Real-time detection and dynamic feeding system and method for NK cell suspension culture nutrient solution

Abstract

The invention provides a real-time detection and dynamic feeding system and method for an NK cell suspension culture nutrient solution. The system comprises a multi-parameter sensing module used for monitoring the concentration of multiple components in a culture solution in real time; the intelligent decision control module is used for dynamically predicting nutritional requirements and generating a feeding strategy according to the multi-parameter data monitored in real time and based on related models and algorithms; the precise feed supplement execution module is used for injecting a nutrient solution and a pH adjusting solution as required through a multi-channel peristaltic pump according to the generated feed supplement strategy, supplementing growth factors at regular time, and is provided with a metabolic waste removal unit for removing lactic acid and ammonia metabolic waste; and the feedback optimization database is used for recording culture process data, optimizing a feeding strategy through a machine learning algorithm, and adapting to individual requirements of different cell strains. The technical problems that in the prior art, a static feeding strategy cannot respond to nutrition dynamic changes in real time, metabolic byproducts lack a real-time removal mechanism, and the cell quality is unstable due to nutrition imbalance are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cell culture, and particularly relates to a system and method for real-time detection and dynamic feeding of a nutrient solution for NK cell suspension culture. BACKGROUND

[0002] NK cells (Natural Killer cells) are important immune cells of the body and have important applications in immunotherapy, and their large-scale production relies on suspension culture technology.

[0003] In the existing process, the consumption of nutrient solutions (such as glucose, amino acids, growth factors, etc.) is calculated by pre-designing feeding, but there are the following defects: the existing NK cell culture mostly adopts a static feeding strategy, relies on an empirical model to pre-calculate the feeding amount, does not fully consider the metabolic differences in the cell growth stage, cannot respond to the dynamic changes of the nutrient solution in real time, and is prone to lead to nutrient deficiency in the later stage, affecting the cell growth, activity and proliferation rate; the metabolic waste such as lactic acid, ammonia and other metabolites of the existing NK cell culture will inhibit the cell activity, and the traditional method lacks an effective real-time removal mechanism; the unbalanced nutrient solution of the existing NK cell culture leads to an increase in cell phenotype heterogeneity and a decrease in killing activity, thereby causing unstable cell quality.

[0004] Therefore, there is an urgent need for an innovative system and method that can achieve precise detection and dynamic feeding of nutrient solutions to improve the efficiency and quality of NK cell culture and adapt to the individualized needs of different cell strains. SUMMARY

[0005] The purpose of the present application is to provide a system and method for real-time detection and dynamic feeding of a nutrient solution for NK cell suspension culture, which solves the technical problems of the prior art that the static feeding strategy cannot respond to the dynamic changes of the nutrient in real time, the metabolic by-products lack a real-time removal mechanism, and the cell quality is unstable due to unbalanced nutrition.

[0006] To solve the above technical problems, the technical solution of the present application is as follows:

[0007] 1. Multi-parameter sensing module: real-time monitoring and data calibration

[0008] 1.1 Online biosensor network

[0009] Monitoring parameters and sensor types:

[0010] Glucose: glucose oxidase electrode sensor (range 0-20 mM, accuracy ±0.1 mM) is used, and data is collected every 10 minutes.

[0011] Glutamine (an essential amino acid for NK cells): a probe based on fluorescence resonance energy transfer (FRET) (detection range 0-8 mM, resolution 0.2 mM) is used, and the reading is updated every 15 minutes.

[0012] Lactic acid / ammonia: Integrated electrochemical sensors (lactic acid range 0-15 mM, ammonia range 0-5 mM), measured every 20 minutes.

[0013] pH and dissolved oxygen (DO): pH electrode (accuracy ±0.05) and optical DO probe (range 0-100% air saturation), continuously monitored in real time.

[0014] Sensor layout: Three sets of sensor arrays were installed in a 5 L bioreactor, at the bottom, middle and below the liquid surface, respectively, to eliminate the error of concentration gradient.

