A method of preparing a product for monkeypox and orthopoxvirus detection and a system
Patent Information
- Application Number
- CN202610711614.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-22
- Publication Date
- 2026-09-15
Smart Images
Figure CN122750893A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of preparation technology for monkeypox and orthopoxvirus detection products, and particularly to a method and system for preparing products for monkeypox and orthopoxvirus detection. Background Technology
[0002] Monkeypox virus belongs to the genus Orthopoxvirus in the family Poxviridae. The preparation of monkeypox and orthopox virus detection products (including fluorescent PCR lyophilized microchip kits, RPA-LFA integrated POCT detection devices, isothermal amplification microfluidic chips, etc.) involves multiple key process steps. The core preparation process for nucleic acid detection reagents includes primer and probe synthesis and purification, lyophilization reaction solution preparation, micro-dispensing, and lyophilization packaging. Immunological detection products involve nitrocellulose membrane application, conjugate pad / sample pad treatment, and colloidal gold labeling. Integrated POCT devices also include complex processes such as microfluidic channel molding, nucleic acid release reagent pre-loading, and RPA reaction ball positioning. These preparation processes involve numerous interconnected key process parameters (KPPs), such as the pre-freezing cooling rate, primary drying temperature, desorption drying heating rate, and vacuum degree change curve in the lyophilization process, and Mg²⁺ in the nucleic acid amplification reaction solution preparation. + Concentration, primer-probe molar ratio, dNTP concentration, buffer pH, and isothermal amplification reaction temperature and time are all important parameters. These parameters not only directly affect the limit of detection, intra-batch / inter-batch coefficient of variation, specificity score, and other key quality attributes (CQA) of the detection reagent, but also exhibit significant nonlinear coupling relationships. For example, the lyophilization vacuum degree and shelf temperature jointly affect the ice crystal sublimation rate, which in turn determines the residual moisture content and long-term stability of the active ingredient.
[0003] The industry primarily relies on two methods to optimize the preparation process parameters of monkeypox and orthopoxvirus detection products: First, offline statistical analysis based on Design of Experiments (DOE), which involves preparing multiple batches of experimental samples, testing their quality indicators separately, and then using regression analysis to find the optimal parameter combination. While this method has some statistical basis, it requires numerous experimental rounds and a long cycle (usually several weeks to months), and it is difficult to reveal the global optimum in high-dimensional parameter space. Especially when the number of parameters exceeds five, the sample size of the full-factor experiment grows exponentially, making it impractical for engineering. Summary of the Invention
[0004] This invention overcomes the shortcomings of the prior art and provides a method and system for preparing products for detecting monkeypox and orthopoxvirus.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for preparing a product for detecting monkeypox and orthopoxvirus, comprising the following steps: Key process parameter data were collected throughout the entire preparation process of monkeypox and orthopoxvirus, and a training set was constructed based on the key process parameter data. A process parameter-quality mapping model is constructed based on a deep learning network, and the process parameter-quality mapping model is trained using the training set. Construct a deep reinforcement learning agent and use it to optimize and recommend the process parameter configuration for each current batch. The process drift detection is carried out by continuously monitoring process parameters using an intelligent agent, and adaptive correction is performed based on the process drift detection results.
[0006] Furthermore, in the preparation method of products for detecting monkeypox and orthopoxvirus, the entire process of preparing monkeypox and orthopoxvirus includes purification of recombinant adenovirus, determination of the optimal dilution of recombinant adenovirus, positive control analysis, identification of recombinant adenovirus shuttle plasmid, synthesis of target gene sequence and primers, construction of recombinant adenovirus shuttle plasmid, packaging of recombinant adenovirus particles, passage amplification and identification of recombinant adenovirus, freeze drying, preparation of nucleic acid amplification reaction solution, and preparation of sample nucleic acid release reagent.
[0007] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, key process parameter data are collected throughout the entire preparation process of monkeypox and orthopoxvirus, and a training set is constructed based on the key process parameter data, specifically as follows: Key process parameter data were collected throughout the entire product preparation process, including: pre-freezing temperature and pre-freezing time in the freeze-drying process, temperature rise curve of the partition in the desorption drying stage, vacuum degree change curve, preparation parameters of nucleic acid amplification reaction solution, isothermal amplification reaction temperature and time, and formulation parameters of sample nucleic acid release reagent. The pre-freezing temperature, pre-freezing time, partition temperature rise curve, vacuum degree change curve, nucleic acid amplification reaction solution preparation parameters, isothermal amplification reaction temperature and time, and sample nucleic acid release reagent formulation parameters of the drying process are collected in real time and the corresponding product quality test results are labeled. A training dataset is constructed based on the pre-freezing temperature, pre-freezing time, partition temperature rise curve during the analytical drying stage, vacuum degree change curve, nucleic acid amplification reaction solution preparation parameters, isothermal amplification reaction temperature and time, sample nucleic acid release reagent formulation parameters, and product quality test results in the aforementioned drying process.
