow chart of a method for generating a mixed formula of a cell culture medium provided by an embodiment of the present disclosure is shown;

[0044] Figure 2 A flow chart showing another method for generating a mixed formula of a cell culture medium provided by an embodiment of the present disclosure is shown;

[0045] Figure 3 A schematic diagram of a cell culture medium mixed formula generating device provided by an embodiment of the present disclosure is shown;

[0046] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. The components of the embodiments of the present disclosure generally 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 disclosure provided in the drawings is not intended to limit the scope of the disclosure for which protection is sought, but merely represents selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present disclosure.

[0048] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0049] The term "and / or" herein simply describes an association relationship, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, and the existence of B alone. In addition, the term "at least one" herein refers to any combination of at least two of any one or more of a plurality of items. For example, "at least one of A, B, and C" can represent any one or more elements selected from the set consisting of A, B, and C.

[0050] Research has found that, at present, the development of cell culture media is still mainly based on "trial and error experiments" or methods such as orthogonal design and response surface analysis. Traditional methods rely on expert experience or fixed templates to select prototype formulas for combination, which makes it difficult to systematically evaluate the differences and complementarities between prototypes, resulting in insufficient coverage of the mixing space and weak sample representativeness. In high-dimensional mixing ratio space, random sampling tends to cluster in local areas, and the sample size required for orthogonal design increases exponentially, resulting in high experimental costs. Existing methods are sensitive to outliers and order of magnitude differences when normalizing component concentrations, which can easily lead to sample distortion and affect the quality of subsequent screening.

[0051] Based on the above research, the present disclosure provides a method, device, and electronic device for generating cell culture medium mixed formulas. The method obtains multiple historical prototype formulas, calculates the component differential entropy of each component in all the historical prototype formulas, and selects a target prototype formula whose component differential entropy is greater than a preset threshold. Using the weight of the target prototype formula as a variable, Latin hypercube sampling is used to generate multiple mixed formulas in a high-dimensional weight space to form a candidate formula set. For each mixed formula in the candidate formula set, adaptive robust normalization is performed based on the median and interquartile range of the concentration of each component in the mixed formula to construct a standardized formula set, which is then screened using a greedy algorithm and a clustering algorithm to form a screened formula set. Preset evaluation indicators corresponding to the screened formula set are calculated, and a relationship model between sample size and coverage is constructed based on the preset evaluation indicators and the number of formulas. The minimum sample size required to achieve target coverage is predicted through Gaussian process regression, and a target mixed formula set corresponding to the minimum sample size is selected from the screened formula set. This method can achieve a high-dimensional mixed space and high sample screening efficiency, improve the intelligent level of cell culture medium formula design, reduce development cycle and cost, and enhance industrial application performance.

[0052] To facilitate understanding of this embodiment, a method for generating a cell culture medium mixing formula disclosed in an embodiment of the present disclosure is first described in detail. The method for generating a cell culture medium mixing formula provided in an embodiment of the present disclosure is generally executed by a computer device with certain computing capabilities. The computer device includes, for example, a terminal device, a server, or other processing device. The terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, an in-vehicle device, a wearable device, etc. In some possible implementations, the method for generating a cell culture medium mixing formula may be implemented by a processor invoking computer-readable instructions stored in a memory.

[0053] See also Figure 1 FIG. 1 is a flow chart of a method for generating a cell culture medium mixing formula according to an embodiment of the present disclosure, wherein the method includes steps S101 to S104, wherein:

[0054] S101 , obtaining multiple historical prototype formulas, calculating the component difference entropy of each component in all the historical prototype formulas, and screening a target prototype formula whose component difference entropy is greater than a preset threshold.

[0055] In the specific implementation, multiple historical prototype formulas are first processed and screened to construct a representative and diverse prototype formula set. Multiple historical prototype formulas are obtained, each containing several ingredients and their corresponding concentration values. These prototype formulas can be derived from public literature, commercial culture media products, or previous experimental records. Assuming the total number of historical prototype formulas is , and the total number of ingredients involved is , a prototype formula concentration matrix is constructed, where each matrix element represents the concentration value of a component in a particular formula.

