Low-temperature crystallization process early warning method and system based on digital twin polypeptide drugs

Abstract

The invention belongs to the technical field of polypeptide drugs, and discloses a low-temperature crystallization process early warning method and system based on digital twin polypeptide drugs, and the method comprises the steps: obtaining batch information of polypeptide drugs, and crystallization process data and online spectrum data during a low-temperature crystallization process, inputting batch information and crystallization process data into a digital twinborn model, predicting to obtain state parameters in a low-temperature crystallization process, extracting feature vectors related to the crystallization process from online spectral data to obtain real-time spectral features, calculating the inconsistency degree between the key state parameters and the real-time spectral features, and calculating the inconsistency degree between the key state parameters and the real-time spectral features. A prediction deviation evaluation result is obtained, and when it is detected that the prediction deviation evaluation result is larger than a preset evaluation threshold value or the real-time spectral features are abnormal data, an early warning signal is sent out; when it is detected that the prediction deviation evaluation result is larger than the preset evaluation threshold value or the real-time spectral characteristics are abnormal data, the early warning signal is sent out, and the early warning efficiency in the low-temperature crystallization process is improved.

Description

Technical Field

[0001] The present application relates to the technical field of polypeptide drugs, and more specifically, to a method and system for early warning of the low-temperature crystallization process of polypeptide drugs based on digital twins. Background Art

[0002] Peptide drugs are an essential component of modern medicine, and their production processes require extremely high purity and quality control. Low-temperature crystallization is a key purification step in the industrial production line of peptide drugs. Its goal is to precipitate peptides from solution as high-purity crystals by controlling process parameters such as temperature, solvent composition, and stirring rate.

[0003] In the industrial multi-batch production environment of peptide drugs, even with seemingly identical process procedures, equipment conditions, and qualified raw materials and solvents, subtle differences between production batches can still be difficult to completely eliminate. This batch-to-batch variability can arise from subtle fluctuations in the purity or impurity profile of the API batches, slight differences in the moisture content or composition ratio between solvent batches, or even minor wear or changes in the equipment's condition after long-term operation. These seemingly insignificant batch-specific factors can cumulatively affect the fine crystallization behavior of peptides in low-temperature environments beyond expectations.

[0004] Existing early warning systems based on digital twins and machine learning face challenges in addressing this inter-batch variability. For one thing, the historical data used to train digital twin and machine learning models may not fully capture certain subtle patterns of variation unique to the current batch. Furthermore, machine learning models significantly decrease their ability to recognize these inadequately learned batch-specific patterns. Due to the limitations of digital twin models in adapting to batch differences and machine learning models in identifying inadequately learned early anomaly patterns, early warning systems may fail to issue accurate warnings at an early stage, potentially resulting in the scrapping of entire production batches or the need for time-consuming and expensive post-processing, significantly increasing production costs.

[0005] Therefore, in order to solve the technical problems that the existing early warning methods for the low-temperature crystallization process of polypeptide drugs have insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings, there is an urgent need for an early warning method and system for the low-temperature crystallization process of polypeptide drugs based on digital twins. Summary of the Invention

[0006] The purpose of this application is to provide an early warning method and system for the low-temperature crystallization process of polypeptide drugs based on digital twins. By calculating the degree of inconsistency between key state parameters and real-time spectral characteristics, a prediction deviation evaluation result is obtained, and when the prediction deviation evaluation result is greater than a preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued. This solves the problem that the existing early warning method for the low-temperature crystallization process of polypeptide drugs has insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the actual state reflected by the digital twin model prediction and the online spectrum, the batch variability and abnormal patterns of the low-temperature crystallization process of polypeptide drugs can be adaptively captured, thereby improving the early warning efficiency of the low-temperature crystallization process of polypeptide drugs.

[0007] In the first aspect, the present application provides an early warning method for the low-temperature crystallization process of a digital twin polypeptide drug, comprising: Obtaining specific batch information of the peptide drug to be tested, and obtaining real-time crystallization process data and online spectral data of the peptide drug during the low-temperature crystallization process; Inputting the specific batch information and the crystallization process data into a preset digital twin model to simulate the low-temperature crystallization process and predict the key state parameters of the test polypeptide drug at the next moment in the low-temperature crystallization process; Extracting characteristic vectors related to the crystallization process from the online spectral data to obtain real-time spectral characteristics; Calculating the degree of inconsistency between the key state parameter and the real-time spectral feature to obtain a prediction deviation assessment result; When it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data, an early warning signal is issued.

[0008] The digital twin-based polypeptide drug low-temperature crystallization process early warning method provided in this application can realize early warning of abnormalities in the low-temperature crystallization process of polypeptide drugs. By calculating the degree of inconsistency between key state parameters and real-time spectral characteristics, a prediction deviation evaluation result is obtained. When the prediction deviation evaluation result is greater than a preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued. This solves the problem that the existing polypeptide drug low-temperature crystallization process early warning method has insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the digital twin model prediction and the actual state reflected by the online spectrum, it can adaptively capture the batch variability and abnormal patterns of the polypeptide drug low-temperature crystallization process, thereby improving the early warning efficiency of the polypeptide drug low-temperature crystallization process.

[0009] Optionally, extracting a characteristic vector related to the crystallization process from the online spectral data to obtain a real-time spectral feature includes: Preprocessing the online spectral data to obtain preprocessed Raman spectral data; The characteristic peak area related to the polypeptide concentration, the characteristic peak position related to the solvent composition, the characteristic peak intensity ratio related to the crystal form, and the specific band integral area related to impurities or intermediate products in the crystallization process are extracted from the pre-processed Raman spectral data to construct corresponding feature vectors and obtain real-time spectral features.

[0010] The digital twin-based polypeptide drug low-temperature crystallization process early warning method provided in this application can realize early warning of abnormalities in the low-temperature crystallization process of polypeptide drugs. The characteristic peak area related to the polypeptide concentration, the characteristic peak position related to the solvent composition, the characteristic peak intensity ratio related to the crystal form, and the specific band integral area related to impurities or intermediate products in the crystallization process are extracted from the pre-processed Raman spectral data. By extracting these features with physicochemical significance, the high-dimensional spectral data can be converted into low-dimensional and information-rich feature vectors. These feature vectors can be more reliably used for subsequent comparison with the key state parameters predicted by the digital twin model, thereby improving the accuracy of the prediction deviation assessment and ultimately improving the reliability of the early warning system.

[0011] Optionally, preprocessing the online spectral data to obtain preprocessed Raman spectral data includes: Performing filtering on the online spectral data to obtain filtered Raman spectral data; Using a smoothing algorithm, smoothing the filtered Raman spectrum data to obtain smoothed Raman spectrum data; The fluorescence background interference of the smoothed Raman spectrum data is corrected by using the least square method to obtain pre-processed Raman spectrum data.

[0012] Optionally, calculating the degree of inconsistency between the key state parameter and the real-time spectral feature to obtain a prediction deviation assessment result includes: performing standardization processing on the key state parameters and the real-time spectral characteristics respectively to obtain standardized key state parameters and standardized real-time spectral characteristics; Based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector; The parameter weighted deviation vector and the spectral weighted deviation vector are input into a preset deviation prediction model to calculate a prediction deviation evaluation result.