[0008] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a process parameter-quality mapping model is constructed based on a deep learning network, and the process parameter-quality mapping model is trained using the training set, specifically as follows: The process parameter space and quality index space are constructed using the training dataset. A multi-layer deep neural network is used to construct a mapping model from the process parameter space to the quality index space, forming a process parameter-quality mapping model. The pre-freezing temperature, primary drying temperature, analytical drying temperature, time parameters, and pressure parameters at each stage of the freeze-drying process are encoded as the first input feature vector, and the annealing temperature of primers and probes, the number of amplification cycles, and the fluorescence collection time point in the amplification reaction system are encoded as the second input feature vector. The ionic strength and pH value of the nucleic acid release reagent are encoded as the third input feature vector. The first, second, and third input feature vectors are combined in a high-order nonlinear manner and then input into a deep neural network. The lowest detection limit (LOD), coefficient of variation (CV), and specificity score (SS) of the deep neural network are used to form predicted values of quality indicators, and a trained process parameter-quality mapping model is obtained.
[0009] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a deep reinforcement learning agent is constructed, specifically as follows: The trained process parameter-quality mapping model is used as an environment simulator for reinforcement learning to construct a process parameter optimization agent based on deep Q-network or proximal policy optimization algorithm. The state space of the agent is defined as the sequence of process parameter vectors for each process in the current batch. The agent's reward function is designed based on the output of the quality prediction model. It gives positive rewards for low detection limits, low coefficient of variation, and high specificity, and imposes penalty terms on abnormal parameter combinations.
[0010] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a deep reinforcement learning agent is used to optimize and recommend the process parameter configuration for each current batch, specifically as follows: In the production line control system, a trained reinforcement learning agent is deployed. The process parameter vector sequence of each process in the current batch is input, and the trained reinforcement learning agent is used to analyze the adjustment amount of key process parameters based on the process parameter vector sequence of each process in the current batch. By analyzing and optimizing the process parameter configuration for each new batch, we continuously explore better combinations of process parameters while ensuring production line stability, and generate optimization results. The optimization results are directly sent to the production line equipment for execution via the programmable logic controller interface, while retaining a manual review interface to allow process engineers to review and confirm the recommended parameters of the intelligent agent.
[0011] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a smart agent is used to continuously monitor process drift of process parameters, and adaptive correction is performed based on the process drift detection results. Specifically: When the production line is under statistical control, a preset number of normal process parameter time series data are collected. Based on the normal process parameter time series data, the multivariate residual vector at each moment is calculated, and the covariance matrix of the residuals is estimated. Calculate the Mahalanobis distance between the residual vector and the vectors in the covariance matrix, introduce the adaptive offset factor and dynamic weight factor, calculate the positive cumulative statistic and the negative cumulative statistic, and calculate the adaptive offset factor. When the positive cumulative statistic is greater than a preset positive threshold or the negative cumulative statistic is lower than a preset negative cumulative threshold, a systematic drift in the process is determined. Furthermore, contribution plot analysis is used to identify the main process parameters causing the drift. The current cumulative residual vector is projected onto each parameter dimension, and the relative contribution percentage of each parameter is calculated. When the contribution percentage of a certain parameter exceeds the threshold for three consecutive sampling points, the parameter is identified as the root cause parameter of drift and used as the process drift detection result. Adaptive correction and compensation are then performed using an intelligent agent.
[0012] A second aspect of the present invention provides a preparation system for a product for detecting monkeypox and orthopoxvirus, comprising a memory and a processor. The memory includes a method program for preparing the product for detecting monkeypox and orthopoxvirus. When the method program for preparing the product for detecting monkeypox and orthopoxvirus is executed by the processor, it implements the steps of the method for preparing the product for detecting monkeypox and orthopoxvirus as described in any one of the present invention.
[0013] This invention addresses the shortcomings of the prior art and has the following beneficial effects: This invention addresses the problem that existing monkeypox detection product process optimization mainly relies on Design of Experiments (DOE) and offline data statistical analysis, resulting in long optimization cycles and difficulty in capturing complex nonlinear coupling relationships between parameters. This solution applies deep reinforcement learning to the real-time dynamic tuning of process parameters in orthopoxvirus detection product preparation. By accelerating the exploration-utilization process through an environment simulator, it significantly shortens the process optimization iteration cycle and solves the technical problem of obtaining the global optimal solution in multi-parameter coupled systems. Attached Figure Description
[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.
[0015] Figure 1 A flowchart illustrating the preparation method of products for detecting monkeypox and orthopoxvirus is shown. Figure 2 A system block diagram of a preparation system for products used in the detection of monkeypox and orthopoxvirus is shown. Detailed Implementation
[0016] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.
[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.
