Drug dissolution prediction method and device and electronic equipment

Abstract

The invention provides a drug dissolution prediction method and device and electronic equipment, and the method comprises the steps: carrying out the feature extraction of near infrared spectrum data of a to-be-processed drug, and obtaining a spectrum feature; a machine learning model is utilized to predict the dissolution rate of the effective components in the medicine to be treated according to the spectral characteristics, and the machine learning model is constructed by extracting the spectral characteristics of the effective components according to the near infrared spectrum data of the sample medicine and the dissolution rates of the effective components in the sample medicine at different moments. In the implementation process of the scheme, the dissolution rate of the effective components in the to-be-treated medicine is predicted through the constructed machine learning model, the dissolution rate of the medicine can be obtained, the damage rate of a medicine sample is reduced, the method is suitable for detection of precious or special medicines, and through feature extraction and the machine learning model, the detection accuracy is improved. The requirements on complex equipment and medicine treatment are reduced, so that the dissolution rate of the medicine can be quickly predicted, and the detection efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical fields of drug analysis, quality control and drug dissolution detection, and in particular to a drug dissolution prediction method, device and electronic equipment. Background Art

[0002] Dissolution refers to the speed and degree of dissolution of active ingredients from solid preparations (such as tablets, capsules, etc.) under specified conditions. It is usually expressed as the percentage of dissolved drugs per unit time and is an important indicator for evaluating drug absorption and bioavailability in the body. Specifically, dissolution testing usually simulates the dissolution process of drugs in the gastrointestinal tract and evaluates their dissolution characteristics by measuring the amount of drugs released over time in a specific medium.

[0003] At present, the determination of drug dissolution mainly relies on the traditional paddle method and basket method. These methods require the use of ultraviolet spectrophotometry (UV) or high performance liquid chromatography (HPLC) and other technologies. In the specific practice process, it is found that these methods need to destroy the sample during each determination process, which is not suitable for the detection of precious or special samples, and it takes a long time to complete the determination of a sample, especially in the process of multi-time point determination, which is more time-consuming. For the determination personnel, the operation is complicated, and professional determination personnel are required to perform multiple sampling and sample processing. Therefore, the current paddle method and basket method for drug dissolution determination have a high destruction rate and low efficiency. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a drug dissolution prediction method, device and electronic equipment, which are used to improve the problem of high damage rate and low efficiency in measuring drug dissolution.

[0005] The embodiment of the present application provides a method for predicting drug dissolution, including: extracting features from near-infrared spectral data of a drug to be processed to obtain spectral features; using a machine learning model to predict the dissolution of an active ingredient in the drug to be processed based on the spectral features, wherein the machine learning model is constructed by extracting the spectral features of the active ingredient from the near-infrared spectral data of a sample drug, and the dissolution of the active ingredient in the sample drug at different times. In the implementation process of the above scheme, the spectral features of the active ingredient are extracted from the near-infrared spectral data of the sample drug, and the machine learning model is constructed by the dissolution of the active ingredient in the drug to be processed at different times, and the dissolution of the active ingredient in the drug to be processed is predicted, so that the dissolution of the drug to be processed can be obtained, which reduces the destruction rate of the drug sample, is suitable for the detection of precious or special drugs, and through feature extraction and machine learning models, reduces the need for complex equipment and drug processing, so that the dissolution of the drug can be quickly predicted, and the efficiency of detection is improved.

[0006] Optionally, in an embodiment of the present application, feature extraction is performed on the near-infrared spectral data of the drug to be treated, including: performing non-negative matrix decomposition on the near-infrared spectral data of the drug to be treated to obtain decomposed spectral data; performing self-modeling hybrid analysis on the near-infrared spectral data of the drug to be treated to obtain spectral features. In the implementation process of the above scheme, purer and more representative spectral features are extracted from complex spectral data through non-negative matrix decomposition and self-modeling hybrid analysis. These features are more stable in different batches and under different conditions, which helps to improve the generalization ability of the model, thereby improving the accuracy of feature extraction. In addition, since non-negative matrix decomposition and self-modeling hybrid analysis can reduce the dimension of high-dimensional spectral data to low-dimensional features, the computational complexity of data processing is reduced.