[0056] Here, historical prototype formulas refer to existing, validated cell culture medium formulas. Sources can include literature databases, internal experimental records, commercial culture medium product instructions, etc. Each formula contains multiple components (such as amino acids, inorganic salts, vitamins, etc.) and their corresponding concentration values. From these historical prototype formulas, prototype formulas with sufficiently diverse ingredients, representativeness, and contribution in subsequent mixing are screened for use in constructing the mixing formula space.

[0057] For each component in the matrix, its concentration sequence across all historical prototype formulas was statistically analyzed. The mean and standard deviation of this sequence were calculated, and the coefficient of variation (CV) for that component was derived from this. This CV measures the fluctuation of that component across historical formulas. To minimize the impact of order of magnitude between different components, the CVs of all components were normalized to obtain a standardized CV value.

[0058] Furthermore, the concentration distribution of each component in each formula is discretized and divided into several concentration intervals. The frequency of occurrence of the component in each interval is counted, and the component difference entropy (CDE) of the component is calculated.

[0059] As a possible implementation method, the standardized value of the coefficient of variation corresponding to each ingredient in the historical prototype formula is determined, and the component difference entropy corresponding to the ingredient is calculated based on the standardized value of the coefficient of variation. The component difference entropy is used to measure the degree of diversity between historical prototype formulas; prototype formulas with component difference entropy greater than a preset threshold are screened, and the concentration range corresponding to each ingredient in the target prototype formula is expanded by a preset proportion to generate a target prototype formula.

[0060] Here, the component difference entropy can be calculated by the following formula:

[0061]

[0062] Where CDE stands for compositional difference entropy, p i It represents the standardized value of the coefficient of variation of the i-th component in the historical prototype formula. The greater the component difference entropy, the more significant the component difference between the prototype formulas.

[0063] Furthermore, based on the differential entropy of the ingredients contained in each prototype formula, the average differential entropy of the formula is calculated to assess its representativeness and diversity within the entire prototype formula set. By comparing it with a preset differential entropy threshold, all prototype formulas with an average differential entropy greater than the threshold are screened out to form the target prototype formula set.

[0064] Here, the extreme value extension method is used to expand the concentration range of each component by a preset ratio to ensure sufficient exploration of the mixing space. It should be noted that the preset ratio of the boundary expansion can be set according to actual needs and is not specifically limited here. An optional range is ±20%.

[0065] In actual implementation, the preset threshold for screening the target prototype formula can be set to 0.4. This value can be set according to actual needs and is not specifically restricted here. The target prototype formula finally screened out has high differences in component composition and concentration distribution, which can effectively improve the coverage and sampling efficiency of the subsequent mixing space, and provide a diversified basis for the subsequent mixing formula design and screening.

[0066] S102 , using the weight of the target prototype recipe as a variable and adopting Latin hypercube sampling to generate multiple mixed recipes in a high-dimensional weight space to form a candidate recipe set.

[0067] In this step, a hybrid formula construction operation is performed based on the multiple target prototype formulas obtained through screening to construct a representative candidate formula set. The target prototype formula set includes multiple target prototype formulas, each of which contains multiple known ingredients and their concentrations. Each prototype formula corresponds to a hybrid weight variable, and these weights are required to satisfy non-negativity and normalization constraints.

[0068] In the specific implementation, the weight range of each target prototype formula is divided into multiple equal probability intervals with the same number as the sample size; a sample point is randomly selected from each equal probability interval, and the preset space filling index of the sample set composed of the target prototype formula is calculated; the sample position is iteratively optimized according to the preset space filling index to generate multiple mixed formulas, and a candidate formula set is formed by all the mixed formulas. Among them, for the sample set with low difference, the Sobol sequence is used to generate the initial sample points, and the corresponding sample positions are adjusted by the simulated annealing algorithm.

[0069] Specifically, the mixing weights for each target prototype recipe are defined as variables, forming a multidimensional mixing weight space. For each weight dimension, its range is divided into equally probable intervals equal to the target sample size. A sample point is randomly selected within each interval to ensure uniform sampling across each dimension. This operation is performed separately for multiple dimensions, and then multiple candidate mixing weight sample points are formed. Each sample is normalized so that the sum of each group of weights is 1.