[0018] like Figure 1 As shown, the first aspect of the present invention provides a method for preparing a product for detecting monkeypox and orthopoxvirus, comprising the following steps: Key process parameter data were collected throughout the entire preparation process of monkeypox and orthopoxvirus, and a training set was constructed based on the key process parameter data. A process parameter-quality mapping model is constructed based on a deep learning network, and the process parameter-quality mapping model is trained using the training set. Construct a deep reinforcement learning agent and use it to optimize and recommend the process parameter configuration for each current batch. The process drift detection is carried out by continuously monitoring process parameters using an intelligent agent, and adaptive correction is performed based on the process drift detection results.
[0019] It should be noted that this invention addresses the problem that existing monkeypox detection product process optimization mainly relies on Design of Experiments (DOE) and offline data statistical analysis, resulting in long optimization cycles and difficulty in capturing complex nonlinear coupling relationships between parameters. This solution applies deep reinforcement learning to the real-time dynamic tuning of process parameters for orthopoxvirus detection product preparation. By accelerating the exploration-utilization process through an environment simulator, it significantly shortens the process optimization iteration cycle and solves the technical problem of obtaining the global optimal solution in multi-parameter coupled systems.
[0020] Furthermore, in the preparation method of products for detecting monkeypox and orthopoxvirus, the entire process of preparing monkeypox and orthopoxvirus includes purification of recombinant adenovirus, determination of the optimal dilution of recombinant adenovirus, positive control analysis, identification of recombinant adenovirus shuttle plasmid, synthesis of target gene sequence and primers, construction of recombinant adenovirus shuttle plasmid, packaging of recombinant adenovirus particles, passage amplification and identification of recombinant adenovirus, freeze drying, preparation of nucleic acid amplification reaction solution, and preparation of sample nucleic acid release reagent.
[0021] It should be noted that the recombinant adenovirus shuttle plasmid was constructed by double digestion of pUC57-MPXV and adenovirus shuttle plasmid pDC316-mCMV-EGFP with restriction endonucleases EcoRV and HindIII, respectively. The digestion products were then recovered using a DNA gel extraction kit after agarose gel electrophoresis to obtain the purified target gene fragment and linearized vector. The target gene fragment and linearized vector were ligated with T4 DNA Ligase and transformed into DH5α supercompetent cells. Both the extracted recombinant plasmid and the shuttle plasmid pDC316-mCMV-EGFP without the target gene fragment were verified by double digestion with EcoRV and HindIII. Correctly identified recombinant plasmids were sent to Megamicro for sequencing. The recombinant plasmid with correct sequencing was named pDC316-MPXV.
[0022] Passaging, amplification, and identification of recombinant adenovirus: 1 mL P0 generation recombinant adenovirus was inoculated into a T25 cell culture flask with approximately 90% confluence. The cells were cultured at 37°C in a 5% CO2 incubator. When most cells showed typical CPE and green fluorescence, and approximately 50% of the cells detached, the supernatant and cells were collected. The cells were subjected to three freeze-thaw cycles at -80°C, centrifuged at 2000 rpm for 10 min, and the supernatant was stored at -80°C. This was the P1 generation recombinant adenovirus. Passaging and propagation using this method yielded a large quantity of recombinant adenovirus. Nucleic acid was extracted from the P5 generation recombinant adenovirus and identified using monkeypox virus real-time fluorescence PCR. Simultaneously, conventional PCR amplification and sequencing were performed using synthesized primers for amplifying the inserted gene fragment.
[0023] The recombinant adenovirus was purified according to the adenovirus purification kit (Adeno-X MaxiPurification Kit instructions). The supernatant of the P6 generation recombinant adenovirus was treated with Benzonase nuclease to digest the nucleic acid and clarified by passing it through a pre-filter at the tip of a syringe. The virus-containing liquid was then pushed through the purification column using a syringe, and the adenovirus particles bound to the filter in the purification column. The bound virus particles were then eluted with a small amount of buffer.
[0024] Determination of the optimal dilution for recombinant adenovirus: The purified recombinant adenovirus was serially diluted 10× using storage buffer (10 mM Tris-HCl pH 8.0, 2 mM MgCl2, 4% sucrose). Nucleic acid was extracted, and each dilution was performed in triplicate. The results were analyzed using the monkeypox virus fluorescent PCR detection method established by our research group to determine the optimal dilution for a positive control. The recombinant adenovirus was diluted to the optimal dilution and aliquoted into 1 mL tubes, which were stored at -80°C as a positive control.
[0025] After the recombinant adenovirus was amplified to generation P5, nucleic acid was extracted from the collected viral fluid and detected and identified using monkeypox virus real-time fluorescence PCR. The P5 generation recombinant adenovirus nucleic acid showed a typical fluorescence amplification curve, while the control group showed no amplification curve. Conventional PCR amplification was performed using primers for amplifying the inserted gene fragment. The results showed a specific band of approximately 831 bp in size, consistent with the size of the inserted target gene fragment, while the control group showed no amplification band. Sequencing results showed that the amplified sequence was consistent with the inserted sequence, indicating that the recombinant adenovirus containing the target gene was correctly constructed.