[0015] 1.2 Off-line mass spectrometry verification and calibration

[0016] Sampling procedure: Micro-samplers (volume 0.5 mL / time, sterile filter pore size 0.22 μm) were used to automatically extract the culture solution every hour. After centrifugation (3000 rpm, 5 minutes) to remove cells, the samples were analyzed for the concentrations of 18 amino acids, vitamins and growth factors (such as IL-2, IL-15) by HPLC-MS system (Agilent 1290 / 6470). Calibration rule: if the deviation between the online sensor data and the mass spectrometry results is >10%, the sensor is automatically cleaned (using 0.1 M NaOH for 5 minutes) and recalibrated.

[0017] 2. Intelligent decision control module: dynamic feeding strategy generation

[0018] 2.1 Metabolic kinetic model construction

[0019] Input parameters: cell density (counted every 30 minutes by an online cell counter, such as Cedex HiRes, with an accuracy of ±5%). Nutrient consumption rate (such as glucose consumption rate Q glc = 0.5-1.2 mM / 10 6 cells / day). Metabolic product inhibition coefficient (such as lactic acid inhibition threshold > 8 mM, cell proliferation rate decreased by 50%).

[0020] Model calculation: based on the mass balance equation to predict the nutrient requirement in the next 2 hours:

[0021] C t+Δt = C t -(Q 消耗 × N 细胞 × Δt) / V + (F 补料 × C 补料 × Δt) / V

[0022] In the formula, C t+Δt is the predicted concentration of the nutrient solution after Δt hours (such as 2 hours); C tConcentration of current nutrient solution; Q 消耗 Nutrient consumption rate per cell; N 细胞 Cell count in culture system; Δt - time increment (e.g. 2 hours); V - volume of culture system; F 补料 Feed rate; C 补料 Nutrient concentration in feed. Dynamic adjustment of feed rate F 补 to maintain target nutrient concentration within optimal window (e.g. glucose 2-5 mM, glutamine 2-4 mM).

[0023] 2.2 Fuzzy PID control algorithm

[0024] Parameter settings: proportional coefficient K p = 0.8, integral time T i = 120 seconds, derivative time T d = 30 seconds.

[0025] Fuzzy rule base: define three-level membership functions for "low concentration", "medium concentration" and "high concentration", output feed pump speed (0-200 rpm).

[0026] Control logic: when glucose < 3 mM, initiate high-priority feed (glucose concentrate, 50 g / L), while reducing flux of lactate removal membrane (to avoid nutrient dilution). When glutamine < 2 mM and lactate > 6 mM, trigger combined feed (glutamine + buffer), and increase dialysate flow rate to 10 mL / min.

[0027] 3. Precise feed execution module: intervention on demand and waste removal

[0028] 3.1 Multi-channel peristaltic pump feed system

[0029] Feed solution types and parameters:

[0030] Basic nutrient solution: contains 10x concentrated glucose, amino acids (Eagle's MEM formulation), feed rate 0.05-0.2 mL / min (corresponding to glucose supplementation 0.5-2 mM / h).

[0031] Growth factors: IL-2 (1000 IU / mL) and IL-15 (50 ng / mL), pulsed feed (10 μL injection every 4 hours to avoid cell exhaustion from continuous stimulation).

[0032] pH adjustment solution: 7.5% NaHCO3 solution, injected in response to pH < 6.8 or > 7.4, flow rate 0.1 mL / min.

[0033] Pump control: High precision peristaltic pump (e.g. Watson-Marlow 323S) with flow error <±1%, equipped with anti-backflow valve.

[0034] 3.2 Metabolic waste removal unit:

[0035] Hollow fiber membrane dialyzer: membrane pore size 10 kDa, effective area 0.5 m 2 , dialysate PBS buffer (flow rate 5-15 mL / min); lactic acid removal efficiency >80% (inlet concentration 10 mM, outlet concentration <2 mM).