[0013] The digital twin-based polypeptide drug low-temperature crystallization process early warning method provided in this application can realize early warning of abnormalities in the low-temperature crystallization process of polypeptide drugs. By weighting the data reliability and using a prediction model for comprehensive evaluation, the accuracy and sensitivity of the assessment of the degree of inconsistency between key state parameters and real-time spectral characteristics are improved, thereby more effectively identifying early or minor process anomalies and supporting subsequent early warning judgments.

[0014] Optionally, based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector, including: Based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, determining the weight of the key state parameter after normalization and the weight of the real-time spectral feature after normalization, which are respectively recorded as a first weight and a second weight; wherein, the greater the uncertainty, the smaller the first weight, and the greater the extraction deviation, the smaller the second weight; Based on the first weight and the second weight, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector.

[0015] Optionally, based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, determining the weight of the key state parameter after the corresponding normalization processing and the weight of the real-time spectral feature after the corresponding normalization processing includes: Determining the uncertainty of the key state parameter by inputting the specific batch information and the crystallization process data into a preset digital twin model multiple times and calculating the variance of the key state parameter; determining an extraction deviation of the real-time spectral feature by extracting characteristic vectors related to the crystallization process from the online spectral data multiple times and calculating actual differences of the real-time spectral feature; Based on the uncertainty and the extraction deviation, the weight of the corresponding normalized key state parameter and the weight of the corresponding normalized real-time spectral feature are determined.

[0016] Optionally, the preset deviation prediction model is a model constructed using historical data to represent the nonlinear relationship between key state parameters and real-time spectral characteristics; the steps of constructing the preset deviation prediction model include: Acquire historical batch information, historical crystallization process data, historical spectral data, and historical nonlinear relationship results between the historical crystallization process data and the historical spectral data of multiple polypeptide drugs from a database; Based on the historical batch information, historical crystallization process data, historical spectral data and the historical nonlinear relationship results, corresponding historical key state parameters and historical spectral characteristics are generated; According to the historical key state parameters and the historical spectral characteristics, a machine learning algorithm is used to construct the preset deviation prediction model.

[0017] Optionally, when it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data, issuing an early warning signal includes: Based on the historical spectral data in the database, the abnormal data determination conditions are predetermined; Based on the abnormal data determination condition, determining whether the real-time spectral feature is abnormal data, and at the same time, comparing the magnitude relationship between the prediction deviation evaluation result and a preset evaluation threshold; When it is detected that the prediction deviation evaluation result is greater than the preset evaluation threshold, or when it is detected that the real-time spectral feature is abnormal data, an early warning signal is issued.

[0018] Optionally, the abnormal data determination condition includes a numerical abnormality condition and a pattern abnormality condition; and judging whether the real-time spectral feature is abnormal data based on the abnormal data determination condition includes: Extracting spectral data of historical normal batches and spectral data of historical abnormal patterns from the historical spectral data; Calculating the Mahalanobis distance between the real-time spectral feature and the spectral data of the historical normal batch, and determining whether the corresponding real-time spectral feature is abnormal data by whether the Mahalanobis distance is greater than a preset normal threshold; when the Mahalanobis distance is greater than the preset normal threshold, determining that the corresponding real-time spectral feature meets the numerical abnormality condition, and determining that the corresponding real-time spectral feature is abnormal data; The Euclidean distance between the real-time spectral feature and the spectral data of the historical abnormal pattern is calculated to determine whether the corresponding real-time spectral feature is abnormal data by determining whether the Euclidean distance is less than a preset abnormal threshold; when the Euclidean distance is less than the preset abnormal threshold, it is determined that the corresponding real-time spectral feature meets the pattern abnormality condition and is determined to be abnormal data.

[0019] In the second aspect, the present application provides an early warning system for the low-temperature crystallization process of polypeptide drugs based on digital twins, comprising: An acquisition module is used to obtain specific batch information of the polypeptide drug to be tested, and to obtain crystallization process data and online spectral data of the polypeptide drug during the low-temperature crystallization process in real time; A prediction module is used to input the specific batch information and the crystallization process data into a preset digital twin model to simulate the low-temperature crystallization process and predict the key state parameters of the test polypeptide drug at the next moment in the low-temperature crystallization process; An extraction module, configured to extract characteristic vectors related to the crystallization process from the online spectral data to obtain real-time spectral features; an extraction module, configured to calculate the degree of inconsistency between the key state parameters and the real-time spectral features, and obtain a prediction deviation assessment result; The early warning module is used to issue an early warning signal when it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data.

[0020] This digital twin-based polypeptide drug low-temperature crystallization process early warning system obtains the prediction deviation evaluation result by calculating the degree of inconsistency between key state parameters and real-time spectral characteristics. When the prediction deviation evaluation result is greater than the preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued. This solves the problem that the existing polypeptide drug low-temperature crystallization process early warning method has insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the digital twin model prediction and the actual state reflected by the online spectrum, it can adaptively capture the batch variability and abnormal patterns of the polypeptide drug low-temperature crystallization process, thereby improving the early warning efficiency of the polypeptide drug low-temperature crystallization process.

[0021] Beneficial effects: The digital twin-based polypeptide drug low-temperature crystallization process early warning method and system provided in this application obtain prediction deviation evaluation results by calculating the degree of inconsistency between key state parameters and real-time spectral characteristics, and when the prediction deviation evaluation result is greater than the preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued, which solves the problem that the existing polypeptide drug low-temperature crystallization process early warning method has insufficient adaptability to batch differences and insufficient abnormal pattern recognition ability when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the actual state reflected by the digital twin model prediction and the online spectrum, it can adaptively capture the batch variability and abnormal patterns of the polypeptide drug low-temperature crystallization process, thereby improving the early warning efficiency of the polypeptide drug low-temperature crystallization process. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Flowchart of the early warning method for the low-temperature crystallization process of digital twin polypeptide drugs provided in the embodiment of this application.

[0023] Figure 2 Schematic diagram of the structure of the early warning system for the low-temperature crystallization process of digital twin polypeptide drugs provided in the embodiment of the present application.

[0024] Explanation of numbers: 1. Acquisition module; 2. Prediction module; 3. Extraction module; 4. Calculation module; 5. Early warning module. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work fall within the scope of protection of the present application.

[0026] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0027] Please refer to Figure 1 , Figure 1 In some embodiments of the present application, a method for early warning of a low-temperature crystallization process of a polypeptide drug based on a digital twin is provided, which is used to provide an early warning of abnormalities in the low-temperature crystallization process of the polypeptide drug, comprising the steps of: Step S101, obtaining batch information of the polypeptide drug, and obtaining crystallization process data and online spectral data of the polypeptide drug during the low-temperature crystallization process; Step S102: Input the batch information and crystallization process data into a preset digital twin model to simulate the low-temperature crystallization process and predict the state parameters of the polypeptide drug at the next moment in the low-temperature crystallization process; Step S103, extracting characteristic vectors related to the crystallization process from the online spectral data to obtain real-time spectral features; Step S104, calculating the inconsistency between the key state parameters and the real-time spectral characteristics to obtain a prediction deviation evaluation result; Step S105: When it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data, an early warning signal is issued.