[0070] Subsequently, based on the generated weighted sample set, a spatial filling index is calculated to assess its distribution uniformity in the high-dimensional mixture space. The maximum-minimum distance index (Maximin) can be used as the optimization objective. By introducing a simulated annealing algorithm, the sample positions are iteratively fine-tuned to optimize the distribution between sample points, resulting in a stronger spatial coverage.

[0071] To achieve uniform sampling in this high-dimensional mixing ratio space, a modified Latin hypercube sampling method is employed. The range of each weight dimension is divided into equally probable intervals equal to the target sample size. A point is randomly sampled within each interval to form multiple weight vector samples. All samples are normalized to ensure that the sum of each set of weights is 1. Each set of normalized weights is then linearly combined with the prototype formula for weighted summation to generate a new mixing formula.

[0072] Here, if the distances between samples are generally small and densely distributed, the sample set is considered low-discrepancy. In this case, the Sobol low-discrepancy sequence is used as the initial sampling point generation method, combined with simulated annealing to further optimize the sample layout to ensure that the sample set has good filling and representativeness in the high-dimensional space.

[0073] To meet the high-dimensional uniformity requirements, a Sobol sequence was used to generate the initial sample distribution, and a simulated annealing algorithm was used to optimize the sample point positions, thereby improving the spatial coverage and distribution uniformity of the candidate formula set. Ultimately, a candidate mixed formula set containing thousands to tens of thousands of formulas was generated, providing a data foundation for subsequent normalization, screening, and evaluation.

[0074] S103. For each mixed formula in the candidate formula set, perform adaptive robust normalization according to the median and interquartile range of the concentration of each component in the mixed formula, construct a standardized formula set, and filter it through a greedy algorithm and a clustering algorithm to form a filtered formula set.

[0075] In this step, to improve data processing robustness and subsequent screening effectiveness, standardization and two-stage screening are performed on multiple blends in the candidate set. First, the ingredient concentration information corresponding to each blend in the candidate set is obtained, and a concentration matrix is constructed. For each column in the concentration matrix (i.e., the concentration distribution of each ingredient), the median and interquartile range (IQR) are calculated. Combining these statistics, adaptive robust normalization is performed on each concentration data point to form a standardized matrix.

[0076] Here, adaptive robust normalization can be performed on the median and interquartile range of the concentrations of the ingredients in the mixture formula using the following formula:

[0077]

[0078] Where x′ represents the standard formula after adaptive robust normalization; IQR represents the interquartile range of the components; Median represents the median of the components; x represents the components in the mixture formula, and ∈ represents a minimal constant introduced to avoid zero denominators. This normalization method effectively eliminates order of magnitude differences between different components and is highly robust to outliers.

[0079] Furthermore, after normalization is completed, a two-stage screening is performed based on the standardized recipe set, and an initial standardized recipe is randomly selected from the standardized recipe set; starting from the initial standardized recipe, the standardized recipe with the largest minimum Euclidean distance from the selected recipe is iteratively selected from the standardized recipe set to form a selected recipe set, until the number of standardized recipes in the selected recipe set reaches a preset number of samples; the local density and relative distance are calculated for the standardized recipes in the selected recipe set, the cluster center is determined according to the preset decision diagram, and the screening recipes with a local density greater than a preset density threshold and a relative distance greater than a preset distance threshold are screened to form a screening recipe set.

[0080] Here, in the first stage, a greedy algorithm is executed to achieve maximum and minimum distance screening. Specifically, an initial formula sample is randomly selected from the standardized formula set, and then the sample with the largest minimum Euclidean distance from the selected sample set is iteratively selected until the preset number of samples is reached to construct the initial screening set. This strategy ensures the maximum difference between the selected samples, thereby enhancing representativeness. In the second stage, a density clustering analysis is performed on the preliminary screening formula set obtained in the first stage. The local density and relative distance index of each formula sample are calculated, and the cluster center is automatically determined according to the preset decision diagram method. On this basis, the mixed formulas with a local density higher than the set density threshold and a significant distance from the high-density sample are screened out to form the final screening formula set.