[0026] The preparation of positive control standards includes procedures for detecting the purification results of recombinant adenovirus and procedures for detecting the optimal dilution of recombinant adenovirus.
[0027] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, key process parameter data are collected throughout the entire preparation process of monkeypox and orthopoxvirus, and a training set is constructed based on the key process parameter data, specifically as follows: It should be noted that key process parameter data are collected throughout the entire product preparation process, including but not limited to: pre-freezing temperature (-80℃ to -96℃) and pre-freezing time (1-1.2h) in the freeze-drying process; the temperature rise curve of the separator in the desorption drying stage (gradual change from -25℃ to 37℃); and the vacuum degree change curve (0.1-0.5mbar); and the preparation parameters of the nucleic acid amplification reaction solution (Mg²). + The concentrations of the reagents are as follows: 1.5-4.0 mM for the reagents, 0.2-0.5 mM for the dNTPs, and the primer-probe ratio; the isothermal amplification reaction temperature and time (usually 60-65℃, amplification time 15-30 min); and the formulation parameters of the sample nucleic acid release reagent (lysis buffer salt ion concentration, pH 6.5-8.5). These data are collected in real-time through an IoT sensing layer and labeled with the corresponding product quality testing results (limit of detection, intra-batch coefficient of variation, specificity score, etc.) to form a training dataset.
[0028] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a process parameter-quality mapping model is constructed based on a deep learning network, and the process parameter-quality mapping model is trained using the training set, specifically as follows: A multi-layer deep neural network was used to construct a mapping model from the process parameter space to the quality index space. Specifically, the three-stage temperature parameters (pre-freezing temperature, primary drying temperature, and analytical drying temperature), time parameters, and pressure parameters of each stage of the freeze-drying process were encoded as the first input feature vector; the annealing temperature of the primers and probes, the number of amplification cycles, and the fluorescence acquisition time point in the amplification reaction system were encoded as the second input feature vector; and the ionic strength and pH value of the nucleic acid release reagent were encoded as the third input feature vector. These three feature vectors were combined nonlinearly through a feature cross-layer and then input into the deep neural network to output the predicted values of the quality indicators, including the limit of detection (LOD), coefficient of variation (CV), and specificity score (SS). The network adopted a multi-task learning architecture, with each quality index task sharing the underlying feature representation layer while maintaining its own independent prediction head. During training, a quality inspection dataset containing different samples, such as the West African branch and the Congo Basin branch of monkeypox virus, was used to ensure the model's generalization ability to detect different clades.
[0029] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a deep reinforcement learning agent is constructed, specifically as follows: The trained process parameter-quality mapping model is used as an environment simulator for reinforcement learning to construct a process parameter optimization agent based on deep Q-network or proximal policy optimization algorithm. The state space of the agent is defined as the sequence of process parameter vectors for each process in the current batch. The agent's reward function is designed based on the output of the quality prediction model. It gives positive rewards for low detection limits, low coefficient of variation, and high specificity, and imposes penalty terms on abnormal parameter combinations.
[0030] It should be noted that the trained quality prediction model is used as an environment simulator for reinforcement learning to construct a process parameter optimization agent based on Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) algorithms. The agent's state space is defined as the sequence of process parameter vectors for each step in the current batch; the action space is defined as the adjustment amount of key process parameters, such as the freeze-drying temperature being continuously adjustable within the range of -25℃ to 37℃ with a step accuracy of 0.5℃; the primer annealing temperature being finely adjusted in 0.2℃ steps within the range of 55℃ to 65℃; and the pH value of the nucleic acid release reagent being adjusted in 0.1-step steps. The reward function is designed based on the output of the quality prediction model, providing positive rewards for low detection limits, low coefficients of variation, and high specificity, and imposing penalties for abnormal parameter combinations. The formula for the reward function is as follows:
[0031] R represents the immediate reward value obtained by the reinforcement learning agent after performing an action in the current state; K represents the total number of categories of the product critical quality attributes (CQA) being studied; k is the index variable of CQA. The weight coefficient for the k-th CQA; The value of the k-th CQA predicted by the deep reinforcement learning environment simulator; Set the target for the k-th CQA; λ This is the motion amplitude penalty coefficient, dimensionless, with a value ranging from 0.01 to 0.1. It is used to balance mass recovery and actuator motion cost, suppressing excessive compensation motion. Let R be the compensation action vector output by the agent at time t. As the predicted CQA value gets closer to the target value, the relative bias term approaches 0, and the reward value R approaches 0.
[0032] It should be noted that this method can solve the problem that key process parameters such as freeze-drying curves, amplification reaction temperatures, and nucleic acid release efficiency rely on manual experience for optimization and that multi-parameter coupling optimization is inefficient in the preparation of monkeypox and orthopox virus detection products.