[0028] This digital twin-based early warning method for the low-temperature crystallization process of polypeptide drugs obtains a prediction deviation evaluation result by calculating the degree of inconsistency between key state parameters and real-time spectral characteristics. When the prediction deviation evaluation result is greater than a preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued. This solves the problem that the existing early warning method for the low-temperature crystallization process of polypeptide drugs has insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the actual state reflected by the digital twin model prediction and the online spectrum, it can adaptively capture the batch variability and abnormal patterns of the low-temperature crystallization process of polypeptide drugs, thereby improving the early warning efficiency of the low-temperature crystallization process of polypeptide drugs.

[0029] Specifically, in step S101, batch information of the peptide drug is obtained, as well as crystallization process data and online spectral data during the low-temperature crystallization process of the peptide drug. Batch information includes information such as raw material batch information and solvent batch information. Crystallization process data, including temperature, pressure, stirring rate, and feed rate, reflects the macroscopic state of the crystallization process and can be collected in real time using various sensors. Online spectral data can be obtained using equipment such as online Raman spectrometers. Online spectral data can reflect the vibration and rotation of the substance molecules and is closely related to the peptide, solvent, impurities, and crystal structure.

[0030] Specifically, in step S102, a preset digital twin model is used to simulate and predict the crystallization process. Batch information and crystallization process data are used as model inputs, enabling the model to reflect the specific state of the current batch and provide key state parameters of the crystallization process for subsequent steps. Key state parameters are important process indicators and include parameters such as peptide concentration and crystal particle size. A digital twin model is a digital model (i.e., a model constructed using digital twin technology) that is highly similar to a physical entity in a virtual space through data collection, transmission, processing, and analysis, modeling and simulation. Specifically, it can include a peptide crystallization kinetic model in a specific solvent system. Digital twin models are prior art and will not be described in detail here.

[0031] Specifically, in step S103, feature vectors related to the crystallization process are extracted from the online Raman spectral data to obtain real-time spectral features, including: Preprocessing the online Raman spectrum data to obtain preprocessed Raman spectrum data; The characteristic peak area related to the peptide concentration, the characteristic peak position related to the solvent composition, the characteristic peak intensity ratio related to the crystal form, and the specific band integral area related to impurities or intermediate products in the crystallization process are extracted from the pre-processed Raman spectral data to construct the corresponding feature vector and obtain real-time spectral features.

[0032] Specifically, in step S103, the online Raman spectrum data is preprocessed to obtain preprocessed Raman spectrum data, including: Performing filtering on the online Raman spectrum data to obtain filtered Raman spectrum data; Using a smoothing algorithm, smoothing the filtered Raman spectrum data to obtain smoothed Raman spectrum data; The least square method is used to correct the fluorescence background interference of the smoothed Raman spectral data to obtain the preprocessed Raman spectral data.

[0033] In step S103, the original online Raman spectral data may contain random noise and fluorescence background interference, which can obscure the true Raman signal and result in inaccurate feature vectors extracted from the data. To address this issue, the online Raman spectral data is first filtered (e.g., Fourier transform filtering or wavelet filtering) to remove high-frequency noise and obtain noise-reduced data. Next, the filtered data is smoothed (e.g., using algorithms such as Savitzky-Golay smoothing or moving average smoothing) to further reduce data fluctuations and make the spectral curve flatter, which is beneficial for subsequent peak identification and measurement. Finally, the smoothed data is corrected for fluorescence background using the least squares method to isolate the true Raman signal and eliminate the effects of fluorescence interference on peak intensity and area measurements. After these preprocessing steps, high-quality Raman spectral data is obtained. Filtering, smoothing, and the least squares method are prior art and will not be described in detail here.

[0034] In step S103, features extracted from the preprocessed Raman spectral data include characteristic peak areas associated with peptide concentration, characteristic peak positions related to solvent composition, characteristic peak intensity ratios associated with crystal form, and the integrated areas of specific bands associated with impurities or crystallization intermediates. These features represent key information about the crystallization process: changes in peptide content, changes in the solvent environment, crystal structure formation and transformation, and potential side reactions or impurity formation. By extracting these physicochemically meaningful features, high-dimensional spectral data can be converted into low-dimensional, information-rich feature vectors.

[0035] Specifically, in step S104, the degree of inconsistency between the key state parameters and the real-time spectral characteristics is calculated to obtain a prediction deviation evaluation result, including: performing standardization processing on the key state parameters and the real-time spectral characteristics respectively to obtain the standardized key state parameters and the standardized real-time spectral characteristics; Based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral characteristics, the weighted deviation of the normalized key state parameters and the normalized real-time spectral characteristics is calculated to obtain the parameter weighted deviation vector and the spectral weighted deviation vector; The parameter weighted deviation vector and the spectral weighted deviation vector are input into the preset deviation prediction model to calculate the prediction deviation evaluation result.

[0036] In step S104, key state parameters and real-time spectral features are standardized using existing standardization methods, such as the Z-score standardization method. Standardization eliminates differences in dimensions and numerical ranges between different parameters and features, making different types of data comparable. Standardization is a prior art technique and will not be described in detail here.

[0037] Specifically, in step S104, based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral features, the weighted deviations of the normalized key state parameters and the normalized real-time spectral features are calculated to obtain the parameter weighted deviation vector and the spectral weighted deviation vector, including: Based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral features, the weights of the key state parameters after the corresponding standardization and the weights of the real-time spectral features after the corresponding standardization are determined and recorded as the first weight and the second weight, respectively; wherein, the greater the uncertainty, the smaller the first weight, and the greater the extraction deviation, the smaller the second weight; Based on the first weight and the second weight, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector.

[0038] Specifically, in step S104, based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral features, the weights of the key state parameters and the real-time spectral features after the normalization processing are determined, including: The uncertainty of the key state parameters is determined by inputting specific batch information and crystallization process data into the preset digital twin model multiple times and calculating the variance of the key state parameters; The actual difference of the real-time spectral features calculated by extracting the characteristic vectors related to the crystallization process from the online Raman spectral data multiple times is used to determine the extraction deviation of the real-time spectral features; Based on the uncertainty and extraction deviation, the weights of the corresponding normalized key state parameters and the weights of the corresponding normalized real-time spectral features are determined.

[0039] In step S104, to quantify the uncertainty of the model prediction, the simulation process (the process of inputting specific batch information and real-time crystallization process data into the digital twin model for simulation) is repeated multiple times. Each simulation may introduce small perturbations or use different model parameter sampling. The variance of the key state parameter results obtained from these multiple simulations is calculated. The variance value reflects the stability of the model prediction, i.e., the uncertainty. Simultaneously, to quantify the deviation of feature extraction, feature extraction (i.e., the step of extracting feature vectors related to the crystallization process from the online Raman spectral data) is repeated multiple times on the same set of raw spectral data. The average of these spectral feature results is calculated, and the actual difference between the real-time spectral features and the average value is calculated. This actual difference reflects the stability of the feature extraction process, i.e., the extraction deviation. Based on the calculated uncertainty (variance) and extraction deviation (actual difference), the weights of the key state parameters and real-time spectral features are determined when calculating the weighted deviation. Key state parameters with large uncertainty are assigned lower weights, and real-time spectral features with large extraction deviations are assigned lower weights. This reduces the impact of uncertain or unreliable data when calculating the weighted deviation between the key state parameters and the real-time spectral features, making the calculation of the weighted deviation more objective and accurate. This more reliable deviation assessment helps to more accurately identify anomalies in the low-temperature crystallization process, especially for small variations between batches or early, rare abnormal patterns. Through more reliable deviation assessment, the sensitivity and accuracy of early warning can be improved.

