Construction method and application of MDCK cell low-serum culture medium based on KAN modeling and Bayesian optimization

Abstract

The invention discloses an MDCK cell low-serum culture medium construction method based on KAN modeling and Bayesian optimization and application thereof.The MDCK cell low-serum culture medium construction method comprises the steps that firstly, key influence factors are screened out from numerous nutrient components through multi-factor experimental design, and then a high-precision nonlinear prediction model between the key factors and cell performance is established through KAN; based on the model, a Bayesian optimization algorithm is adopted for rapid optimization, and the optimal concentration of each component is determined. The cell line culture medium has universality and can be migrated and applied to culture medium development of other cell lines. According to the MDCK cell low-serum culture medium constructed through the method, the adding amount of fetal calf serum (FBS) is only 2%-4% (v / v), the formula is systematically optimized, efficient growth of cells can be directly supported, and pre-domestication is not needed. The culture medium has obvious effects on promoting MDCK cell proliferation, improving virus (such as influenza virus) titer and reducing production cost, and is suitable for large-scale production of vaccines.

Description

Technical Field

[0001] This invention belongs to the field of biotechnology and relates to a method for constructing and applying a low-serum culture medium for MDCK cells based on KAN modeling and Bayesian optimization. Background Technology

[0002] MDCK cell lines, due to their high sensitivity to various viruses and stable growth characteristics, have become the core cell matrix for the production of viral vaccines such as influenza vaccines. Currently, industrial MDCK cell culture mainly uses culture media supplemented with 10%-20% fetal bovine serum (FBS). This high-serum culture mode has significant drawbacks: First, FBS is expensive, accounting for a large proportion of the total cost of the culture medium; second, as a biologically derived product, FBS has a complex composition and large batch-to-batch variability, directly affecting the stability of cell growth and virus yield, increasing the difficulty and risk of quality control in the production process; finally, the large amount of contaminating proteins in serum burdens the downstream virus isolation and purification.

[0003] To address these challenges, developing culture media with clearly defined chemical compositions and low or serum-free or low-serum content has become an industry trend. However, existing culture media development technologies face several bottlenecks: First, optimization efficiency is low, with traditional trial-and-error methods or single-factor experiments requiring enormous workloads and failing to examine interactions between multiple components. Second, model capabilities are insufficient; even with Design of Experiments (DOE) and response surface methodology, the quadratic polynomial models they rely on struggle to accurately fit the complex nonlinear dose-response relationships prevalent in biological systems. Third, some existing serum-free or low-serum culture media often require a "acclimatization" process for cells to adapt to the new nutrient environment, which prolongs the process development cycle and limits the range of compatible cell lines.

[0004] In recent years, machine learning methods have been introduced into biological process optimization, but their application in culture medium development still faces problems such as poor model interpretability, insufficient consideration of biological system specificity, and weak integration with experimental iteration. Therefore, there is an urgent need for an intelligent culture medium development method that can efficiently analyze complex nutrient interactions, accurately predict cell performance, and quickly guide to optimal formulations, as well as the resulting culture medium products that can be efficiently applied without complex acclimatization. Summary of the Invention

[0005] This invention provides a method and application for constructing low-serum culture medium for MDCK cells based on KAN modeling and Bayesian optimization. It combines Kolmogorov-Arnold network (KAN) nonlinear modeling technology with Bayesian global optimization algorithm to form a closed-loop optimization framework, so as to systematically solve the efficiency and accuracy problems in traditional culture medium development.

[0006] The technical solution adopted in this invention is as follows:

[0007] A method for constructing a low-serum culture medium for MDCK cells based on KAN modeling and Bayesian optimization includes the following steps:

[0008] (1) Screening of key components: The candidate nutrient library includes amino acids, vitamins, inorganic salts, lipids and culture medium additives. Key factors that have a significant impact on the growth or function of target cells are screened out. The key factors screened out include at least three of the following: amino acids, vitamins, inorganic salts, lipids and culture medium additives. The criterion for judging the significant impact is a p-value of ANOVA < 0.05.

[0009] (2) Dataset construction: Based on the selected key components, a multi-level experimental scheme is constructed using response surface design. Through cell culture experiments, a dataset of the correlation between the concentration of key components and at least one cell performance index is obtained. (3) Nonlinear relationship modeling: The dataset obtained in step (2) is trained using the Kolmogorov-Arnold network to construct a prediction model with the concentration of key components as input and the cell performance index as output. (4) Global optimization: Using the KAN prediction model constructed in step (3) as the objective function, the Bayesian optimization method is used for iterative search to determine the combination of key component concentrations that optimizes the cell performance index.