[0033] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a deep reinforcement learning agent is used to optimize and recommend the process parameter configuration for each current batch, specifically as follows: In the production line control system, a trained reinforcement learning agent is deployed. The process parameter vector sequence of each process in the current batch is input, and the trained reinforcement learning agent is used to analyze the adjustment amount of key process parameters based on the process parameter vector sequence of each process in the current batch. By analyzing and optimizing the process parameter configuration for each new batch, we continuously explore better combinations of process parameters while ensuring production line stability, and generate optimization results. The optimization results are directly sent to the production line equipment for execution via the programmable logic controller interface, while retaining a manual review interface to allow process engineers to review and confirm the recommended parameters of the intelligent agent.
[0034] It should be noted that applying deep reinforcement learning to the real-time dynamic optimization of process parameters in the preparation of monkeypox and orthopoxvirus detection products, and accelerating the exploration-exploitation process through an environmental simulator, significantly shortens the process optimization iteration cycle and solves the technical problem of obtaining the global optimum in multi-parameter coupled systems. Furthermore, the deep reinforcement learning-based adaptive optimization system for process parameters reduces the optimization cycle of multi-parameter coupled processes from several weeks to several days, significantly reducing batch variations in the preparation of monkeypox detection products, and effectively improving the consistency and stability of product quality.
[0035] Furthermore, in the preparation method of products for monkeypox and orthopoxvirus detection, a smart agent is used to continuously monitor process drift of process parameters, and adaptive correction is performed based on the process drift detection results. Specifically: When the production line is under statistical control, a preset number of normal process parameter time series data are collected. Based on the normal process parameter time series data, the multivariate residual vector at each moment is calculated, and the covariance matrix of the residuals is estimated. Calculate the Mahalanobis distance between the residual vector and the vectors in the covariance matrix, introduce the adaptive offset factor and dynamic weight factor, calculate the positive cumulative statistic and the negative cumulative statistic, and calculate the adaptive offset factor. When the positive cumulative statistic is greater than a preset positive threshold or the negative cumulative statistic is lower than a preset negative cumulative threshold, it is determined that the process has a systematic drift. Furthermore, the main process parameters causing the drift are analyzed through contribution plot analysis. The current cumulative residual vector is projected onto each parameter dimension, and the relative contribution percentage of each parameter is calculated. When the contribution percentage of a certain parameter exceeds the threshold for three consecutive sampling points, the parameter is identified as the root cause parameter of drift and used as the process drift detection result. Adaptive correction and compensation are then performed using an intelligent agent.
[0036] It should be noted that while positive and negative cumulative statistics are defined, a fixed offset parameter is no longer used. Instead, an adaptive offset factor and a dynamic weighting factor are introduced to enhance the sensitivity to different drift rates.
[0037] in It is the standard deviation of Mahalanobis distance under normal conditions. Mahalanobis distance value, adaptive offset factor Adjustments will be made dynamically based on recent trends.
[0038] here This is the baseline offset (usually 0.5-1.0). For the past The slope of the Mahalanobis distance change at each time point This represents the maximum expected drift slope under normal conditions. The sensitivity coefficient is (0.5-2.0). This design allows for operation even at high drift rates. The time required decreases, the accumulation rate increases, and thus the alarm is triggered more quickly.
[0039] when or At this point, a systematic drift in the process is determined. Further, the main process parameters causing the drift are analyzed using a contribution plot. The current cumulative residual vector is then... Projecting onto each parameter dimension, calculate the relative contribution percentage of each parameter. When the contribution percentage of a parameter exceeds the threshold for three consecutive sampling points (which can be preset to twice 1 / d), the parameter is identified as the root cause parameter of the drift.
[0040] It should be noted that the multivariate adaptive accumulation and collaborative drift mode that can simultaneously capture multiple coupled parameters proposed in this invention, and the automatic location of root cause parameters through contribution maps, solve the problem of difficulty in identifying coupled drift of multiple process parameters in the preparation of orthopoxvirus detection products. By dynamically adjusting the CUSUM detection speed through an adaptive offset factor, compared with CUSUM with fixed parameters, this invention has better detection timeliness and lower false alarm rate for both slowly rising drift and sudden step drift.
[0041] In addition, this method also includes: Train a counterfactual reasoning network to receive the current process state, detected drift parameters and their drift amplitudes, using a conditional variational autoencoder architecture; Using drift information as a conditional variable, a counterfactual sample is generated that is aligned with the current state distribution but whose drift parameters are corrected to the target value, thereby outputting a counterfactual state; The difference between the counterfactual state and the current actual state is used as the compensation target and input into the deep reinforcement learning agent. The agent's action space is defined as a compensation instruction vector that can be adjusted by each actuator; The generated compensation instruction vector is verified by the safety constraint module to ensure that the compensated process parameters do not exceed the physical limits of the equipment or cause irreversible degradation of product quality. After the compensation instruction is executed, the process parameters and CQA data of subsequent batches are continuously monitored. The compensation effect is fed back to the experience replay pool of the reinforcement learning agent. A priority experience replay mechanism is used to weight the training samples, enabling the agent to learn more frequently from successful and unsuccessful drift correction cases, and to fine-tune and update the deep reinforcement learning model using the latest accumulated compensation experience.