[0007] Optionally, in an embodiment of the present application, before using a machine learning model to predict the dissolution of the active ingredient in the drug to be processed according to the spectral features, it also includes: obtaining near-infrared spectral data of the sample drug and the dissolution of the active ingredient in the sample drug at different times; standardizing and correcting the near-infrared spectral data to obtain the corrected spectral data; extracting the spectral features of the active ingredient in the sample drug from the corrected spectral data; using the spectral features of the active ingredient in the sample drug and the dissolution of the active ingredient in the sample drug at different times to construct a machine learning model. In the implementation process of the above scheme, by standardizing and correcting the near-infrared spectral data and extracting features, noise and interference can be removed, key information can be retained, thereby improving the prediction accuracy of the model, and using the spectral data after standardization and correction to remove noise and interference and the dissolution of the active ingredient at different times to construct a machine learning model can improve the generalization ability of the model under different conditions. Further, through standardization and feature extraction, the data dimension can be reduced, the data processing process can be simplified, the computational complexity can be reduced, the influence of random errors and systematic errors can be reduced, and the stability and robustness of the model can be improved.

[0008] Optionally, in an embodiment of the present application, obtaining near-infrared spectral data of sample drugs includes: producing multiple sample drugs through drug manufacturing equipment with different production processes and formulations; scanning multiple sample drugs at the same time through a near-infrared spectral scanner to obtain near-infrared spectral data of the sample drugs. In the implementation of the above scheme, by using sample drugs with different production processes and formulations, more variability and complexity can be captured, thereby improving prediction accuracy, and diversified sample drug data can help the model better identify and deal with various interference factors, thereby improving the robustness of the model.

[0009] Optionally, in an embodiment of the present application, constructing a machine learning model includes: dividing the spectral characteristics of the active ingredient in the sample drug and the dissolution rate of the active ingredient in the sample drug at different times into a validation set and a calibration set; using multiple machine learning algorithms to construct multiple calibration models based on the calibration set; using the validation set to evaluate the performance of the multiple calibration models, and determining a calibration model with the best performance screened out from the multiple calibration models as the machine learning model. In the implementation process of the above scheme, by dividing the validation set and the calibration set, and using the validation set to evaluate the performance of multiple calibration models and the screening model, the risk of model overfitting can be effectively reduced, ensuring that the final selected model has high accuracy and reliability.

[0010] Optionally, in an embodiment of the present application, a correction model with the best performance selected from multiple correction models is determined as a machine learning model, including: using parameter indicators of the correction set, parameter indicators of the validation set and / or preset performance indicators to screen multiple correction models to obtain the correction model with the best performance. In the implementation process of the above scheme, through the evaluation and screening of parameter indicators of the correction set, parameter indicators of the validation set and / or preset performance indicators, the generalization ability of the model under different conditions can be ensured, and the applicability and generalization of the model can be improved.

[0011] Optionally, in an embodiment of the present application, the machine learning model includes: a partial least squares regression PLSR model, a back propagation neural network BPNN model, or a support vector regression SVR model.

[0012] The embodiment of the present application also provides a drug dissolution prediction device, including: a spectral feature acquisition module, used to extract features from near-infrared spectral data of a drug to be processed to obtain spectral features; a drug dissolution prediction module, used to predict the solubility of an active ingredient in the drug to be processed based on the spectral features using a machine learning model, wherein the machine learning model is constructed by extracting the spectral features of the active ingredient based on the near-infrared spectral data of the sample drug, and the solubility of the active ingredient in the sample drug at different times.

[0013] Optionally, in an embodiment of the present application, the spectral feature acquisition module includes: a spectral data decomposition submodule, used to perform non-negative matrix decomposition on the near-infrared spectral data of the drug to be treated to obtain decomposed spectral data; a data mixing analysis submodule, used to perform self-modeling mixing analysis on the near-infrared spectral data of the drug to be treated to obtain spectral features.

[0014] Optionally, in an embodiment of the present application, the drug dissolution prediction device also includes: a sample drug data acquisition module, used to obtain near-infrared spectral data of the sample drug, and the solubility of the active ingredient in the sample drug at different times; a near-infrared spectral data correction module, used to standardize the near-infrared spectral data to obtain corrected spectral data; a drug spectral feature extraction module, used to extract the spectral features of the active ingredient in the sample drug from the corrected spectral data; a machine learning model construction module, used to use the spectral features of the active ingredient in the sample drug and the solubility of the active ingredient in the sample drug at different times to construct a machine learning model.