[0081] In this way, through the above-mentioned normalization and two-stage screening steps, it is possible to effectively remove redundant and duplicate samples while maintaining sample diversity, providing high-quality candidate inputs for subsequent formulation evaluation and experimental design.

[0082] S104. Calculate the preset evaluation index corresponding to the screening formula set, construct a relationship model between sample size and coverage based on the preset evaluation index and the number of formulas, predict the minimum sample size required to achieve the target coverage through Gaussian process regression, and select the target mixed formula set corresponding to the minimum sample size from the screening formula set.

[0083] In this step, in order to determine the minimum number of experimental samples required to ensure formula coverage, the evaluation index calculation and sample size prediction operations are further performed on the screening formula set. First, a set of preset multi-dimensional evaluation indicators are calculated for all mixed formula samples in the screening formula set.

[0084] In the specific implementation, the preset evaluation indicators and the number of recipes corresponding to the screening recipes are used as input to construct a relationship model between sample size and coverage; the number of cost rows is defined according to the preset experimental resource consumption, and the minimum sample size required to achieve the target coverage is predicted through Gaussian process regression based on the relationship model, and experimental plans under different sample numbers are generated, and the Pareto optimal number of experimental groups that balances experimental resource consumption and sample size is output.

[0085] Here, the preset evaluation indicators include at least diversity dimension and composition dimension; for the diversity dimension, the preset evaluation indicators include at least average pairwise Euclidean distance, minimum pairwise Euclidean distance, candidate pool coverage distance and spatial distribution width; for the composition dimension, the preset evaluation indicators include at least minimum concentration, maximum concentration, concentration fluctuation range and concentration fluctuation multiple.

[0086] Among them, the average pairwise Euclidean distance (APD) is the average of the Euclidean distances between all the recipes in the screening recipe set, which is used to evaluate the overall diversity and distribution breadth of the sample set. The larger the average pairwise Euclidean distance, the greater the difference between the recipes in the screening recipe set, which reflects a higher diversity. The minimum pairwise Euclidean distance (MPD) is the minimum of the Euclidean distances between all the recipes in the screening recipe set, and the minimum of the Euclidean distances between all the recipes. If the minimum pairwise Euclidean distance is large, it means that there are no duplicate recipes that are too similar in the screening recipe set, which helps to ensure the distinction between the recipes. The candidate pool coverage distance (CD) calculates the distance from each recipe in the screening recipe set to the nearest recipe in the library, and then calculates the average (or maximum) distance, that is, the average or maximum value of the distance from each sample in the candidate recipe set to the nearest recipe in the screening set. The smaller the average distance, the better the final library covers the set of screening recipes. This means that each recipe in the library can "represent" a large candidate region. The maximum distance reflects the "representativeness" of the most marginal candidate samples; the smaller the value, the better. The spatial distribution width (such as the determinant of the covariance matrix) calculates the determinant of the covariance matrix of each dimension of the recipes in the screening recipe set after normalization. This is used to measure the "volume" covered by the recipes in the multidimensional feature space. A larger determinant indicates a larger "volume" of these recipes in the feature space. From the perspective of experimental design, this means better spatial coverage.

[0087] Among them, the minimum concentration is the minimum value of each component in the mixed formula library, reflecting the lowest concentration level of the component; the maximum concentration is the maximum value of each component in the screening formula set, reflecting the highest concentration level of the component; the concentration fluctuation range is determined by the difference between the maximum concentration and the minimum concentration, indicating the absolute range of variation of the component concentration; the fluctuation multiple is the maximum concentration divided by the sum of the minimum concentration and a very small constant. This indicator gives the relative proportion (fold) of the change in component concentration, reflecting the concentration dispersion of the component in the screening formula set. If the concentration fluctuation multiple value is large, it means that the component has large fluctuations in the screening formula set, otherwise it is relatively stable.