[0040] In step S104, the normalized key state parameters and the normalized real-time spectral features are multiplied by corresponding weights to obtain parameter weighted deviation vectors and spectral weighted deviation vectors. Each component of the weighted deviation vector represents the difference between the predicted value and the actual value of the corresponding parameter or feature.

[0041] The parameter-weighted deviation vector and the spectral-weighted deviation vector are input into a pre-set deviation prediction model, which outputs a prediction deviation assessment result. This assessment reflects the degree of deviation between the current process state and the normal state. The deviation prediction model can be a machine learning model, such as a support vector machine, random forest, or neural network algorithm. This deviation prediction model is used to accurately assess the deviation between the key state parameters predicted by the digital twin model and the spectral features acquired in real time.

[0042] The preset deviation prediction model is a model constructed using historical data to represent the nonlinear relationship between key state parameters and real-time spectral characteristics; the steps of constructing the preset deviation prediction model include: Obtain historical batch information, historical crystallization process data, historical spectral data, and historical nonlinear relationship results between historical crystallization process data and historical spectral data for multiple peptide drugs from a database; Generate corresponding historical key state parameters and historical spectral characteristics based on historical batch information, historical crystallization process data, historical spectral data and historical nonlinear relationship results; Based on historical key state parameters and historical spectral characteristics, a machine learning algorithm is used to construct a preset deviation prediction model.

[0043] Specifically, historical data is collected from a database, including information on peptide drug batches (normal batches), crystallization process data, online Raman spectral data, and known nonlinear relationships between these historical data. This historical data provides the foundational information needed for model training, covering conditions across different batches and process conditions. Based on this historical data, corresponding historical key state parameters and historical spectral features are generated. This step transforms the raw historical data into a form directly related to the deviation to be assessed, enabling the model to learn the corresponding relationships between key state parameters and spectral features. A machine learning algorithm is then used to construct a pre-defined deviation prediction model based on the generated historical key state parameters and historical spectral features. The machine learning algorithm learns and establishes complex nonlinear mappings between key state parameters and real-time spectral features from the historical data. By learning these nonlinear relationships, the model can more accurately understand the associations between key state parameters and real-time spectral features. In practical applications, when input with standardized key state parameters and standardized real-time spectral features, it can output a prediction deviation assessment result that reflects the degree of inconsistency between them.

[0044] Specifically, in step S105, when it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data, an early warning signal is issued, including: Based on the historical spectral data in the database, the abnormal data determination conditions are predetermined; Based on the abnormal data determination conditions, determine whether the real-time spectral characteristics are abnormal data, and at the same time, compare the size relationship between the prediction deviation evaluation result and the preset evaluation threshold; When it is detected that the prediction deviation evaluation result is greater than the preset evaluation threshold, or when it is detected that the real-time spectral feature is abnormal data, an early warning signal is issued.

[0045] In step S105, the historical spectral data accumulated in the database is analyzed to identify normal batch data and known abnormal patterns. Specific criteria for identifying abnormal data are pre-defined. These criteria can be based on the numerical range of spectral features, the presence or absence of specific peaks, peak area ratios, and other factors. This step provides a basis for subsequent real-time judgment.

[0046] Among them, the abnormal data determination conditions include numerical abnormality conditions and pattern abnormality conditions.

[0047] In step S105, based on the abnormal data determination condition, determining whether the real-time spectrum feature is abnormal data includes: Extracting spectral data of historical normal batches and spectral data of historical abnormal patterns from historical spectral data; Calculate the Mahalanobis distance between the real-time spectral feature and the spectral data of the historical normal batch, and determine whether the corresponding real-time spectral feature is abnormal data by whether the Mahalanobis distance is greater than the preset normal threshold; when the Mahalanobis distance is greater than the preset normal threshold, determine that the corresponding real-time spectral feature meets the numerical abnormality condition and determine that the corresponding real-time spectral feature is abnormal data; The Euclidean distance between the real-time spectral feature and the spectral data of the historical abnormal pattern is calculated to determine whether the corresponding real-time spectral feature is abnormal data by whether the Euclidean distance is less than the preset abnormal threshold; when the Euclidean distance is less than the preset abnormal threshold, it is determined that the corresponding real-time spectral feature meets the pattern abnormality condition and is determined to be abnormal data.

[0048] In step S105, once the real-time spectral signature is extracted, the Mahalanobis distance between the real-time spectral signature and the mean of historical normal batch spectral data is calculated and compared with a preset normal threshold. If the Mahalanobis distance exceeds the normal threshold, it indicates that the real-time spectral signature deviates from the normal distribution, meeting the numerical anomaly condition, and a warning signal is issued. The preset normal threshold can be set according to actual needs.

[0049] At the same time, the Euclidean distance between the real-time spectral feature and each abnormal pattern in the historical abnormal pattern library is calculated and compared with the preset abnormal threshold. If the Euclidean distance with any historical abnormal pattern is less than the abnormal threshold, it indicates that the real-time spectral feature is similar to the known abnormal pattern and meets the pattern abnormality condition, and an early warning signal is issued. Through this dual judgment mechanism, even if the numerical deviation of the real-time spectral feature is not significant, if its pattern is similar to the historical abnormal pattern, it can be identified, thereby effectively addressing the inherent small batch variability in the low-temperature crystallization process of peptide drugs and the early and rare abnormal patterns with scarce samples in historical data, and achieving rapid and accurate early warning of rare abnormal patterns before the process state deviates significantly.

[0050] In step S105, it is determined whether the real-time spectral feature is abnormal data, and the size relationship between the predicted deviation evaluation result and the preset evaluation threshold is compared. When it is detected that the predicted deviation evaluation result is greater than the preset evaluation threshold, or when it is detected that the real-time spectral feature is abnormal data, an early warning signal is issued.

[0051] By introducing abnormal data judgment criteria based on historical data and comparing the predicted deviation assessment results with a preset assessment threshold, a dual independent anomaly detection dimension is added. This not only detects anomalies based on the deviation between the digital twin model's predicted state and the actual spectral characteristics, but also directly identifies abnormal patterns in the spectral data itself, even if these abnormal patterns have not yet led to significant prediction deviations. This dual judgment mechanism improves the sensitivity and reliability of early warnings, provides more comprehensive coverage of potential anomalies, and resolves the problems caused by relying solely on deviation judgment or unclear abnormal data judgment methods.