[0010] The method of this invention combines the predictive power of artificial intelligence with the search efficiency of optimization algorithms, achieving a significant reduction in experimental workload (typically more than 50%) and an improvement in optimization accuracy while ensuring biological understanding. This method is transferable; by changing the target cells and the initial nutrient component library, it can be applied to the development of culture media for other commonly used industrial cell lines such as CHO and Vero.

[0011] In another aspect, the present invention provides a low-serum culture medium for MDCK cells obtained by the above-described construction method. The core innovation of this culture medium is that its formulation is a direct product of data-driven and model-optimized methods, rather than the result of empirical adjustments.

[0012] The MDCK cell low serum culture medium contains 9 lipids, 20 amino acids, 8 vitamins, 7 salts, 3 trace elements, 1 buffer, 1 acid-base indicator, and 8 culture medium additives.

[0013] The concentration ranges of each component are as follows:

[0014] (1) Lipids: Arachidonic acid 0.0016-0.0024 mg / L, Linoleic acid 0.008-0.012 mg / L, Linolenic acid 0.008-0.012 mg / L, Myristic acid 0.008-0.012 mg / L, Oleic acid 0.008-0.012 mg / L, Palmitic acid 0.008-0.012 mg / L, Stearic acid 0.008-0.012 mg / L, Cholesterol 0.176-0.264 mg / L, Polyoxyethylene sorbitan monooleate 1.76-2.64 mg / L;

[0015] (2) Amino acids: L-arginine 67.2-100.8 mg / L, L-cysteine ​​50.4-75.6 mg / L, L-isoleucine 84.0-126.0 mg / L, L-leucine 84.0-126.0 mg / L, L-lysine 116.8-175.2 mg / L, L-phenylalanine 52.8-79.2 mg / L, L-threonine 76.0-114.0 mg / L, disodium tyrosine 83.2-124.8 mg / L, L-valine 75.2-112.8 mg / L, L-alanyl-L-glutamine 347.55-521.33 mg / L, L-glutamine 292.35-876.92 mg / L, L- Tyrosine 54.36-163.08 mg / L, glycine 24.0-36.0 mg / L, L-histidine 33.6-50.4 mg / L, L-methionine 24.0-36.0 mg / L, L-serine 33.6-50.4 mg / L, L-tryptophan 12.8-19.2 mg / L, L-asparagine 10.57-31.71 mg / L, L-aspartic acid 8.0-12.0 mg / L, L-proline 13.6-20.4 mg / L;

[0016] (3) Vitamins: choline chloride 3.2-4.8 mg / L, D-calcium pantothenate 3.2-4.8 mg / L, folic acid 3.2-4.8 mg / L, nicotinamide 3.2-4.8 mg / L, pyridoxine hydrochloride 3.2-4.8 mg / L, riboflavin 0.32-0.48 mg / L, thiamine hydrochloride 3.2-4.8 mg / L, tocopherol acetate 0.056-0.084 mg / L;

[0017] (4) Salts: calcium chloride 160.0-240.0 mg / L, magnesium sulfate 78.16-117.24 mg / L, potassium chloride 320.0-480.0 mg / L, sodium bicarbonate 2960.0-4440.0 mg / L, sodium chloride 3800.0-5700.0 mg / L, sodium dihydrogen phosphate 87.2-130.8 mg / L, disodium hydrogen phosphate 56.8-85.2 mg / L;

[0018] (5) Trace elements: sodium selenite 80.0-120.0 mg / L, ferric nitrate 0.08-0.12 mg / L, copper sulfate 12.8-19.2 mg / L;

[0019] (6) Buffer: 4-hydroxyethylpiperazine ethanesulfonic acid 4766.4-7149.6 mg / L;

[0020] (7) Acid-base indicator: Phenol red 12.0-18.0 mg / L;

[0021] (8) Culture medium supplements: D-glucose 3600.0-5400.0 mg / L, sodium pyruvate 88.0-132.0 mg / L, peptone 200.0-300.0 mg / L, bovine serum albumin 120.0-180.0 mg / L, transferrin 4.4-6.6 mg / L, insulin 8.0-12.0 mg / L, hypoxanthine 4.0-6.0 mg / L, ethanolamine 1.6-2.4 mg / L;