[0088] Furthermore, the above-mentioned evaluation indicators constitute the input feature set, which is used together with the current sample size as modeling data. Next, a mapping relationship model between sample size and coverage is constructed. Specifically, with sample size as the independent variable and coverage index (such as CD) as the dependent variable, the Gaussian Process Regression (GPR) algorithm is used to fit the sample size-coverage function. The GPR model has non-parametric properties and is suitable for function modeling and uncertainty assessment under small sample data.

[0089] Here, based on the established relationship model, according to the target coverage threshold set by the user, the minimum sample size required to meet the target coverage is obtained through reverse deduction through the GPR prediction model. Optionally, an experimental resource consumption function can also be introduced to perform Pareto optimization on the sample size selection based on the relationship between the number of samples and the experimental overhead to achieve a dual balance between coverage and cost.

[0090] Finally, the first mixed formula sample is selected from the screening formula set according to the evaluation index ranking or representative score to form the final target mixed formula set, which serves as the input set for subsequent experimental verification and performance testing.

[0091] In this way, this step can significantly compress the scale of the experiment and maximize the experimental efficiency and resource utilization while maintaining the representativeness and spatial coverage quality.

[0092] See also Figure 2 FIG. 1 is a flow chart of another method provided by an embodiment of the present disclosure, wherein the method includes steps S201 to S204, and is applied to Figure 1 After step S104 in the method shown in , wherein:

[0093] S201 : For candidate recipes that are not selected in the candidate recipe set, calculate corresponding outliers, perform uncertainty sampling, and generate unsampled areas.

[0094] S202: Construct a random forest prediction model using the screened formula set as a training set, determine the component distribution diversity prediction results of the unsampled area, and generate a supplementary formula.

[0095] S203: adding the supplementary formula to the candidate formula set, and iterating the screening process based on the greedy algorithm and the clustering algorithm and the calculation process of the preset evaluation index.

[0096] S204. If the average pairwise distance corresponding to the candidate formula increases by less than a preset first amplitude threshold in a consecutive preset number of iterations, and the coverage distance of the candidate formula set decreases by less than a preset second amplitude threshold, the iteration is terminated and the final target mixed formula set is output.

[0097] In the specific implementation, in order to further improve the diversity and spatial coverage capability of the screening recipe set, after the target mixed recipe set is preliminarily screened out based on the greedy algorithm and clustering algorithm, the unselected candidate recipes are analyzed and reused in combination with the active learning strategy to achieve adaptive iterative optimization of the candidate space. First, for the remaining mixed recipes in the candidate recipe set that have not yet been included in the screening recipe set, their standardized representation is constructed, and the outlier index of each recipe sample is calculated to identify samples with high uncertainty or lack of representativeness.

[0098] Preferably, Mahalanobis distance, local anomaly factor and other anomaly detection methods can be used to calculate the degree of outlier of each unsampled recipe point relative to the screened recipe.

[0099] On this basis, several representative samples were selected from the recipes with the highest outlier rankings to form a subset of samples from the unsampled regions. Using the samples from the selected recipe set as training data, the recipe ingredient concentrations as input features, and preset diversity indicators (such as local density, Euclidean distance, and coefficient of variation) as prediction targets, a random forest prediction model was constructed to estimate the diversity potential of the uncovered regions in the candidate space.

[0100] Here, the model is used to score each recipe point in the unsampled region. Samples with predicted scores above a threshold are selected as new sampling points to generate a supplementary recipe subset. The supplementary recipes are then incorporated into the original candidate recipe set. Normalization, greedy maximum-minimum distance screening, and density peak clustering are re-performed. Pre-set evaluation metrics are calculated based on the latest recipe set, and the selected recipe set and target mixed recipe set are updated.

[0101] The above-mentioned active learning-supplementary screening-indicator evaluation process constitutes a complete iterative cycle. To avoid redundant iterations and performance overfitting, an iteration termination condition is set: when the average pairwise distance (APD) of the screening recipe set increases by less than a preset first threshold (such as 5%) within a preset number of consecutive times (such as 3 rounds), and the coverage distance (CD) of the candidate pool decreases by less than a preset second threshold (such as 3%), the spatial exploration is considered to have converged, the iteration is stopped, and the final target mixed recipe set is output as the final output result of the recipe optimization.