[0052] For example, when pre-determining abnormal data, historical normal batch spectral data can be analyzed to establish normal fluctuation ranges for key spectral features (such as specific peak intensity, peak area, and peak position). Simultaneously, historical abnormal batch spectral data can be analyzed to identify spectral patterns associated with anomalies, such as the location and intensity threshold of specific impurity peaks. After acquiring real-time online Raman spectral data from the low-temperature crystallization process of a peptide drug, real-time spectral features are extracted, for example, by calculating the area and position of key characteristic peaks. To determine whether a real-time spectral feature is abnormal, the extracted feature can be compared with a pre-determined normal fluctuation range. If it exceeds the range, it is determined to be a numerical anomaly. Simultaneously, the system checks for features that match historical abnormal patterns. For example, if a signal matching a historical abnormal pattern is detected in a specific wavelength band (using Mahalanobis distance and Euclidean distance), it is determined to be a pattern anomaly. If a real-time spectral feature meets either the numerical anomaly or pattern anomaly conditions, it is determined to be abnormal data. Simultaneously, the system calculates the current prediction deviation assessment result and compares it with a preset assessment threshold. For example, the preset assessment threshold is set to 0.8. If the predicted deviation evaluation result is 0.9, which is greater than 0.8, the warning condition is met. Alternatively, if the real-time spectral signature is judged to be abnormal data, the warning condition is met even if the predicted deviation evaluation result is less than or equal to 0.8. When either condition is met, the system immediately issues a warning signal, such as an audible or visual alarm or a notification to the operator, indicating that there may be an abnormality in the low-temperature crystallization process and that further inspection or intervention is required.

[0053] From the above, it can be seen that the digital twin-based polypeptide drug low-temperature crystallization process early warning method obtains the batch information of the polypeptide drug, as well as the crystallization process data and online spectral data of the polypeptide drug during the low-temperature crystallization process, and inputs the batch information and crystallization process data into the preset digital twin model to simulate the low-temperature crystallization process, and predicts the state parameters of the polypeptide drug at the next moment in the low-temperature crystallization process. The characteristic vector related to the crystallization process is extracted from the online spectral data to obtain the real-time spectral characteristics, and the degree of inconsistency between the key state parameters and the real-time spectral characteristics is calculated to obtain the prediction deviation evaluation result. When it is detected that the prediction deviation evaluation result is greater than the preset evaluation threshold, or the real-time spectral characteristics are abnormal numbers When the data is obtained, an early warning signal is issued; thus, by calculating the degree of inconsistency between the key state parameters and the real-time spectral characteristics, the prediction deviation evaluation result is obtained, and when the prediction deviation evaluation result is greater than the preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued, which solves the problem that the existing early warning method for the low-temperature crystallization process of polypeptide drugs has insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the actual state reflected by the digital twin model prediction and the online spectrum, it is possible to adaptively capture the batch variability and abnormal patterns of the low-temperature crystallization process of polypeptide drugs, thereby improving the early warning efficiency of the low-temperature crystallization process of polypeptide drugs.

[0054] refer to Figure 2 This application provides a digital twin-based peptide drug low-temperature crystallization process early warning system for abnormalities in the peptide drug low-temperature crystallization process, including: Acquisition module 1 is used to obtain batch information of the peptide drug, as well as crystallization process data and online spectral data of the peptide drug during the low-temperature crystallization process; Prediction module 2 is used to input batch information and crystallization process data into a preset digital twin model to simulate the low-temperature crystallization process and predict the state parameters of the peptide drug at the next moment in the low-temperature crystallization process; Extraction module 3, used to extract characteristic vectors related to the crystallization process from the online spectral data to obtain real-time spectral features; Calculation module 4, used to calculate the inconsistency between the key state parameters and the real-time spectral characteristics to obtain the prediction deviation evaluation result; The early warning module 5 is used to issue an early warning signal when it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data.

[0055] This digital twin-based polypeptide drug low-temperature crystallization process early warning system obtains the prediction deviation evaluation result by calculating the degree of inconsistency between key state parameters and real-time spectral characteristics. When the prediction deviation evaluation result is greater than the preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued. This solves the problem that the existing polypeptide drug low-temperature crystallization process early warning method has insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the digital twin model prediction and the actual state reflected by the online spectrum, it can adaptively capture the batch variability and abnormal patterns of the polypeptide drug low-temperature crystallization process, thereby improving the early warning efficiency of the polypeptide drug low-temperature crystallization process.

[0056] Specifically, when executing, Acquisition Module 1 acquires batch information of the peptide drug, as well as crystallization process data and online spectral data during the low-temperature crystallization process of the peptide drug. Batch information includes information such as raw material batch information and solvent batch information. Crystallization process data includes various data reflecting the macroscopic state of the crystallization process, such as temperature, pressure, stirring rate, and feed rate, and can be collected in real time using various sensors. Online spectral data can be acquired using equipment such as online Raman spectrometers. Online spectral data can reflect the vibration and rotation information of the substance molecules and is closely related to the peptide, solvent, impurities, and crystal structure.

[0057] Specifically, when Prediction Module 2 is executed, it utilizes a preset digital twin model to simulate and predict the crystallization process. Batch information and crystallization process data serve as model inputs, enabling the model to reflect the specific state of the current batch and provide key state parameters of the crystallization process for subsequent steps. Key state parameters are important process indicators and include parameters such as peptide concentration and crystal particle size. A digital twin model is a digital model (i.e., a model constructed using digital twin technology) that is highly similar to a physical entity in a virtual space through data collection, transmission, processing, and analysis, modeling and simulation. Specifically, it can include a peptide crystallization kinetics model in a specific solvent system. Digital twin models are prior art and will not be described in detail here.

[0058] Specifically, when extracting the feature vectors related to the crystallization process from the online Raman spectral data to obtain the real-time spectral features, the extraction module 3 executes: Preprocessing the online Raman spectrum data to obtain preprocessed Raman spectrum data; The characteristic peak area related to the peptide concentration, the characteristic peak position related to the solvent composition, the characteristic peak intensity ratio related to the crystal form, and the specific band integral area related to impurities or intermediate products in the crystallization process are extracted from the pre-processed Raman spectral data to construct the corresponding feature vector and obtain real-time spectral features.

[0059] Specifically, when the extraction module 3 preprocesses the online Raman spectrum data to obtain the preprocessed Raman spectrum data, it executes: Performing filtering on the online Raman spectrum data to obtain filtered Raman spectrum data; Using a smoothing algorithm, smoothing the filtered Raman spectrum data to obtain smoothed Raman spectrum data; The least square method is used to correct the fluorescence background interference of the smoothed Raman spectral data to obtain the preprocessed Raman spectral data.

[0060] When extraction module 3 is executed, the raw online Raman spectral data may contain random noise and fluorescence background interference. These factors can obscure the true Raman signal and result in inaccurate feature vectors extracted from the data. To address this issue, the online Raman spectral data is first filtered (e.g., Fourier transform filtering or wavelet filtering) to remove high-frequency noise and obtain noise-reduced data. Next, the filtered data is smoothed (e.g., using algorithms such as Savitzky-Golay smoothing or moving average smoothing) to further reduce data fluctuations and make the spectral curve flatter, facilitating subsequent peak identification and measurement. Finally, the smoothed data is corrected for fluorescence background using the least squares method to isolate the true Raman signal and eliminate the effects of fluorescence interference on peak intensity and area measurements. After these preprocessing steps, high-quality Raman spectral data is obtained. Filtering, smoothing, and the least squares method are conventional techniques and will not be described in detail here.