[0022] The optimal concentrations of each component are as follows:

[0023] (1) Lipids: arachidonic acid 0.002 mg / L, linoleic acid 0.01 mg / L, linolenic acid 0.01 mg / L, myristic acid 0.01 mg / L, oleic acid 0.01 mg / L, palmitic acid 0.01 mg / L, stearic acid 0.01 mg / L, cholesterol 0.22 mg / L, polyoxyethylene sorbitan monooleate 2.2 mg / L;

[0024] (2) Amino acids: L-arginine 84 mg / L, L-cysteine ​​63 mg / L, L-isoleucine 105 mg / L, L-leucine 105 mg / L, L-lysine 146 mg / L, L-phenylalanine 66 mg / L, L-threonine 95 mg / L, disodium tyrosine 104 mg / L, L-valine 94 mg / L, L-alanyl-L-glutamine 434.44 mg / L, L-glutamine 292.3 mg / L, L-tyrosine 54.36 mg / L, glycine 30 mg / L, L-histidine 42 mg / L, L-methionine 30 mg / L, L-serine 42 mg / L, L-tryptophan 16 mg / L, L-asparagine 11.34 mg / L, L-aspartic acid 10 mg / L, L-proline 17 mg / L;

[0025] (3) Vitamins: choline chloride 4 mg / L, D-calcium pantothenate 4 mg / L, folic acid 4 mg / L, nicotinamide 4 mg / L, pyridoxine hydrochloride 4 mg / L, riboflavin 0.4 mg / L, thiamine hydrochloride 4 mg / L, tocopherol acetate 0.07 mg / L;

[0026] (4) Salts: calcium chloride 200 mg / L, magnesium sulfate 97.7 mg / L, potassium chloride 400 mg / L, sodium bicarbonate 3700 mg / L, sodium chloride 4750 mg / L, sodium dihydrogen phosphate 109 mg / L, disodium hydrogen phosphate 71 mg / L;

[0027] (5) Trace elements: sodium selenite 100 mg / L, ferric nitrate 0.1 mg / L, copper sulfate 16 mg / L;

[0028] (6) Buffer: 4-hydroxyethylpiperazine ethanesulfonic acid 5958 mg / L;

[0029] (7) Acid-base indicator: Phenol red 15 mg / L;

[0030] (8) Culture medium supplements: D-glucose 4500 mg / L, sodium pyruvate 110 mg / L, peptone 250 mg / L, bovine serum albumin 150 mg / L, transferrin 5.5 mg / L, insulin 10 mg / L, hypoxanthine 5 mg / L, ethanolamine 2 mg / L.

[0031] The relevant parameters of the MDCK cells in the low-serum culture medium are as follows:

[0032] Serum content: The amount of fetal bovine serum (FBS) added is significantly reduced to 2%-4% (v / v), which is a significant reduction in cost compared to the traditional 10% dosage.

[0033] No acclimatization required: Because the formula is precisely matched to the nutritional requirements and metabolic characteristics of MDCK cells under low serum pressure, cells can be directly inoculated using conventional methods and exhibit excellent growth performance when transferred from conventional medium containing 10% FBS, without any adaptation process. This simplifies the process and shortens production preparation time.

[0034] Key components: Through optimization methods, L-glutamine, L-tyrosine, and L-asparagine were identified as key amino acids affecting the proliferation of MDCK cells in low serum environments. Their optimized concentration range and ratio are one of the core components of this culture medium's "acclimatization-free" and high-performance characteristics.

[0035] Complete formulation: In addition to key amino acids, the culture medium also contains optimized lipids, vitamins, salts, trace elements, buffer systems (such as HEPES), energy substances (such as glucose), protein / peptide supplements (such as peptone, BSA), and other growth-promoting factors (such as insulin, transferrin), forming a complete system that is nutritionally balanced and supports efficient metabolism.