[0102] In this way, through the above method, the system can achieve adaptive optimization driven by model feedback, effectively improve the representativeness, redundancy and experimental value of samples, and significantly improve the efficiency and performance of formula screening.

[0103] The disclosed embodiments provide a method for generating a cell culture medium mixed formula. The method comprises the following steps: obtaining multiple historical prototype formulas, calculating the component differential entropy of each component in all the historical prototype formulas, and screening a target prototype formula whose component differential entropy is greater than a preset threshold; using the weight of the target prototype formula as a variable, Latin hypercube sampling is used to generate multiple mixed formulas in a high-dimensional weight space to form a candidate formula set; for each mixed formula in the candidate formula set, adaptive robust normalization is performed based on the median and interquartile range of the concentration of each component in the mixed formula to construct a standardized formula set, which is then screened using a greedy algorithm and a clustering algorithm to form a screened formula set; calculating a preset evaluation index corresponding to the screened formula set, constructing a relationship model between sample size and coverage based on the preset evaluation index and the number of formulas, predicting the minimum sample size required to achieve target coverage through Gaussian process regression, and selecting a target mixed formula set corresponding to the minimum sample size from the screened formula set. The method can achieve a high-dimensional mixed space and high sample screening efficiency, improve the intelligent level of cell culture medium formula design, reduce development cycle and cost, and enhance industrial application performance.

[0104] Those skilled in the art will understand that in the above-mentioned method of the specific implementation method, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0105] Based on the same inventive concept, the embodiments of the present disclosure also provide a cell culture medium mixed formula generation device corresponding to the cell culture medium mixed formula generation method. Since the principle of solving the problem by the device in the embodiments of the present disclosure is similar to that of the above-mentioned cell culture medium mixed formula generation method in the embodiments of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0106] See also Figure 3 , Figure 3 Schematic diagram of a cell culture medium mixing formula generating device provided in an embodiment of the present disclosure. Figure 3 As shown in FIG, the cell culture medium mixed formula generating device 300 provided by the embodiment of the present disclosure includes:

[0107] The prototype formula generation module 310 is used to obtain multiple historical prototype formulas, calculate the component difference entropy of each component in all the historical prototype formulas, and select a target prototype formula whose component difference entropy is greater than a preset threshold.

[0108] The candidate recipe set sampling module 320 is configured to generate multiple mixed recipes in a high-dimensional weight space by using the weight of the target prototype recipe as a variable using Latin hypercube sampling to form a candidate recipe set.

[0109] The recipe screening module 330 is used to perform adaptive robust normalization on each of the mixed recipes in the candidate recipe set according to the median and interquartile range of the concentration of each component in the mixed recipe, construct a standardized recipe set, and screen it through the greedy algorithm and clustering algorithm to form a screened recipe set.

[0110] The experimental recipe output module 340 is used to calculate the preset evaluation indicators corresponding to the screening recipe set, construct a relationship model between sample size and coverage based on the preset evaluation indicators and the number of recipes, predict the minimum sample size required to achieve the target coverage through Gaussian process regression, and select the target mixed recipe set corresponding to the minimum sample size in the screening recipe set.

[0111] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference can be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0112] The disclosed embodiments provide a device for generating a cell culture medium mixed formula. The device obtains multiple historical prototype formulas, calculates the component differential entropy of each component in all the historical prototype formulas, and selects a target prototype formula whose component differential entropy is greater than a preset threshold. Using the weight of the target prototype formula as a variable, Latin hypercube sampling is used to generate multiple mixed formulas in a high-dimensional weight space to form a candidate formula set. For each mixed formula in the candidate formula set, adaptive robust normalization is performed based on the median and interquartile range of the concentration of each component in the mixed formula to construct a standardized formula set, which is then screened using a greedy algorithm and a clustering algorithm to form a screened formula set. Preset evaluation indicators corresponding to the screened formula set are calculated, and a relationship model between sample size and coverage is constructed based on the preset evaluation indicators and the number of formulas. The device predicts the minimum sample size required to achieve target coverage through Gaussian process regression, and selects a target mixed formula set corresponding to the minimum sample size from the screened formula set. This device can achieve a high-dimensional mixed space and high sample screening efficiency, improve the intelligent level of cell culture medium formula design, reduce development cycle and cost, and enhance industrial application performance.