[0061] When Extraction Module 3 is executed, features extracted from the preprocessed Raman spectral data include characteristic peak areas associated with peptide concentration, characteristic peak positions associated with solvent composition, characteristic peak intensity ratios associated with crystal form, and the integrated areas of specific bands associated with impurities or intermediate products during crystallization. These features represent key information about the crystallization process: changes in peptide content, changes in the solvent environment, crystal structure formation and transformation, and potential side reactions or impurity formation. By extracting these physicochemically meaningful features, high-dimensional spectral data can be converted into low-dimensional, information-rich feature vectors.

[0062] Specifically, when calculating the inconsistency between the key state parameters and the real-time spectral characteristics and obtaining the prediction deviation evaluation result, the calculation module 4 executes: performing standardization processing on the key state parameters and the real-time spectral characteristics respectively to obtain the standardized key state parameters and the standardized real-time spectral characteristics; Based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral characteristics, the weighted deviation of the normalized key state parameters and the normalized real-time spectral characteristics is calculated to obtain the parameter weighted deviation vector and the spectral weighted deviation vector; The parameter weighted deviation vector and the spectral weighted deviation vector are input into the preset deviation prediction model to calculate the prediction deviation evaluation result.

[0063] When executing, calculation module 4 utilizes existing standardization methods, such as the Z-score standardization method, to standardize key state parameters and real-time spectral features. This standardization eliminates differences in the dimensions and numerical ranges of different parameters and features, making different types of data comparable. Standardization is a prior art technique and will not be described in detail here.

[0064] Specifically, when the calculation module 4 calculates the weighted deviation of the normalized key state parameters and the normalized real-time spectral features based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral features, and obtains the parameter weighted deviation vector and the spectral weighted deviation vector, the following is executed: Based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral features, the weights of the key state parameters after the corresponding standardization and the weights of the real-time spectral features after the corresponding standardization are determined and recorded as the first weight and the second weight, respectively; wherein, the greater the uncertainty, the smaller the first weight, and the greater the extraction deviation, the smaller the second weight; Based on the first weight and the second weight, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector.

[0065] Specifically, the calculation module 4 determines the weights of the key state parameters and the weights of the real-time spectral features after normalization based on the uncertainty of the key state parameters and the extraction deviation of the real-time spectral features, and executes: The uncertainty of the key state parameters is determined by inputting specific batch information and crystallization process data into the preset digital twin model multiple times and calculating the variance of the key state parameters; The actual difference of the real-time spectral features calculated by extracting the characteristic vectors related to the crystallization process from the online Raman spectral data multiple times is used to determine the extraction deviation of the real-time spectral features; Based on the uncertainty and extraction deviation, the weights of the corresponding normalized key state parameters and the weights of the corresponding normalized real-time spectral features are determined.

[0066] During execution, calculation module 4 repeats the simulation process (the process of inputting specific batch information and real-time crystallization process data into the digital twin model for simulation) multiple times to quantify the uncertainty of the model prediction. Each simulation may introduce small perturbations or use different model parameter sampling. The variance of the key state parameter results obtained from these multiple simulations is calculated. The variance value reflects the stability of the model prediction, i.e., the uncertainty. Simultaneously, to quantify the deviation of feature extraction, feature extraction (i.e., the step of extracting feature vectors related to the crystallization process from the online Raman spectral data) is repeated multiple times on the same set of raw spectral data. The average of these spectral feature results is calculated, and the actual difference between the real-time spectral features and the average value is calculated. This actual difference reflects the stability of the feature extraction process, i.e., the extraction deviation. Based on the calculated uncertainty (variance) and extraction deviation (actual difference), the weights of the key state parameters and real-time spectral features are determined when calculating the weighted deviation. Key state parameters with large uncertainty are assigned lower weights, and real-time spectral features with large extraction deviations are assigned lower weights. This reduces the impact of uncertain or unreliable data when calculating the weighted deviation between the key state parameters and the real-time spectral features, making the calculation of the weighted deviation more objective and accurate. This more reliable deviation assessment helps to more accurately identify anomalies in the low-temperature crystallization process, especially for small variations between batches or early, rare abnormal patterns. Through more reliable deviation assessment, the sensitivity and accuracy of early warning can be improved.

[0067] When the calculation module 4 is executed, the standardized key state parameters and the standardized real-time spectral characteristics are multiplied by the corresponding weights to calculate the parameter weighted deviation vector and the spectral weighted deviation vector. Each component of the weighted deviation vector represents the difference between the predicted value and the actual value of the corresponding parameter or characteristic.

[0068] The parameter-weighted deviation vector and the spectral-weighted deviation vector are input into a pre-set deviation prediction model, which outputs a prediction deviation assessment result. This assessment reflects the degree of deviation between the current process state and the normal state. The deviation prediction model can be a machine learning model, such as a support vector machine, random forest, or neural network algorithm. This deviation prediction model is used to accurately assess the deviation between the key state parameters predicted by the digital twin model and the spectral features acquired in real time.

[0069] The preset deviation prediction model is a model constructed using historical data to represent the nonlinear relationship between key state parameters and real-time spectral characteristics; the steps of constructing the preset deviation prediction model include: Obtain historical batch information, historical crystallization process data, historical spectral data, and historical nonlinear relationship results between historical crystallization process data and historical spectral data for multiple peptide drugs from a database; Generate corresponding historical key state parameters and historical spectral characteristics based on historical batch information, historical crystallization process data, historical spectral data and historical nonlinear relationship results; Based on historical key state parameters and historical spectral characteristics, a machine learning algorithm is used to construct a preset deviation prediction model.

[0070] Specifically, historical data is collected from a database, including information on peptide drug batches (normal batches), crystallization process data, online Raman spectral data, and known nonlinear relationships between these historical data. This historical data provides the foundational information needed for model training, covering conditions across different batches and process conditions. Based on this historical data, corresponding historical key state parameters and historical spectral features are generated. This step transforms the raw historical data into a form directly related to the deviation to be assessed, enabling the model to learn the corresponding relationships between key state parameters and spectral features. A machine learning algorithm is then used to construct a pre-defined deviation prediction model based on the generated historical key state parameters and historical spectral features. The machine learning algorithm learns and establishes complex nonlinear mappings between key state parameters and real-time spectral features from the historical data. By learning these nonlinear relationships, the model can more accurately understand the associations between key state parameters and real-time spectral features. In practical applications, when input with standardized key state parameters and standardized real-time spectral features, it can output a prediction deviation assessment result that reflects the degree of inconsistency between them.

[0071] Specifically, when the warning module 5 detects that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data, and issues a warning signal, it executes: Based on the historical spectral data in the database, the abnormal data determination conditions are predetermined; Based on the abnormal data determination conditions, determine whether the real-time spectral characteristics are abnormal data, and at the same time, compare the size relationship between the prediction deviation evaluation result and the preset evaluation threshold; When it is detected that the prediction deviation evaluation result is greater than the preset evaluation threshold, or when it is detected that the real-time spectral feature is abnormal data, an early warning signal is issued.