[0042] It should be noted that the current process state includes the current values of all KPPs and the trajectory over the last 30 minutes, while the counterfactual state is the ideal process state assuming no drift has occurred. The compensation command is as follows: Freeze-drying process: shelf temperature correction value (-2℃~+2℃, in 0.1℃ increments), vacuum setpoint adjustment (-0.05~+0.05mbar), and desorption drying time extension (0~30min).
[0043] Amplification process: hot cap temperature fine-tuning (±1℃), heating rate compensation (±0.2℃ / s), annealing temperature offset for each cycle (±0.5℃).
[0044] Fluid dispensing process: Solenoid valve opening pulse width correction (±5%), liquid supply pressure adjustment (±1kPa).
[0045] It should be noted that safety constraints include: the freeze-drying shelf temperature does not exceed the equipment's upper limit (e.g., 60℃), the amplifier's temperature change rate does not exceed the maximum rise / fall rate (e.g., 4℃ / s), and the dispensing flow rate does not exceed the minimum / maximum pumping capacity. The verified compensation command is sent to the PLC or directly drives the actuator (e.g., proportional control valve, electric heating power module) via the OPCUA protocol to achieve closed-loop adjustment. Traditional methods typically return to zero directly or use fixed rules for compensation after drift detection, ignoring the causal structure between parameters. This invention utilizes a counterfactual reasoning network driven by digital twins to establish a causal inference of "what the current process state should be like if the drift had not occurred," ensuring that the compensation command directly targets the root cause of the deviation rather than the surface phenomenon. The generation of compensation actions is modeled as a reinforcement learning problem, balancing quality recovery and action costs through a reward function, avoiding the trial-and-error costs of manual parameter tuning. In particular, the safety constraint module ensures that the compensation process does not introduce new quality risks.
[0046] In addition, this method also includes: Based on computational fluid dynamics and finite element analysis methods, heat and mass transfer models for the freeze-drying process, fluid dynamics models inside microfluidic chips, and thermodynamic and kinetic models for nucleic acid amplification reactions were constructed respectively. The heat and mass transfer model of the freeze-drying process, the fluid dynamics model inside the microfluidic chip, and the thermodynamic and kinetic models of the nucleic acid amplification reaction are fused with offline process experimental data through a system identification method to establish a digital twin mapping relationship between process parameters and product quality. Before formal production, a large-scale virtual trial production simulation is carried out using a digital twin. Within the feasible domain of process parameters, thousands of parameter combinations are generated in the parameter space through Latin hypercube sampling or Sobol sequence. The digital twin evaluates the predicted product quality value under each combination one by one. Based on the predicted product quality values, parameters that may deviate from specifications and the expected deviation range are marked. When the predicted deviation exceeds the quality threshold, compensatory adjustment measures are recommended.
[0047] It is important to note the following: the heat and mass transfer model for the freeze-drying process simulates the spatial non-uniformity of temperature distribution, ice crystal growth morphology, and sublimation rate at various locations within the sample vial in the freeze-drying chamber, incorporating microcavity structural parameters unique to the freeze-drying microchip; the fluid dynamics model within the microfluidic chip simulates the flow characteristics of the sample within the microchannels, the risk of bubble entrainment, and mixing uniformity, with microchannel geometric parameters (width 50–200 μm, depth 30–100 μm) serving as model input variables; and the thermodynamic and kinetic models for the nucleic acid amplification reaction simulate the heating and cooling rates, temperature field uniformity, and primer-template binding kinetics of the reaction solution within the PCR amplification instrument or isothermal amplification module. Model parameters are calibrated using sensor data from the actual production line. The digital twin predicts the trend of product quality indicators after scale-up and marks parameters that may deviate from specifications and the expected deviation range, such as a deviation in freeze-drying temperature leading to an increase in residual moisture from 1% to 5%. When the predicted deviation exceeds the quality threshold, compensatory adjustment measures are recommended, such as extending the desorption drying time and adjusting the plate heating and cooling rates.
[0048] It should be noted that this method enables the construction of a multi-physics digital twin covering multiple processes such as freeze-drying, microfluidics, and amplification reactions in the preparation of vaccinia virus detection products, achieving full-process collaborative simulation from virtual pilot production to online mirroring. The digital twin's process scale-up risk assessment and root cause inference functions solve the technical problems of high trial-and-error costs in process development and difficulty in predicting online anomalies in existing technologies. Moreover, multi-physics model coupling prediction can further improve the prediction accuracy of process scale-up risk assessment and root cause inference.