[0036] Ammonia adsorption column: packed with zeolite molecular sieve (pore size 0.5 nm), activated when ammonia concentration >3 mM, adsorption capacity >50 mg / g.

[0037] 4. Feedback optimization database: process self-adaptive learning

[0038] 4.1 Data acquisition and storage

[0039] Recorded parameters: time stamp, cell density, nutrient concentration, metabolite level, feed amount, environmental parameters (temperature 37±0.2℃, CO2 5%). About 100,000 data points were generated for each batch of culture and stored in SQL database (sampling interval 1 minute).

[0040] 4.2 Machine learning optimization

[0041] Model training: using random forest algorithm (Python scikit-learn library), input historical data to predict the optimal feeding schedule.

[0042] Feature engineering: extract the slope of nutrient consumption curve, cell growth inflection point time, metabolite accumulation rate.

[0043] Output results: for NK-92 cell line, the optimized feeding frequency was reduced by 20%, while the primary NK cells required 3 times more IL-15 feeding.

[0044] Process flow chart and key parameter summary

[0045]

[0046] Implementation effect verification

[0047] Proliferation curve comparison: the dynamic feeding group reached a cell density of 8×10 6 cells / mL at day 7, while the static feeding group only reached 4×10 6 cells / mL, with significant difference p<0.01 (t-test).

[0048] Cell activity indicators: CD107a degranulation rate increased to 65% (control group 45%), IFN-γ secretion increased by 40%.

[0049] Through the above step-by-step process and parameter configuration, the system can realize precise regulation of the NK cell culture process, and provide a standardized solution for industrial production.

[0050] Compared with the prior art, the present application at least includes the following beneficial effects:

[0051] 1. The present application realizes real-time and accurate monitoring of multiple components in the culture medium by integrating a multi-parameter sensing module and offline mass spectrometry calibration; based on a metabolic kinetics model and a fuzzy PID control algorithm, the nutritional requirements are dynamically predicted and personalized feeding strategies are generated to ensure that the nutrient concentration is always maintained within the optimal window.

[0052] 2. The present application introduces a hollow fiber membrane dialyzer and an ammonia adsorption column to selectively remove lactic acid and ammonia, with removal efficiencies of >80% and >90%, respectively. Through the linkage control of feeding and waste removal, the dilution of nutrients and metabolic inhibition are avoided, and the cell proliferation efficiency and functional activity are significantly improved.

[0053] 3. The present application uses machine learning algorithms (such as random forest model) to optimize the feeding strategy, which adapts to the individual needs of different cell lines (such as NK-92 and primary NK cells), and improves the process stability and repeatability.

[0054] 4. The system uses modular design, integrating sensing, decision-making, feeding, and removal functions in one, supporting fully automated control; at the same time, it is configured with an abnormality handling mechanism (such as automatic cleaning of sensors and switching of feeding pipeline blockage), ensuring process continuity and reliability; thereby significantly improving the NK cell expansion efficiency, greatly improving the cell quality and functional activity, reducing production costs and resource waste, adapting to various cell lines and culture conditions, and meeting the GMP standard to support industrial application. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0056] Figure 1 The structure and principle diagram of the detection and feeding system and method provided by the present application;

[0057] Figure 2 The value-added effect comparison diagram of the dynamic feeding process of the present application and the traditional static feeding process.

[0058] Figure 3 The comparison chart of the removal efficiency of the waste metabolism removal process of the present application and the traditional process. DETAILED DESCRIPTION

[0059] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0060] Some embodiments and specific implementation solutions of the present application will be described in detail below with reference to the drawings.

[0061] Embodiment one:

[0062] NK-92 cells were cultured in a 5L bioreactor with an initial density of 1x10 6 cells / mL.

[0063] When the system detected that the glucose concentration decreased to 3mM, the feeding pump was triggered to inject high-sugar medium (feeding rate 0.1mL / min), and the dialysis membrane was started to remove lactic acid.