[0113] Corresponding to Figure 1 and Figure 2 The method for generating a mixed formula of a cell culture medium in the present disclosure also provides an electronic device 400, such as Figure 4 FIG. 4 is a schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present disclosure, including:

[0114] Processor 41, memory 42, and bus 43; memory 42 is used to store execution instructions, including memory 421 and external memory 422; the memory 421 here is also called internal memory, which is used to temporarily store the operation data in the processor 41 and the data exchanged with the external memory 422 such as the hard disk. The processor 41 exchanges data with the external memory 422 through the memory 421. When the electronic device 400 is running, the processor 41 and the memory 42 communicate through the bus 43, so that the processor 41 executes Figure 1 and Figure 2 Steps in the method for generating a cell culture medium mix recipe.

[0115] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program executes the steps of the method for generating a cell culture medium mixture formula described in the above method embodiment. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0116] The present disclosure also provides a computer program product including computer instructions. When the computer instructions are executed by a processor, the steps of the method for generating a cell culture medium mixing formula described in the above method embodiment can be performed. For details, please refer to the above method embodiment, which will not be repeated here.

[0117] The computer program product may be implemented in hardware, software, or a combination thereof. In one embodiment, the computer program product is implemented as a computer storage medium. In another embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0118] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in 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. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0119] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0120] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0121] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0122] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The scope of protection of the present disclosure is not limited thereto. Although the present disclosure has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed in the present disclosure, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure shall be subject to the scope of protection of the claims.

Claims

1. A method for generating a mixed formula of a cell culture medium, characterized in that: include: Acquire multiple historical prototype formulas, calculate the component difference entropy of each component in all the historical prototype formulas, and select a target prototype formula whose component difference entropy is greater than a preset threshold; Using the weight of the target prototype recipe as a variable, Latin hypercube sampling is used to generate multiple mixed recipes in a high-dimensional weight space to form a candidate recipe set; For each of the mixed formulas in the candidate formula set, adaptive robust normalization is performed according to the median and interquartile range of the concentration of each component in the mixed formula to construct a standardized formula set and screen it through a greedy algorithm and a clustering algorithm to form a screened formula set; Calculate the preset evaluation indicators corresponding to the screening formula set, construct a relationship model between sample size and coverage based on the preset evaluation indicators and the number of formulas, predict the minimum sample size required to achieve the target coverage through Gaussian process regression, and select the target mixed formula set corresponding to the minimum sample size from the screening formula set.

2. The method according to claim 1, characterized in that Calculating the component difference entropy of each component in all the historical prototype formulas, and screening the target prototype formula whose component difference entropy is greater than a preset threshold, specifically including: Determine a standardized value of the coefficient of variation corresponding to each ingredient in the historical prototype formula, and calculate the ingredient difference entropy corresponding to the ingredient based on the standardized value of the coefficient of variation, wherein the ingredient difference entropy is used to measure the degree of diversity between the historical prototype formulas; The prototype formula having the component difference entropy greater than the preset threshold is screened, and the concentration range corresponding to each component in the target prototype formula is expanded by a preset ratio to generate the target prototype formula.

3. The method according to claim 1, characterized in that Using the weight of the target prototype recipe as a variable, Latin hypercube sampling is used to generate multiple mixed recipes in a high-dimensional weight space to form a candidate recipe set, specifically including: Dividing the weight range of each target prototype formula into a plurality of equal probability intervals whose number is equal to the sample size; Randomly selecting a sample point for each of the equal probability intervals, and calculating a preset space filling index of the sample set consisting of the target prototype formula; The sample positions are iteratively optimized according to the preset space filling index to generate a plurality of the mixed formulas, and the candidate formula set is formed by all the mixed formulas. Specifically, for the sample set with low difference, the Sobol sequence is used to generate the initial sample points, and the corresponding sample positions are adjusted by the simulated annealing algorithm.