[0049] It should be noted that the overall process is as follows: Data acquisition (step S1): Data is collected in real time via IoT sensors throughout the entire process of recombinant adenovirus construction, purification, dilution determination, and lyophilization. Freeze-drying parameters: pre-freezing temperature -90℃, pre-freezing for 1.1h; primary drying temperature -25℃; desorption drying temperature gradually changed from -5℃ to 37℃, vacuum degree 0.3mbar.
[0050] Amplification and identification parameters: annealing temperature 60℃, cycle number 40.
[0051] Purification parameters: Benzobonase nuclease concentration, elution buffer volume, etc.
[0052] Simultaneously, the quality control results of this batch of positive control were recorded: LOD was 10 copies / mL, CV was 3.2%, and specificity score was 99.5%. This forms a training sample.
[0053] Model Training (Step S2): Construct a deep neural network with 3 hidden layers (128 neurons per layer). The input layer receives encoded freeze-drying curve features, amplification response features, etc. The output layer predicts LOD, CV, and SS, respectively. The model is trained using data from the past 100 batches (including quality control data from different samples of the monkeypox virus West African branch and Congo Basin branch) until the prediction error on the validation set converges.
[0054] Reinforcement learning agent optimization (step S3): Environment simulator: Uses the deep neural network model trained above.
[0055] The agent employs the PPO algorithm. Its state is the parameter vector for the current batch (e.g., [pre-freezing temperature, desorption drying temperature, vacuum level, annealing temperature, pH value]). Its action is the adjustment amount to the above parameters, for example, an action of [+0.5℃, -0.2℃, +0.01mbar, +0.2℃, -0.1].
[0056] Execution: After the process engineer approves the optimization suggestions through the system interface, the suggestions are automatically sent to the freeze dryer and PCR instrument via the PLC interface to guide the preparation of the new batch.
[0057] Process drift detection and correction (step S4): In a certain batch of production, the drift detection module continuously calculates the Mahalanobis distance between the "partition temperature" and "vacuum degree" in the freeze-drying process. The adaptive CUSUM statistic C_t^+ exceeds the threshold h.
[0058] Root cause identification: The contribution plot shows that the contribution percentage of "partition temperature" exceeds the threshold for 3 consecutive points, and is therefore identified as the root cause parameter.
[0059] Counterfactual reasoning: The counterfactual reasoning network (CVAE) receives the current state (temperature offset +1.2℃, vacuum level normal) and drift information, and generates a counterfactual state (temperature should be corrected back to the set value).
[0060] Compensation instruction: Based on the compensation target (temperature -1.2℃), the reinforcement learning agent outputs the instruction "reduce shelf temperature setpoint by 1.2℃".
[0061] Safety Verification: The safety constraint module confirms that the reduced temperature does not exceed the equipment's lower limit (-25℃) and will not cause the product to collapse. The instruction is issued via the OPCUA protocol, and the actuator adjusts the heating power to complete the calibration.
[0062] like Figure 2As shown, a second aspect of the present invention provides a preparation system for a product for detecting monkeypox and orthopoxvirus, including a memory and a processor. The memory includes a method program for preparing the product for detecting monkeypox and orthopoxvirus. When the processor executes the method program for preparing the product for detecting monkeypox and orthopoxvirus, it implements the steps of the method for preparing the product for detecting monkeypox and orthopoxvirus as described in any one of the claims.
[0063] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.
[0064] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.
[0065] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.
[0066] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0067] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.
[0068] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for preparing a product for detection of monkeypox and orthopoxvirus, characterized by, Includes the following steps: Key process parameter data were collected throughout the entire preparation process of monkeypox and orthopoxvirus, and a training set was constructed based on the key process parameter data. A process parameter-quality mapping model is constructed based on a deep learning network, and the process parameter-quality mapping model is trained using the training set. Construct a deep reinforcement learning agent and use it to optimize and recommend the process parameter configuration for each current batch. The process drift detection is carried out by continuously monitoring process parameters using an intelligent agent, and adaptive correction is performed based on the process drift detection results.
2. The method for preparing a product for detecting monkeypox and orthopoxvirus according to claim 1, characterized in that, The entire process for preparing monkeypox and orthopoxvirus includes the purification of recombinant adenovirus, determination of the optimal dilution of recombinant adenovirus, positive control analysis, identification of recombinant adenovirus shuttle plasmid, synthesis of target gene sequence and primers, construction of recombinant adenovirus shuttle plasmid, packaging of recombinant adenovirus particles, passage amplification and identification of recombinant adenovirus, freeze drying, preparation of nucleic acid amplification reaction solution, and preparation of sample nucleic acid release reagent.