[0064] Results: The cell density reached 1.2x10 7 cells / mL (6x10 6 cells / mL for the traditional process) on the 10th day of culture. The cytotoxicity experiment showed that the killing activity was increased by 25%.

[0065] Embodiment two:

[0066] For primary NK cells, the system automatically identified glutamine as a limiting factor (threshold <2mM), preferentially fed and increased the IL-15 concentration to 10ng / mL.

[0067] Results: The expansion rate of primary cells was increased by 3 times, and the stability of CD16 expression was enhanced.

[0068] Specific implementation solutions

[0069] The present application proposes a real-time detection and dynamic feeding system and method for NK cell suspension culture nutrient solution, as shown in Figure 1 , which is realized by the following technical solutions:

[0070] I. System initialization and culture preparation

[0071] 1. Device configuration and installation

[0072] Bioreactor: 5L disposable bioreactor (e.g. Sartorius CultiBag RM) equipped with three sets of sensor arrays (bottom, middle, 5cm below liquid surface).

[0073] Sensor calibration:

[0074] Glucose sensor: Three-point calibration using standard solutions (0mM, 5mM, 10mM glucose) with error < ±0.1mM.

[0075] pH electrode: Calibrated at 37°C with pH = 4.0, 7.0, 10.0 buffers ensuring accuracy ±0.05.

[0076] Feed module: Connected to a four-channel peristaltic pump (Watson-Marlow 323S) loaded with:

[0077] Channel 1: 10x concentrated basal nutrient solution (containing glucose 50g / L, essential amino acids 10x concentration).

[0078] Channel 2: Growth factor mix (IL-21 1000IU / mL + IL-15 50ng / mL).

[0079] Channel 3: pH adjustment solution (7.5% NaHCO3).

[0080] Channel 4: Dialysis buffer (PBS, pH 7.2).

[0081] 2. Initial culture condition setup

[0082] Cell inoculation: NK-92 cells were inoculated at a density of 1x10 6 cells / mL in RPMI 1640 medium containing 10% FBS, 1% penicillin-streptomycin.

[0083] Environmental parameters: Temperature 37.0 ± 0.1°C, dissolved oxygen (DO) 50% air saturation, CO2 concentration 5%, agitation speed 60rpm (to avoid shear damage).

[0084] II. Real-time monitoring and data calibration

[0085] 1. Online sensor data acquisition

[0086] Monitoring frequency:

[0087] Glucose, glutamine: Updated every 10 minutes.

[0088] Lactate, ammonia: Measured every 20 minutes.

[0089] pH, DO: Real-time continuous monitoring (sampling every second).

[0090] Abnormality detection: If glucose readings fluctuate >10% for 3 times in a row, trigger sensor auto-washing (0.1M NaOH rinse for 5 minutes, sterile PBS rinse for 3 times).

[0091] 2. Off-line mass spectrometry verification

[0092] Sampling and processing:

[0093] 0.5mL culture broth was withdrawn by sterile sampler every hour, centrifuged at 3000rpm for 5 minutes, and the supernatant was taken.

[0094] HPLC-MS (Agilent 1290 / 6470) was used to analyze the concentrations of amino acids, vitamins, and IL-15, and the data was compared with the sensor data.

[0095] Calibration trigger condition: If the glucose sensor deviates from the mass spectrometry data by >10%, the system pauses feeding and recalibrates the sensor.

[0096] III. Dynamic feeding execution

[0097] 1. Intelligent decision-making control flow

[0098] Metabolic model input:

[0099] Cell density data was obtained every 30 minutes (Cedex HiRes cell counter).

[0100] Real-time nutrient consumption rate was calculated (e.g., glucose Q glc = (C 初始 -C 当前 ) / (cell number x time)).

[0101] Feeding trigger logic:

[0102] Priority 1: Glucose <3mM → Start channel 1 feeding at a rate of 0.1mL / min (corresponding to glucose supplementation of 1mM / h).