4. The method according to claim 1, wherein The standardized formula set is filtered by the greedy algorithm and the clustering algorithm to form the filtered formula set, which specifically includes: Randomly selecting an initial standardized formula from the standardized formula set; Starting from the initial standardized recipe, iteratively selecting the standardized recipe with the largest minimum Euclidean distance from the selected recipes in the standardized recipe set to form a selected recipe set, until the number of the standardized recipes in the selected recipe set reaches a preset sample number; The local density and relative distance are calculated for the standardized recipes in the selected recipe set, the cluster center is determined according to a preset decision diagram, and the filtered recipes whose local density is greater than a preset density threshold and whose relative distance is greater than a preset distance threshold are screened to form the filtered recipe set.

5. The method according to claim 1, wherein The preset evaluation indicators include at least diversity dimension and component dimension; For the diversity dimension, the preset evaluation indicators include at least the average pairwise Euclidean distance, the minimum pairwise Euclidean distance, the candidate pool coverage distance, and the spatial distribution width; For the component dimension, the preset evaluation indicators include at least minimum concentration, maximum concentration, concentration fluctuation range and concentration fluctuation multiple.

6. The method according to claim 1, characterized in that The relationship model between sample size and coverage is constructed based on the preset evaluation index and the number of formulations, and the minimum sample size required to achieve the target coverage is predicted by Gaussian process regression, specifically including: Taking the preset evaluation index and the number of formulations corresponding to the screening formulations as input, constructing the relationship model between sample size and coverage; The number of cost rows is defined according to the preset experimental resource consumption, and the minimum sample size required to achieve the target coverage is predicted through Gaussian process regression based on the relationship model. Experimental plans under different sample numbers are generated, and the Pareto optimal number of experimental groups that balances experimental resource consumption and sample size is output.

7. The method according to claim 1, characterized in that After selecting a target mixed formula set corresponding to the minimum sample size from the screening formula set, the method further includes: For the candidate recipes that are not selected in the candidate recipe set, calculating the corresponding outliers and performing uncertainty sampling to generate unsampled areas; Constructing a random forest prediction model using the screened formula set as a training set, determining the predicted results of the component distribution diversity in the unsampled area, and generating a supplementary formula; Adding the supplementary formula to the candidate formula set, and iterating the screening process based on the greedy algorithm and the clustering algorithm and the calculation process of the preset evaluation index; If the average pairwise distance corresponding to the candidate formula increases by less than a preset first amplitude threshold in a consecutive preset number of iterations, and the coverage distance of the candidate formula set decreases by less than a preset second amplitude threshold, the iteration is terminated and the final target mixed formula set is output.

8. A device for generating a cell culture medium mixed formula, characterized in that: include: A prototype formula generation module is used to obtain multiple historical prototype formulas, calculate the component difference entropy of each ingredient in all the historical prototype formulas, and select a target prototype formula whose component difference entropy is greater than a preset threshold; A candidate recipe set sampling module is used to generate multiple mixed recipes in a high-dimensional weight space using the weight of the target prototype recipe as a variable using Latin hypercube sampling to form a candidate recipe set; A recipe screening module is used to perform adaptive robust normalization on each of the mixed recipes in the candidate recipe set according to the median and interquartile range of the concentration of each component in the mixed recipe, construct a standardized recipe set, and screen it using a greedy algorithm and a clustering algorithm to form a screened recipe set; The experimental formula output module is used to calculate the preset evaluation indicators corresponding to the screening formula set, construct a relationship model between sample size and coverage based on the preset evaluation indicators and the number of formulas, predict the minimum sample size required to achieve the target coverage through Gaussian process regression, and select the target mixed formula set corresponding to the minimum sample size from the screening formula set.

9. An electronic device, characterized in that: include: A processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the steps of the method for generating a cell culture medium mixing formula according to any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the method for generating a cell culture medium mixing formula according to any one of claims 1 to 7.

Images