3. The method for preparing a product for detecting monkeypox and orthopoxvirus according to claim 1, characterized in that, Key process parameter data were collected throughout the entire preparation process of monkeypox and orthopoxvirus, and a training set was constructed based on this data. Specifically: Key process parameter data were collected throughout the entire product preparation process, including: pre-freezing temperature and pre-freezing time in the freeze-drying process, temperature rise curve of the partition in the desorption drying stage, vacuum degree change curve, preparation parameters of nucleic acid amplification reaction solution, isothermal amplification reaction temperature and time, and formulation parameters of sample nucleic acid release reagent. The pre-freezing temperature, pre-freezing time, partition temperature rise curve, vacuum degree change curve, nucleic acid amplification reaction solution preparation parameters, isothermal amplification reaction temperature and time, and sample nucleic acid release reagent formulation parameters of the drying process are collected in real time and the corresponding product quality test results are labeled. A training dataset is constructed based on the pre-freezing temperature, pre-freezing time, partition temperature rise curve during the analytical drying stage, vacuum degree change curve, nucleic acid amplification reaction solution preparation parameters, isothermal amplification reaction temperature and time, sample nucleic acid release reagent formulation parameters, and product quality test results in the aforementioned drying process.
4. The method for preparing a product for detecting monkeypox and orthopoxvirus according to claim 1, characterized in that, A process parameter-quality mapping model is constructed based on a deep learning network, and the model is trained using the training set, specifically as follows: The process parameter space and quality index space are constructed using the training dataset. A multi-layer deep neural network is used to construct a mapping model from the process parameter space to the quality index space, forming a process parameter-quality mapping model. The pre-freezing temperature, primary drying temperature, analytical drying temperature, time parameters, and pressure parameters at each stage of the freeze-drying process are encoded as the first input feature vector, and the annealing temperature of primers and probes, the number of amplification cycles, and the fluorescence collection time point in the amplification reaction system are encoded as the second input feature vector. The ionic strength and pH value of the nucleic acid release reagent are encoded as the third input feature vector. The first, second, and third input feature vectors are combined in a high-order nonlinear manner and then input into a deep neural network. The lowest detection limit (LOD), coefficient of variation (CV), and specificity score (SS) of the deep neural network are used to form predicted values of quality indicators, and a trained process parameter-quality mapping model is obtained.
5. The method for preparing a product for detecting monkeypox and orthopoxvirus according to claim 4, characterized in that, Constructing a deep reinforcement learning agent, specifically: The trained process parameter-quality mapping model is used as an environment simulator for reinforcement learning to construct a process parameter optimization agent based on deep Q-network or proximal policy optimization algorithm. The state space of the agent is defined as the sequence of process parameter vectors for each process in the current batch. The agent's reward function is designed based on the output of the quality prediction model. It gives positive rewards for low detection limits, low coefficient of variation, and high specificity, and imposes penalty terms on abnormal parameter combinations.
6. The method for preparing a product for detecting monkeypox and orthopoxvirus according to claim 1, characterized in that, The process parameter configuration for each current batch is optimized and recommended using a deep reinforcement learning agent, specifically as follows: In the production line control system, a trained reinforcement learning agent is deployed. The process parameter vector sequence of each process in the current batch is input, and the trained reinforcement learning agent is used to analyze the adjustment amount of key process parameters based on the process parameter vector sequence of each process in the current batch. By analyzing and optimizing the process parameter configuration for each new batch, we continuously explore better combinations of process parameters while ensuring production line stability, and generate optimization results. The optimization results are directly sent to the production line equipment for execution via the programmable logic controller interface, while retaining a manual review interface to allow process engineers to review and confirm the recommended parameters of the intelligent agent.
7. The method for preparing a product for detecting monkeypox and orthopoxvirus according to claim 1, characterized in that, The process drift detection utilizes an intelligent agent to continuously monitor process parameters and performs adaptive correction based on the process drift detection results. Specifically: When the production line is under statistical control, a preset number of normal process parameter time series data are collected. Based on the normal process parameter time series data, the multivariate residual vector at each moment is calculated, and the covariance matrix of the residuals is estimated. Calculate the Mahalanobis distance between the residual vector and the vectors in the covariance matrix, introduce the adaptive offset factor and dynamic weight factor, calculate the positive cumulative statistic and the negative cumulative statistic, and calculate the adaptive offset factor. When the positive cumulative statistic is greater than a preset positive threshold or the negative cumulative statistic is lower than a preset negative cumulative threshold, it is determined that the process has a systematic drift. Furthermore, the main process parameters causing the drift are analyzed through contribution plot analysis. The current cumulative residual vector is projected onto each parameter dimension, and the relative contribution percentage of each parameter is calculated. When the contribution percentage of a certain parameter exceeds the threshold for three consecutive sampling points, the parameter is identified as the root cause parameter of drift and used as the process drift detection result. Adaptive correction and compensation are then performed using an intelligent agent.
8. A preparation system for products used in the detection of monkeypox and orthopoxvirus, characterized in that, The device includes a memory and a processor. The memory includes a method program for preparing a product for detecting monkeypox and orthopoxvirus. When the processor executes the method program for preparing a product for detecting monkeypox and orthopoxvirus, it implements the steps of the method for preparing a product for detecting monkeypox and orthopoxvirus as described in any one of claims 1-7.