[0103] Priority 2: Glutamine <2mM → Start channel 1 feeding (containing glutamine) at a rate of 0.05mL / min.

[0104] Priority 3: Lactate >8mM → Start channel 4 dialysate (flow rate 10mL / min), simultaneously reduce feeding rate by 20%.

[0105] 2. Feeding solution injection and regulation

[0106] Basic nutrient feeding:

[0107] Pump speed control: dynamically adjusted according to fuzzy PID control algorithm (e.g. 15 min to increase glucose concentration from 3 mM to 4 mM, feed rate 0.12 mL / min).

[0108] Anti-backflow design: pump reversed for 0.5 s (pressure -0.1 bar) after each feeding to prevent residual in the pipeline.

[0109] Growth factor pulse feeding: 10 μL growth factor mixture (IL-21 00 IU / mL + IL-15 50 ng / mL) injected every 4 hours to avoid over-activation of cell receptors.

[0110] Four, metabolic waste removal

[0111] 1. Membrane dialysis and adsorption combination

[0112] Hollow fiber membrane operation: when lactate > 6 mM, start dialysate circulation (PBS, flow rate 8 mL / min), membrane surface area 0.5 m 2 , removal efficiency > 80%; dialysate temperature maintained at 37°C to avoid fluctuations in culture medium temperature.

[0113] Ammonia adsorption column activation: when ammonia concentration > 3 mM, switch to zeolite molecular sieve adsorption column (adsorption capacity 50 mg / g), backwash every 2 hours (flow rate 5 mL / min for 2 min).

[0114] Five, data recording and feedback optimization

[0115] 1. Real-time database construction

[0116] Stored parameters: time stamp, cell density, nutrient concentration, feeding amount, metabolite level, environmental parameters (temperature, DO, pH).

[0117] Data sampling interval: 1 minute / time, stored in SQL database (total data volume about 100,000 pieces / batch).

[0118] 2. Machine learning optimization strategy

[0119] Model training: use historical data to train random forest model (Python scikit-learn, n_estimators = 100) to predict the optimal feeding time.

[0120] Input features: glucose consumption rate, lactate accumulation slope, cell growth inflection point time.

[0121] Dynamic adjustment: update model parameters and optimize feeding strategy every 3 batches (e.g. NK-92 cell feeding frequency reduced by 15%).

[0122] Six, verification of implementation effect

[0123] 1. Proliferation Efficiency Detection

[0124] Cell counting: Samples were taken every day and viable cell density was calculated by trypan blue staining.

[0125] Example of result: Figure 2 As shown in the figure, the density of the dynamic feeding group reached 8×10 on the 7th day. 6 cells / mL, the traditional group only had 4×10 6 (p<0.01, t-test).

[0126] 2. Cell Function Assessment

[0127] Cytotoxicity: The cells were co-cultured with K562 cells (effector-target ratio 10:1) and CD107a expression was detected (dynamic group 65% vs traditional group 45%).

[0128] Cell phenotype: Flow cytometry was used to detect the proportion of CD56+CD16+ subpopulations (dynamic group 85% vs traditional group 70%).

[0129] 3. Metabolite Control

[0130] Lactate concentration: dynamic group maintained <6mM, traditional group >12mM (day 7).

[0131] Ammonia concentration: dynamic group <2.5mM, traditional group >4mM.

[0132] The lactic acid removal efficiency of the present invention is 85%, and the ammonia removal efficiency is 92%. Figure 3 shown.

[0133] 7. Exception handling and maintenance:

[0134] 1. Sensor failure response

[0135] If the sensor fails to calibrate three times in a row, it switches to the backup sensor and triggers a manual inspection instruction.

[0136] 2. Treatment of feed blockage

[0137] When the pump pressure is greater than 1.5 bar, it automatically switches to the backup line and sends an alarm to remind you to clean the filter (0.22μm filter element).