[0036] The beneficial effects of this invention are:

[0037] This invention's construction method deeply integrates KAN nonlinear modeling with Bayesian optimization, providing an efficient and precise universal solution for optimizing complex systems like cell culture media that are multi-factorial and highly nonlinear. This method significantly reduces the workload and time cost of experimental screening and optimization. FBS usage is reduced to 2%-4%, and direct raw material costs are reduced by approximately 68%. Cells can be directly transferred from conventional serum culture media, saving weeks of acclimatization time, simplifying the process, and improving production flexibility. Under low serum conditions, superior cell growth rate and virus yield are achieved compared to traditional high serum culture media. The chemical composition is more clearly defined, the formulation is optimized by the model, batch-to-batch reproducibility is good (CV≤5%), and product quality is stable. This culture medium and its optimization method provide strong technical support and product selection for the low-cost, high-efficiency, and high-quality large-scale production of biological products such as influenza vaccines, and have broad market application prospects.

[0038] The low-serum culture medium for MDCK cells obtained in this invention contains only 2%-4% (v / v) fetal bovine serum (FBS). Its formulation has been systematically optimized to directly support efficient cell growth without prior acclimatization. This medium is highly effective in promoting MDCK cell proliferation, increasing viral titers (such as influenza virus), and reducing production costs, making it suitable for large-scale vaccine production. This low-serum MDCK cell culture medium supports high-density MDCK cell growth, with a significantly shorter population doubling time than traditional high-serum media and unoptimized low-serum media, and cell viability remains at a high level (>95%) throughout the culture cycle. It is particularly suitable for the amplification of viruses such as influenza virus. After inoculation with the virus, a higher peak viral titer can be obtained, and the virus harvest time is advanced, improving production efficiency. The culture medium composition is clearly defined, and with the optimized formulation, the batch-to-batch consistency (CV value) of cell growth and virus production is far superior to media containing a high proportion of serum, meeting the quality control requirements of large-scale production. Due to the more defined composition and fewer impurities, downstream purification processes are simplified, resulting in higher recovery rates; the production cycle is shortened, and overall production capacity is increased. Attached Figure Description

[0039] Figure 1 The example uses a Plackett-Burman design to screen key nutrient components, resulting in a standardized Pareto plot.

[0040] Figure 2 The bar chart in this example shows the comparison between the predicted and experimental values ​​of the KAN model in the first round of Bayesian optimization iteration.

[0041] Figure 3 This is a bar chart comparing the predicted and experimental values ​​of the KAN model in the second round of Bayesian optimization iterations, as shown in the example.

[0042] Figure 4 Comparison of MDCK cell growth curves cultured in culture medium, commercial high-serum culture medium, and basal low-serum culture medium.

[0043] Figure 5 A comparison of virus titers harvested after culturing MDCK cells in different culture media and inoculating them with the virus.

[0044] Figure 6 A metabolic comparison of glucose consumption and lactate accumulation when MDCK cells are cultured in different culture media. Detailed Implementation

[0045] The specific embodiments of the present invention are described in detail below with reference to the technical solutions and accompanying drawings.

[0046] Example 1: The optimal formulation of low serum culture medium for MDCK cells was gradually obtained by applying intelligent optimization methods.

[0047] 1. Target Determination and Candidate Library Establishment

[0048] Optimization objective: To maximize the proliferation rate of MDCK cells under low serum (2% FBS) conditions.

[0049] Initial candidate nutrient component library: 27 in total, covering amino acids (such as arginine, glutamine, etc.), vitamins (such as choline, calcium pantothenate, etc.), lipid precursors (such as fatty acids, cholesterol, etc.), trace elements and other commonly used additives (such as glucose, peptone, etc.).

[0050] 2. Screening of key components (Step 1)

[0051] Experimental design: A Plackett-Burman design was used to screen 27 candidate factors at two levels (low concentration / high concentration). A total of 32 experimental runs were arranged, including 4 center points to estimate experimental error. All experiments were performed with 3 biological replicates.

[0052] Experimental procedure: Logarithmic growth phase MDCK cells were seeded into 96-well plates. After adhesion, the medium was replaced with the test medium containing different combinations of components and 2% FBS. After 72 hours of culture, cell viability was detected using the CCK-8 assay, and the relative proliferation rate was calculated.

[0053] Data Analysis: Statistical software (such as Minitab) was used to perform analysis of variance on the experimental results. Using a p-value < 0.05 as the criterion, three key factors with highly significant effects on cell proliferation were identified: L-glutamine (p = 0.00002), L-tyrosine (p = 0.00209), and L-asparagine (p = 0.00687). The model was statistically significant overall (p < 0.0023), with a high R² value, indicating effective screening.