[0015] Optionally, in an embodiment of the present application, the sample drug data acquisition module includes: a sample drug generation submodule, which is used to produce multiple sample drugs through drug manufacturing equipment with different production processes and formulas; and a sample drug scanning submodule, which is used to scan multiple sample drugs at the same time through a near-infrared spectral scanner to obtain near-infrared spectral data of the sample drugs.

[0016] Optionally, in an embodiment of the present application, the machine learning model construction module includes: a validation and correction set division submodule, which is used to divide the spectral characteristics of the active ingredients in the sample medicine and the solubility of the active ingredients in the sample medicine at different times into a validation set and a correction set; a correction model construction submodule, which is used to use multiple machine learning algorithms to construct multiple correction models according to the correction set; and a correction model screening submodule, which is used to evaluate the performance of multiple correction models using the validation set, and determine a correction model with the best performance screened out from the multiple correction models as the machine learning model.

[0017] Optionally, in an embodiment of the present application, the correction model screening submodule includes: a correction model screening unit, which is used to screen multiple correction models using parameter indicators of the correction set, parameter indicators of the verification set and / or preset performance indicators.

[0018] Optionally, in an embodiment of the present application, the machine learning model includes: a partial least squares regression PLSR model, a back propagation neural network BPNN model, or a support vector regression SVR model.

[0019] An embodiment of the present application further provides an electronic device, including: a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the machine-readable instructions execute the method described above when executed by the processor.

[0020] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described above is executed.

[0021] The embodiment of the present application also provides a computer program product, including: a computer program or a computer instruction, and the computer program or the computer instruction executes the method described above when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 The schematic diagram of the process of the drug dissolution prediction method provided in the embodiment of the present application is shown; Figure 2 A schematic diagram of a process for acquiring near-infrared spectral data of a sample drug provided in an embodiment of the present application is shown; Figure 3 The schematic diagram of the structure of the drug dissolution prediction device provided in the embodiment of the present application is shown; Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is shown.

[0024] Icons: drug dissolution prediction device-200; spectral feature acquisition module-210; drug dissolution prediction module-220; electronic device-300; processor-310; memory-320; storage medium-330. DETAILED DESCRIPTION

[0025] To make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the drawings in the embodiment of the present application. It should be understood that the drawings in the embodiment of the present application only serve the purpose of explanation and description, and are not used to limit the protection scope of the embodiment of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in the embodiment of the present application shows the operation implemented according to some embodiments of the embodiment of the present application. It should be understood that the operation of the flowchart can be implemented out of order, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the embodiment of the present application, or remove one or more operations from the flowchart.

[0026] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments. The components of the embodiments of the present application described and shown in the drawings generally here can be arranged and designed in various 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 embodiments of the present application claimed, but merely represents the selected embodiments of the embodiments of the present application.

[0027] It is understandable that the "first" and "second" in the embodiments of the present application are used to distinguish similar objects. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity and execution order, and the words "first" and "second" do not necessarily limit the difference. In the description of the embodiments of the present application, the term "and / or" is merely a description of the association relationship of the associated objects, indicating that there may be three relationships, such as A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the related objects before and after are in an "or" relationship. The term "multiple" refers to more than two (including two), and similarly, "multiple groups" refers to more than two groups (including two groups).

[0028] It should be noted that the drug dissolution prediction method provided in the embodiment of the present application can be executed by an electronic device, where the electronic device refers to a device terminal or server with the function of executing a computer program, such as a smart phone, a personal computer, a tablet computer, a personal digital assistant or a mobile Internet device. A server refers to a device that provides computing services through a network, such as an x86 server and a non-x86 server, and a non-x86 server includes a mainframe, a minicomputer and a UNIX server.

[0029] See also Figure 1 The schematic diagram of the process of the drug dissolution prediction method provided in the embodiment of the present application is shown; the embodiment of the present application provides a drug dissolution prediction method, comprising: Step S110: extracting features from the near-infrared spectrum data of the drug to be processed to obtain spectrum features.