[0138] 8. Summary

[0139] 1. Summary of the operation of the embodiment:

[0140] Taking NK-92 cell culture as an example:

[0141] Day 0: System initialization, sensor calibration, and cell seeding.

[0142] Day 1-3: Real-time monitoring shows stable glucose consumption (0.8 mM / 10 6 cells / day), feeding 0.1 mL / min every 6 hours.

[0143] Day 4: Lactate rises to 7 mM, start dialysis (flow rate 10 mL / min), adjust feeding rate to 0.08 mL / min synchronously.

[0144] Day 7: Cell density reaches 8 x 10 6 cells / mL, harvest cells and verify activity (CD107a 65%).

[0145] Through the above step-by-step implementation process, the system can realize the full automation control of NK cell culture, significantly improve the expansion efficiency and product quality, and is suitable for GMP-level cell therapy product production.

[0146] 2. Technology and effect summary

[0147] (1) Significantly improve NK cell expansion efficiency

[0148] Dynamic feeding extends the logarithmic growth phase of NK cells by 30%-50%, and increases the final yield by 1.5-2 times (traditional process density 4 x 10 6 cells / mL, this patent 8 x 10 6 cells / mL); the expansion ratio of primary NK cells is increased by 3 times, meeting the high demand for cell quantity in clinical treatment; providing sufficient cell sources for advanced therapies such as CAR-NK and NK-T, accelerating the industrialization process.

[0149] (2) Greatly improve cell quality and functional activity

[0150] The concentration of metabolic waste is reduced by 60% (lactate <6 mM, ammonia <2.5 mM), and cell activity is significantly improved; the proportion of CD56+CD16+ functional subpopulation is increased to more than 85%, CD107a degranulation rate is increased to 65% (traditional process 45%), IFN-γ secretion is increased by 40%; ensuring the consistency and effectiveness of cell products, improving the clinical treatment effect.

[0151] (3) Reduce production cost and resource waste

[0152] Dynamic feeding reduces nutrient waste, saving about 20% of the cost; automatic control reduces the frequency of manual intervention, improves production efficiency; provides an economically feasible solution for the large-scale production of cell therapy products.

[0153] (4) Adapt to various cell strains and culture conditions

[0154] The system supports the culture of multiple cell lines such as NK-92 and primary NK cells, and optimizes personalized feeding strategies through machine learning; it can be extended to the culture process optimization of other immune cells (such as T cells and CIK cells); it provides a universal technology platform for the field of immune cell therapy and promotes the progress of the industry.

[0155] (5) Comply with GMP standards to support industrial applications

[0156] The system design complies with GMP specifications, supports sterile operation and data traceability; the modular structure facilitates scale-up (e.g., from a 5L reactor to a 200L bioreactor); and it provides reliable technical support for the clinical conversion and commercial production of cell therapy products.

[0157] Through the above step-by-step implementation process, the system can achieve full automation control of NK cell culture, significantly improving the amplification efficiency and product quality, and is suitable for GMP-level cell therapy product production.

[0158] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time detection and dynamic feeding system for NK cell suspension culture nutrient solution, characterized in that: include: Multi-parameter sensing module: used for real-time monitoring of the concentrations of various components of the basic nutrient solution, lactic acid, ammonia, pH and dissolved oxygen in the culture medium; Intelligent decision-making control module: used to dynamically predict nutrient requirements and generate feeding strategies based on real-time monitoring of multi-parameter data of culture medium, relevant models and algorithms; The precision feeding execution module is used to inject basal nutrient solution and pH adjustment solution as needed through a multi-channel peristaltic pump according to the generated feeding strategy, and regularly replenish growth factors. At the same time, this module is equipped with a metabolic waste removal unit to remove metabolic wastes such as lactic acid and ammonia. Feedback optimization database: used to record culture process data and optimize feeding strategies through machine learning algorithms to adapt to the personalized needs of different cell lines.