[0054] 3. Construct the modeling dataset (Step 2)

[0055] Experimental Design: A three-factor, three-level Box-Behnken response surface methodology was employed for the three key amino acids mentioned above. Each factor was set at low, medium, and high concentrations (e.g., L-glutamine: 292.35, 584.64, 876.92 mg / L; L-tyrosine: 54.36, 108.72, 163.08 mg / L; L-asparagine: 10.57, 21.14, 31.71 mg / L). This design generated 17 experimental sites, including 5 center-point replicates to assess pure error.

[0056] Data acquisition: Prepare culture media according to the design and conduct cell culture experiments. Measure the proliferation rate of MDCK cells at each experimental point to obtain a "concentration-response" dataset for modeling.

[0057] 4. KAN Model Training and Validation (Step 3)

[0058] Model Construction: The 17 sets of data were randomly divided into a training set (12 sets) and an independent test set (5 sets). A 3-layer KAN network was built using Python and other tools (3 nodes in the input layer corresponding to the concentrations of 3 factors, 16 neurons in each of the 2 hidden layers, and 1 node in the output layer corresponding to the proliferation rate). The activation function used was the Gaussian radial basis function.

[0059] Training and Optimization: The Adam optimizer was used to train the network with an initial learning rate of 0.001, and learning rate decay was applied. L1 regularization was introduced to penalize network connection weights. After training, pruning was performed to remove redundant connections and nodes with weights below a threshold (e.g., 1e-4) to improve model simplicity and interpretability.

[0060] Model Validation: The model achieved a goodness-of-fit R² of 0.996 on the training set. More importantly, the Q²cv calculated using leave-one-out cross-validation was 0.987, and the prediction error was small on the independent test set, demonstrating that the model has excellent fitting ability and generalization performance, and can reliably predict cell proliferation rates under different combinations of the three factors.

[0061] 5. Bayesian optimization (step 4)

[0062] Optimization settings: The trained KAN model is used as the "black box" objective function to be optimized. The Bayesian optimization toolbox is used, and a Gaussian process is selected as the surrogate model to model the objective function and its uncertainties. The expected improvement (EI) is chosen as the acquisition function.

[0063] Iterative process:

[0064] In the first round, the optimization algorithm, based on the initial dataset (i.e., the 17 points of Box-Behnken) and the KAN model, recommended a new concentration combination (A1) that is expected to maximize the proliferation rate. Experimental verification showed a proliferation rate of 154.8%, which highly matched the model's predicted value of 154.7%.

[0065] In the second round, new data points (A1) were added to the training dataset, updating the KAN model and the Bayesian-optimized surrogate model. The algorithm then recommended the next optimal combination (A2). Experiments confirmed that the proliferation rate was further increased to 158.6%.

[0066] Convergence judgment: The deviation between the second round of predicted values ​​and experimental values ​​is extremely small, and the increase in proliferation rate is already very small, so the optimization is considered to have converged.

[0067] Optimal solution: Through two rounds of iteration, the optimal combination of concentrations of the three key amino acids was determined: L-glutamine 292.3 mg / L, L-tyrosine 54.36 mg / L, and L-asparagine 11.34 mg / L.

[0068] Example 2: Formulation and preparation of low-serum culture medium for MDCK cells

[0069] Based on the optimal concentrations of key amino acids obtained in Example 1, and considering the basic cellular nutritional requirements, the complete formulation of 1L of culture medium was determined according to claim 10. The following is a specific example of a preferred formulation:

[0070] Basic components and their amounts:

[0071] (1) Lipids: arachidonic acid 0.002 mg, linoleic acid 0.01 mg, linolenic acid 0.01 mg, myristic acid 0.01 mg, oleic acid 0.01 mg, palmitic acid 0.01 mg, stearic acid 0.01 mg, cholesterol 0.22 mg, polyoxyethylene sorbitan monooleate 2.2 mg;

[0072] (2) Amino acids: L-arginine 84 mg, L-cysteine ​​63 mg, L-isoleucine 105 mg, L-leucine 105 mg, L-lysine 146 mg, L-phenylalanine 66 mg, L-threonine 95 mg, disodium tyrosine 104 mg, L-valine 94 mg, L-alanyl-L-glutamine 434.44 mg, L-glutamine 292.3 mg, L-tyrosine 54.36 mg, glycine 30 mg, L-histidine 42 mg, L-methionine 30 mg, L-serine 42 mg, L-tryptophan 16 mg, L-asparagine 11.34 mg, L-aspartic acid 10 mg, L-proline 17 mg;