[0072] During execution, the early warning module 5 utilizes the historical spectral data accumulated in the database, analyzes the normal batch data and known abnormal pattern data contained therein, and pre-defines specific conditions for identifying abnormal data. These conditions can be based on the numerical range of spectral features, the presence or absence of specific peaks, peak area ratios, and other factors. This step provides a basis for subsequent real-time judgment.

[0073] Among them, the abnormal data determination conditions include numerical abnormality conditions and pattern abnormality conditions.

[0074] When the early warning module 5 determines whether the real-time spectrum feature is abnormal data based on the abnormal data determination condition, it executes: Extracting spectral data of historical normal batches and spectral data of historical abnormal patterns from historical spectral data; Calculate the Mahalanobis distance between the real-time spectral feature and the spectral data of the historical normal batch, and determine whether the corresponding real-time spectral feature is abnormal data by whether the Mahalanobis distance is greater than the preset normal threshold; when the Mahalanobis distance is greater than the preset normal threshold, determine that the corresponding real-time spectral feature meets the numerical abnormality condition and determine that the corresponding real-time spectral feature is abnormal data; The Euclidean distance between the real-time spectral feature and the spectral data of the historical abnormal pattern is calculated to determine whether the corresponding real-time spectral feature is abnormal data by whether the Euclidean distance is less than the preset abnormal threshold; when the Euclidean distance is less than the preset abnormal threshold, it is determined that the corresponding real-time spectral feature meets the pattern abnormality condition and is determined to be abnormal data.

[0075] When executing, early warning module 5 extracts a real-time spectral signature and calculates the Mahalanobis distance between the real-time spectral signature and the mean of historical normal batch spectral data. This distance is then compared with a preset threshold for normality. If the Mahalanobis distance exceeds the threshold, it indicates that the real-time spectral signature deviates from a normal distribution, meeting the condition for numerical anomaly, and a warning signal is issued. The preset threshold for normality can be set as needed.

[0076] At the same time, the Euclidean distance between the real-time spectral feature and each abnormal pattern in the historical abnormal pattern library is calculated and compared with the preset abnormal threshold. If the Euclidean distance with any historical abnormal pattern is less than the abnormal threshold, it indicates that the real-time spectral feature is similar to the known abnormal pattern and meets the pattern abnormality condition, and an early warning signal is issued. Through this dual judgment mechanism, even if the numerical deviation of the real-time spectral feature is not significant, if its pattern is similar to the historical abnormal pattern, it can be identified, thereby effectively addressing the inherent small batch variability in the low-temperature crystallization process of peptide drugs and the early and rare abnormal patterns with scarce samples in historical data, and achieving rapid and accurate early warning of rare abnormal patterns before the process state deviates significantly.

[0077] In step S105, it is determined whether the real-time spectral feature is abnormal data, and the size relationship between the predicted deviation evaluation result and the preset evaluation threshold is compared. When it is detected that the predicted deviation evaluation result is greater than the preset evaluation threshold, or when it is detected that the real-time spectral feature is abnormal data, an early warning signal is issued.

[0078] By introducing abnormal data judgment criteria based on historical data and comparing the predicted deviation assessment results with a preset assessment threshold, a dual independent anomaly detection dimension is added. This not only detects anomalies based on the deviation between the digital twin model's predicted state and the actual spectral characteristics, but also directly identifies abnormal patterns in the spectral data itself, even if these abnormal patterns have not yet led to significant prediction deviations. This dual judgment mechanism improves the sensitivity and reliability of early warnings, provides more comprehensive coverage of potential anomalies, and resolves the problems caused by relying solely on deviation judgment or unclear abnormal data judgment methods.

[0079] For example, when pre-determining abnormal data, historical normal batch spectral data can be analyzed to establish normal fluctuation ranges for key spectral features (such as specific peak intensity, peak area, and peak position). Simultaneously, historical abnormal batch spectral data can be analyzed to identify spectral patterns associated with anomalies, such as the location and intensity threshold of specific impurity peaks. After acquiring real-time online Raman spectral data from the low-temperature crystallization process of a peptide drug, real-time spectral features are extracted, for example, by calculating the area and position of key characteristic peaks. To determine whether a real-time spectral feature is abnormal, the extracted feature can be compared with a pre-determined normal fluctuation range. If it exceeds the range, it is determined to be a numerical anomaly. Simultaneously, the system checks for features that match historical abnormal patterns. For example, if a signal matching a historical abnormal pattern is detected in a specific wavelength band (using Mahalanobis distance and Euclidean distance), it is determined to be a pattern anomaly. If a real-time spectral feature meets either the numerical anomaly or pattern anomaly conditions, it is determined to be abnormal data. Simultaneously, the system calculates the current prediction deviation assessment result and compares it with a preset assessment threshold. For example, the preset assessment threshold is set to 0.8. If the predicted deviation evaluation result is 0.9, which is greater than 0.8, the warning condition is met. Alternatively, if the real-time spectral signature is judged to be abnormal data, the warning condition is met even if the predicted deviation evaluation result is less than or equal to 0.8. When either condition is met, the system immediately issues a warning signal, such as an audible or visual alarm or a notification to the operator, indicating that there may be an abnormality in the low-temperature crystallization process and that further inspection or intervention is required.

[0080] From the above, it can be seen that the digital twin-based polypeptide drug low-temperature crystallization process early warning system obtains the batch information of the polypeptide drug, as well as the crystallization process data and online spectral data of the polypeptide drug during the low-temperature crystallization process, and inputs the batch information and crystallization process data into the preset digital twin model to simulate the low-temperature crystallization process, and predicts the state parameters of the polypeptide drug at the next moment in the low-temperature crystallization process. The characteristic vector related to the crystallization process is extracted from the online spectral data to obtain the real-time spectral characteristics, and the degree of inconsistency between the key state parameters and the real-time spectral characteristics is calculated to obtain the prediction deviation evaluation result. When it is detected that the prediction deviation evaluation result is greater than the preset evaluation threshold, or the real-time spectral characteristics are abnormal numbers When the data is obtained, an early warning signal is issued; thus, by calculating the degree of inconsistency between the key state parameters and the real-time spectral characteristics, the prediction deviation evaluation result is obtained, and when the prediction deviation evaluation result is greater than the preset evaluation threshold or the real-time spectral characteristics are abnormal data, an early warning signal is issued, which solves the problem that the existing early warning method for the low-temperature crystallization process of polypeptide drugs has insufficient adaptability to batch differences and insufficient ability to recognize abnormal patterns when using digital twins and machine learning methods, making it difficult to issue accurate early warnings. Through the actual state reflected by the digital twin model prediction and the online spectrum, it is possible to adaptively capture the batch variability and abnormal patterns of the low-temperature crystallization process of polypeptide drugs, thereby improving the early warning efficiency of the low-temperature crystallization process of polypeptide drugs.

[0081] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0082] In addition, the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected based on actual needs to achieve the purpose of the solution of this embodiment.