2. The NK cell suspension culture nutrient solution real-time detection and dynamic feeding system according to claim 1, characterized in that: The multi-parameter sensing module is equipped with an offline mass spectrometry calibration function. By comparing the mass spectrometry results with the online sensor data, when the deviation is greater than a preset value, the sensor is triggered to be cleaned and recalibrated to ensure the accuracy of the monitoring data.

3. The NK cell suspension culture nutrient solution real-time detection and dynamic feeding system according to claim 1, characterized in that: The intelligent decision-making control module dynamically predicts nutritional needs and generates feeding strategies based on a metabolic kinetic model and a fuzzy PID control algorithm.

4. The NK cell suspension culture nutrient solution real-time detection and dynamic feeding system according to claim 3, characterized in that: The metabolic kinetic model was constructed based on the mass balance equation: C t+Δt =C t -(Q 消耗 ×N 细胞 ×Δt) / V+(F 补料 ×C 补料 ×Δt) / V Among them, C t+Δt To predict the concentration of the nutrient solution after Δt hours, C t is the current concentration of the nutrient solution, Q 消耗 is the nutrient consumption rate per unit cell, N 细胞 is the cell count in the culture system, Δt is the time increment, V is the volume of the culture system, F 补料 is the feeding rate, C 补料 is the nutrient concentration in the feed.

5. The NK cell suspension culture nutrient solution real-time detection and dynamic feeding system according to claim 3, characterized in that: The proportional coefficient K of the fuzzy PID control algorithm p =0.8, integration time T i =120 seconds, differential time T d =30 seconds.

6. The NK cell suspension culture nutrient solution real-time detection and dynamic feeding system according to claim 1, characterized in that: The metabolic waste removal unit uses a hollow fiber membrane dialyzer and an ammonia adsorption column to remove metabolic waste.

7. The NK cell suspension culture nutrient solution real-time detection and dynamic feeding system according to claim 6, characterized in that: The hollow fiber membrane dialyzer has a membrane pore size of 10 kDa and an effective area of ​​0.5 m 2 , the dialysate flow rate is 5-15 mL / min; The ammonia adsorption column is filled with zeolite molecular sieve with a pore size of 0.5 nm and an adsorption capacity of ≥50 mg / g.

8. The NK cell suspension culture nutrient solution real-time detection and dynamic feeding system according to claim 1, characterized in that: The machine learning algorithm uses a random forest model, with input features including nutrient consumption rate, lactate accumulation slope and cell growth inflection point time, and outputs the optimal feeding schedule. Moreover, the model parameters are updated after each batch of culture is completed to optimize the feeding strategy.

9. A method for real-time detection and dynamic feeding of nutrient solution for NK cell suspension culture, using the system of claims 1-8, characterized in that: The following steps are involved: Inoculate cells at a certain density, culture them with the initial culture medium, and then add culture medium; Real-time monitoring of the concentrations of basic nutrient solution components, lactic acid, ammonia, pH and dissolved oxygen in the culture medium; Compare the online sensor data with the mass spectrometry results. When the deviation is greater than the preset value, the sensor is cleaned and recalibrated. Based on the real-time monitoring of multi-parameter data of culture medium, metabolic kinetic model and fuzzy PID control algorithm, dynamic prediction of nutrient demand and generation of feeding strategy; According to the generated feeding strategy, the basic nutrient solution and pH adjustment solution are injected as needed through a multi-channel peristaltic pump; Pulse-wise quantitative injection of growth factor mixture every preset time; Remove lactic acid or ammonia metabolic waste by activating a hollow fiber membrane dialyzer or ammonia adsorption column; After culturing for the preset days, the cells were harvested when the cell density reached the target density.

10. The method for real-time detection and dynamic feeding of NK cell suspension culture nutrient solution according to claim 9, characterized in that: The following steps are also included: After each batch of cultivation is completed, the cultivation data is entered into the feedback optimization database; Use the random forest algorithm to train metabolic kinetic models and optimize feeding timing and amount to meet the personalized needs of different cell lines.

Images