[0073] (3) Vitamins: choline chloride 4 mg, D-calcium pantothenate 4 mg, folic acid 4 mg, nicotinamide 4 mg, pyridoxine hydrochloride 4 mg, riboflavin 0.4 mg, thiamine hydrochloride 4 mg, tocopherol acetate 0.07 mg;

[0074] (4) Salts: Calcium chloride 200 mg, magnesium sulfate 97.7 mg, potassium chloride 400 mg, sodium bicarbonate 3700 mg, sodium chloride 4750 mg, sodium dihydrogen phosphate 109 mg, disodium hydrogen phosphate 71 mg;

[0075] (5) Trace elements: sodium selenite 100 mg, ferric nitrate 0.1 mg, copper sulfate 16 mg;

[0076] (6) Buffer: 4-hydroxyethylpiperazine ethanesulfonic acid 5958 mg;

[0077] (7) Acid-base indicator: Phenol red 15 mg;

[0078] (8) Culture medium supplements: D-glucose 4500 mg, sodium pyruvate 110 mg, peptone 250 mg, bovine serum albumin 150 mg, transferrin 5.5 mg, insulin 10 mg, hypoxanthine 5 mg, ethanolamine 2 mg.

[0079] Serum addition: Before use, add fetal bovine serum (FBS) at a ratio of 2% (v / v) (20 ml).

[0080] Preparation method:

[0081] Dissolve all components except serum in approximately 950 mL of water for injection in a sterile container. After thorough mixing, adjust the pH to 7.3 ± 0.1 with hydrochloric acid or sodium hydroxide solution. Add water to a final volume of 1 L and filter aseptically through a 0.22 μm filter membrane. The culture medium can be stored at 2–8°C protected from light. Just before use, add the appropriate volume of FBS under sterile conditions and mix gently.

[0082] Example 3: Experiment to verify the performance of the culture medium

[0083] This embodiment compares the culture medium of the present invention (experimental group), commercial DMEM culture medium containing 10% FBS (positive control group), and a basic 2% FBS culture medium that has not been validated and optimized (negative control group).

[0084] 1. Cell growth performance testing

[0085] Methods: MDCK cells were seeded at the same density in three different culture media and cultured for 7 days at 37°C and 5% CO2. Viable cell density was counted periodically, and viability was calculated.

[0086] result:

[0087] Population doubling time: The experimental group had the shortest doubling time, approximately 11.05 hours, which was 34% shorter than the positive control group (approximately 16.75 hours) and significantly shorter than the negative control group.

[0088] Peak cell density: The experimental group reached approximately 8.85 × 10⁻⁶ cells / day at 72 hours. 5 The number of cells / mL was higher than that of the two control groups.

[0089] Cell viability: Throughout the culture period, the cell viability of the experimental group remained above 95%, and the viability at the end of the culture period was significantly higher than that of the control group, showing better growth and lower cell death.

[0090] 2. Virus amplification capacity test

[0091] Methods: When MDCK cells reached a suitable density, they were inoculated with an influenza virus strain (e.g., BVR-26) at a low multiplicity of infection (MOI=0.01). Post-infection culture continued, and the viral fluid was harvested after 48 hours and analyzed using TCID45. 50 The viral titer was determined by a method.

[0092] Results: The viral titer of cells cultured in the experimental group reached 3.52 ± 0.06 log after 48 hours. 10 TCID 50 / mL. This titer was significantly higher than that of the positive control group (approximately 2.48 log₂ / mL). 10 TCID 50 ( / mL), demonstrating that virus production capacity was actually increased when serum usage was reduced by 80%.

[0093] 3. Metabolic Characteristic Analysis

[0094] Methods: Monitor changes in glucose and lactic acid concentrations during the culture process.

[0095] Results: In the experimental group, the glucose consumption rate and lactate production rate of cells were at a better level, and the lactate / glucose yield coefficient (YLac / Glu) was lower, indicating that the cell metabolism was more efficient, tended to a more complete oxidation pathway, reduced the accumulation of metabolic waste, and was beneficial for high-density long-term culture.