[0083] Furthermore, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0084] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0085] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for early warning of low-temperature crystallization process of polypeptide drugs based on digital twins, which is used to warn of abnormalities in the low-temperature crystallization process of polypeptide drugs, characterized in that: Including steps: Obtaining specific batch information of the peptide drug to be tested, and obtaining real-time crystallization process data and online spectral data of the peptide drug during the low-temperature crystallization process; Inputting the specific batch information and the crystallization process data into a preset digital twin model to simulate the low-temperature crystallization process and predict the key state parameters of the test polypeptide drug at the next moment in the low-temperature crystallization process; Extracting characteristic vectors related to the crystallization process from the online spectral data to obtain real-time spectral characteristics; Calculating the degree of inconsistency between the key state parameter and the real-time spectral feature to obtain a prediction deviation assessment result; When it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data, an early warning signal is issued.

2. The early warning method for low-temperature crystallization of a digital twin polypeptide drug according to claim 1 is characterized in that: Extracting characteristic vectors related to the crystallization process from the online spectral data to obtain real-time spectral features includes: Preprocessing the online spectral data to obtain preprocessed Raman spectral data; The characteristic peak area related to the polypeptide concentration, the characteristic peak position related to the solvent composition, the characteristic peak intensity ratio related to the crystal form, and the specific band integral area related to impurities or intermediate products in the crystallization process are extracted from the pre-processed Raman spectral data to construct corresponding feature vectors and obtain real-time spectral features.

3. The early warning method for low-temperature crystallization of a digital twin polypeptide drug according to claim 2, characterized in that: Preprocessing the online spectral data to obtain preprocessed Raman spectral data includes: Performing filtering on the online spectral data to obtain filtered Raman spectral data; Using a smoothing algorithm, smoothing the filtered Raman spectrum data to obtain smoothed Raman spectrum data; The fluorescence background interference of the smoothed Raman spectrum data is corrected by using the least square method to obtain pre-processed Raman spectrum data.

4. The early warning method for low-temperature crystallization of a digital twin polypeptide drug according to claim 1, characterized in that: Calculating the degree of inconsistency between the key state parameter and the real-time spectral feature to obtain a prediction deviation assessment result includes: performing standardization processing on the key state parameters and the real-time spectral characteristics respectively to obtain standardized key state parameters and standardized real-time spectral characteristics; Based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector; The parameter weighted deviation vector and the spectral weighted deviation vector are input into a preset deviation prediction model to calculate a prediction deviation evaluation result.

5. The early warning method for low-temperature crystallization of a digital twin polypeptide drug according to claim 4 is characterized in that: Based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector, including: Based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, determining the weight of the key state parameter after normalization and the weight of the real-time spectral feature after normalization, which are respectively recorded as a first weight and a second weight; wherein, the greater the uncertainty, the smaller the first weight, and the greater the extraction deviation, the smaller the second weight; Based on the first weight and the second weight, a weighted deviation between the normalized key state parameter and the normalized real-time spectral feature is calculated to obtain a parameter weighted deviation vector and a spectral weighted deviation vector.

6. The method for early warning of low-temperature crystallization process of digital twin polypeptide drugs according to claim 5 is characterized in that: Based on the uncertainty of the key state parameter and the extraction deviation of the real-time spectral feature, determining the weight of the key state parameter after the corresponding normalization processing and the weight of the real-time spectral feature after the corresponding normalization processing, including: Determining the uncertainty of the key state parameter by inputting the specific batch information and the crystallization process data into a preset digital twin model multiple times and calculating the variance of the key state parameter; determining an extraction deviation of the real-time spectral feature by calculating actual differences of the real-time spectral feature by extracting characteristic vectors related to the crystallization process from the online spectral data multiple times; Based on the uncertainty and the extraction deviation, the weight of the corresponding normalized key state parameter and the weight of the corresponding normalized real-time spectral feature are determined.

7. The early warning method for low-temperature crystallization of a digital twin polypeptide drug according to claim 4 is characterized in that: The preset deviation prediction model is a model constructed using historical data to represent the nonlinear relationship between key state parameters and real-time spectral characteristics; The steps of constructing the preset deviation prediction model include: Acquire historical batch information, historical crystallization process data, historical spectral data, and historical nonlinear relationship results between the historical crystallization process data and the historical spectral data of multiple polypeptide drugs from a database; Based on the historical batch information, the historical crystallization process data, the historical spectral data and the historical nonlinear relationship results, generating corresponding historical key state parameters and historical spectral characteristics; According to the historical key state parameters and the historical spectral characteristics, a machine learning algorithm is used to construct the preset deviation prediction model.

8. The early warning method for low-temperature crystallization of a digital twin polypeptide drug according to claim 1 is characterized in that: When it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data, an early warning signal is issued, including: Based on the historical spectral data in the database, the abnormal data determination conditions are predetermined; Based on the abnormal data determination condition, determining whether the real-time spectral feature is abnormal data, and at the same time, comparing the magnitude relationship between the prediction deviation evaluation result and a preset evaluation threshold; When it is detected that the prediction deviation evaluation result is greater than the preset evaluation threshold, or when it is detected that the real-time spectral feature is abnormal data, an early warning signal is issued.

9. The method for early warning of low-temperature crystallization process of a digital twin polypeptide drug according to claim 8, characterized in that: The abnormal data determination conditions include numerical abnormality conditions and pattern abnormality conditions; Determining whether the real-time spectral feature is abnormal data based on the abnormal data determination condition includes: Extracting spectral data of historical normal batches and spectral data of historical abnormal patterns from the historical spectral data; Calculating the Mahalanobis distance between the real-time spectral feature and the spectral data of the historical normal batch, and determining whether the corresponding real-time spectral feature is abnormal data by whether the Mahalanobis distance is greater than a preset normal threshold; when the Mahalanobis distance is greater than the preset normal threshold, determining that the corresponding real-time spectral feature meets the numerical abnormality condition, and determining that the corresponding real-time spectral feature is abnormal data; The Euclidean distance between the real-time spectral feature and the spectral data of the historical abnormal pattern is calculated to determine whether the corresponding real-time spectral feature is abnormal data by determining whether the Euclidean distance is less than a preset abnormal threshold; when the Euclidean distance is less than the preset abnormal threshold, it is determined that the corresponding real-time spectral feature meets the pattern abnormality condition and is determined to be abnormal data.

10. A digital twin-based polypeptide drug low-temperature crystallization process early warning system for abnormalities in the low-temperature crystallization process of polypeptide drugs, characterized by: include: An acquisition module is used to obtain specific batch information of the polypeptide drug to be tested, and to obtain crystallization process data and online spectral data of the polypeptide drug during the low-temperature crystallization process in real time; A prediction module is used to input the specific batch information and the crystallization process data into a preset digital twin model to simulate the low-temperature crystallization process and predict the key state parameters of the test polypeptide drug at the next moment in the low-temperature crystallization process; An extraction module, configured to extract characteristic vectors related to the crystallization process from the online spectral data to obtain real-time spectral features; an extraction module, configured to calculate the degree of inconsistency between the key state parameters and the real-time spectral features, and obtain a prediction deviation assessment result; The early warning module is used to issue an early warning signal when it is detected that the prediction deviation evaluation result is greater than a preset evaluation threshold, or the real-time spectral feature is abnormal data.

Images