Claims

1. A method for constructing a low-serum culture medium for MDCK cells based on KAN modeling and Bayesian optimization, characterized in that, The process includes the following steps: (1) Screening of key components: using amino acids, vitamins, inorganic salts, lipids and culture medium additives as candidate nutrient libraries, screening out key factors that have a significant impact on the growth or function of target cells, and the screened key factors include at least three of the following: amino acids, vitamins, inorganic salts, lipids and culture medium additives; the judgment criterion for the significant impact is an ANOVA p-value < 0.05; (2) Dataset construction: based on the screened key components, a multi-level experimental scheme is constructed using response surface design, and a correlation dataset between the concentration of key components and at least one cell performance index is obtained through cell culture experiments; (3) Nonlinear relationship modeling: using the Kolmogorov-Arnold network to train the dataset obtained in step (2), a prediction model is constructed with the concentration of key components as input and the cell performance index as output; (4) Global optimization: using the KAN prediction model constructed in step (3) as the objective function, an iterative search is performed using the Bayesian optimization method to determine the combination of key component concentrations that optimizes the cell performance index.

2. The method for constructing a low-serum culture medium for MDCK cells according to claim 1, characterized in that, The multi-factor screening design described in step (1) is Plackett-Burman.

3. The method for constructing a low-serum culture medium for MDCK cells according to claim 1, characterized in that, The response surface design in step (2) is Box-Behnken; the cell performance indicators include cell proliferation rate, cell viability or virus titer.

4. The method for constructing a low-serum culture medium for MDCK cells according to claim 1, characterized in that, In step (3), the KAN model uses Gaussian radial basis functions as activation functions and undergoes L1 regularization pruning; the model is cross-validated, and its leave-one-out cross-validation determination coefficient Q²cv ≥ 0.

90.

5. The method for constructing a low-serum culture medium for MDCK cells according to claim 1, characterized in that, In step (4), the Bayesian optimization uses a Gaussian process as a surrogate model and the acquisition function is the expected improvement function; the iteration process is carried out for 2 to 4 rounds, and the stopping condition is that the relative deviation between the predicted values ​​of cell performance indicators obtained from two consecutive rounds of optimization and the experimental verification values ​​is ≤5%.

6. A low-serum culture medium for MDCK cells, the formulation of which is determined by the construction method according to any one of claims 1 to 5, characterized in that, The culture medium contains 2%-4% (v / v) fetal bovine serum (FBS) and can be used directly for inoculation and culture without serum adaptation of MDCK cells.

7. The MDCK cell low-serum culture medium according to claim 6, characterized in that, The culture medium contains 9 lipids, 20 amino acids, 8 vitamins, 7 salts, 3 trace elements, 1 buffer, 1 acid-base indicator, and 8 culture medium additives.

8. The MDCK cell low-serum culture medium according to claim 6 or 7, characterized in that, The concentration ranges of each component are as follows: (1) Lipids: Arachidonic acid 0.0016-0.0024 mg / L, Linoleic acid 0.008-0.012 mg / L, Linolenic acid 0.008-0.012 mg / L, Myristic acid 0.008-0.012 mg / L, Oleic acid 0.008-0.012 mg / L, Palmitic acid 0.008-0.012 mg / L, Stearic acid 0.008-0.012 mg / L, Cholesterol 0.176-0.264 mg / L, Polyoxyethylene sorbitan monooleate 1.76-2.64 mg / L; (2) Amino acids: L-arginine 67.2-100.8 mg / L, L-cysteine ​​50.4-75.6 mg / L, L-isoleucine 84.0-126.0 mg / L, L-leucine 84.0-126.0 mg / L, L-lysine 116.8-175.2 mg / L, L-phenylalanine 52.8-79.2 mg / L, L-threonine 76.0-114.0 mg / L, disodium tyrosine 83.2-124.8 mg / L, L-valine 75.2-112.8 mg / L, L-alanyl-L-glutamine 347.55-521.33 mg / L, L-glutamine 292.35-876.92 mg / L, L- Tyrosine 54.36-163.08 mg / L, glycine 24.0-36.0 mg / L, L-histidine 33.6-50.4 mg / L, L-methionine 24.0-36.0 mg / L, L-serine 33.6-50.4 mg / L, L-tryptophan 12.8-19.2 mg / L, L-asparagine 10.57-31.71 mg / L, L-aspartic acid 8.0-12.0 mg / L, L-proline 13.6-20.4 mg / L; (3) Vitamins: choline chloride 3.2-4.8 mg / L, D-calcium pantothenate 3.2-4.8 mg / L, folic acid 3.2-4.8 mg / L, nicotinamide 3.2-4.8 mg / L, pyridoxine hydrochloride 3.2-4.8 mg / L, riboflavin 0.32-0.48 mg / L, thiamine hydrochloride 3.2-4.8 mg / L, tocopherol acetate 0.056-0.084 mg / L; (4) Salts: calcium chloride 160.0-240.0 mg / L, magnesium sulfate 78.16-117.24 mg / L, potassium chloride 320.0-480.0 mg / L, sodium bicarbonate 2960.0-4440.0 mg / L, sodium chloride 3800.0-5700.0 mg / L, sodium dihydrogen phosphate 87.2-130.8 mg / L, disodium hydrogen phosphate 56.8-85.2 mg / L; (5) Trace elements: sodium selenite 80.0-120.0 mg / L, ferric nitrate 0.08-0.12 mg / L, copper sulfate 12.8-19.2 mg / L; (6) Buffer: 4-hydroxyethylpiperazine ethanesulfonic acid 4766.4-7149.6 mg / L; (7) Acid-base indicator: Phenol red 12.0-18.0 mg / L; (8) Culture medium supplements: D-glucose 3600.0-5400.0 mg / L, sodium pyruvate 88.0-132.0 mg / L, peptone 200.0-300.0 mg / L, bovine serum albumin 120.0-180.0 mg / L, transferrin 4.4-6.6 mg / L, insulin 8.0-12.0 mg / L, hypoxanthine 4.0-6.0 mg / L, ethanolamine 1.6-2.4 mg / L; The above concentration range was determined through KAN modeling and Bayesian optimization to adapt to the cellular metabolic needs under low serum conditions.

9. The MDCK cell low-serum culture medium according to claim 8, characterized in that, The optimal concentrations of each component are as follows: (1) Lipids: arachidonic acid 0.002 mg / L, linoleic acid 0.01 mg / L, linolenic acid 0.01 mg / L, myristic acid 0.01 mg / L, oleic acid 0.01 mg / L, palmitic acid 0.01 mg / L, stearic acid 0.01 mg / L, cholesterol 0.22 mg / L, polyoxyethylene sorbitan monooleate 2.2 mg / L; (2) Amino acids: L-arginine 84 mg / L, L-cysteine ​​63 mg / L, L-isoleucine 105 mg / L, L-leucine 105 mg / L, L-lysine 146 mg / L, L-phenylalanine 66 mg / L, L-threonine 95 mg / L, disodium tyrosine 104 mg / L, L-valine 94 mg / L, L-alanyl-L-glutamine 434.44 mg / L, L-glutamine 292.3 mg / L, L-tyrosine 54.36 mg / L, glycine 30 mg / L, L-histidine 42 mg / L, L-methionine 30 mg / L, L-serine 42 mg / L, L-tryptophan 16 mg / L, L-asparagine 11.34 mg / L, L-aspartic acid 10 mg / L, L-proline 17 mg / L; (3) Vitamins: choline chloride 4 mg / L, D-calcium pantothenate 4 mg / L, folic acid 4 mg / L, nicotinamide 4 mg / L, pyridoxine hydrochloride 4 mg / L, riboflavin 0.4 mg / L, thiamine hydrochloride 4 mg / L, tocopherol acetate 0.07 mg / L; (4) Salts: calcium chloride 200 mg / L, magnesium sulfate 97.7 mg / L, potassium chloride 400 mg / L, sodium bicarbonate 3700 mg / L, sodium chloride 4750 mg / L, sodium dihydrogen phosphate 109 mg / L, disodium hydrogen phosphate 71 mg / L; (5) Trace elements: sodium selenite 100 mg / L, ferric nitrate 0.1 mg / L, copper sulfate 16 mg / L; (6) Buffer: 4-hydroxyethylpiperazine ethanesulfonic acid 5958 mg / L; (7) Acid-base indicator: Phenol red 15 mg / L; (8) Culture medium supplements: D-glucose 4500 mg / L, sodium pyruvate 110 mg / L, peptone 250 mg / L, bovine serum albumin 150 mg / L, transferrin 5.5 mg / L, insulin 10 mg / L, hypoxanthine 5 mg / L, ethanolamine 2 mg / L.

10. The use of a low-serum culture medium for MDCK cells in amplifying a virus or a viral vaccine, wherein the virus is an influenza virus.

Images