[0030] The drug to be processed refers to the drug whose dissolution needs to be predicted. The drug can be in the form of tablets or capsules. The above-mentioned drug to be processed can be specifically a Chinese medicine compound preparation, such as Yinqiao Jiedu Tablets, Ganmao Qingre Tablets and Niuhuang Jiedu Tablets, etc. Of course, the drug can also be a Western medicine, such as aspirin tablets and ibuprofen capsules. For the sake of ease of understanding and explanation, the following is an example of Yinqiao Jiedu Tablets. Yinqiao Jiedu Tablets are mainly made of honeysuckle, mint, forsythia, schizonepeta, light black beans, burdock seeds, light bamboo leaves, platycodon, licorice, etc.

[0031] Near infrared spectroscopy data refers to the data obtained by using near infrared spectroscopy technology to process the drugs to be processed, such as using a near infrared spectroscopy scanner to scan the drugs to be processed. Among them, near infrared spectroscopy technology does not require pre-treatment steps for samples, avoiding the situation of damaging samples, and can penetrate glass and plastic packaging for direct scanning, achieving the effect of non-destructive testing.

[0032] Step S120: Use a machine learning model to predict the dissolution rate of the active ingredient in the drug to be processed based on the spectral characteristics. The machine learning model is constructed by extracting the spectral characteristics of the active ingredient based on the near-infrared spectral data of the sample drug, as well as the dissolution rate of the active ingredient in the sample drug at different times.

[0033] The active ingredients in the above-mentioned drugs to be treated refer to the ingredients that are effective in improving the condition of the patient. For example, the main active ingredients in Yinqiao Jiedu Tablets are chlorogenic acid and forsythiaside.

[0034] The above-mentioned machine learning model is a model constructed by extracting the spectral characteristics of the active ingredient based on the near-infrared spectral data of the sample drug and the dissolution rate of the active ingredient in the sample drug at different times. It can be a neural network model or a regression model.

[0035] It is understandable that compared with the traditional paddle method and basket method that require sample destruction, the use of near-infrared technology can penetrate glass and plastic packaging for direct scanning to obtain near-infrared spectral data, and the machine learning model can predict the dissolution rate of the active ingredients in the drug to be processed based on the spectral features extracted from the near-infrared spectral data, which can effectively improve the efficiency of dissolution and is therefore suitable for the dissolution rate detection of a large number of pharmaceutical tablets. In addition, In the process of implementing the above scheme, the spectral characteristics of the active ingredient are extracted from the near-infrared spectral data of the sample drug, and the machine learning model is constructed based on the dissolution of the active ingredient in the sample drug at different times. The dissolution of the active ingredient in the drug to be treated is predicted, and the solubility of the drug to be treated can be obtained, which reduces the destruction rate of the drug sample and is suitable for the detection of precious or special drugs. In addition, through feature extraction and machine learning models, the need for complex equipment and drug processing is reduced, so that the solubility of the drug can be quickly predicted, thereby improving the efficiency of detection.

[0036] As an optional implementation of the above step S110, the above implementation of feature extraction of near infrared spectrum data of the drug to be processed may include: Step S111: performing non-negative matrix decomposition on the near-infrared spectrum data of the drug to be processed to obtain decomposed spectrum data.

[0037] The implementation method of the above step S111 is, for example: using a non-negative matrix factorization (NMF) algorithm to perform non-negative matrix factorization on the near-infrared spectrum data of the drug to be treated to obtain the decomposed spectrum data. Among them, NMF is an effective algorithm for approximating a high-dimensional matrix from two low-dimensional matrices. NMF can mathematically decompose a non-negative matrix Vm×n into two non-negative matrices, and achieve the approximate error through a multiplication iterative update algorithm.

[0038] Step S112: Perform self-modeling hybrid analysis on the near-infrared spectrum data of the drug to be processed to obtain spectral features.

[0039] The implementation method of the above step S112 is, for example: using the Self-Modelling Mixture Analysis (SMMA) algorithm to perform self-modeling mixture analysis on the near-infrared spectral data of the drug to be treated to obtain spectral features. Among them, SMMA is used to improve the problem that NMF for ternary mixtures or more complex mixtures, random initialization without restrictions requires a large number of iterations, and it is difficult to obtain a unique spectral decomposition with a physical interpretation. After standardization and correction of baseline drift, the spectrum extracted by SMMA is highly consistent with the measured spectrum, which greatly facilitates the rapid identification of the components of the mixture.

[0040] As an optional implementation of the above-mentioned drug dissolution prediction method, before using the machine learning model to predict the dissolution of the active ingredient in the drug to be processed according to the spectral characteristics, the machine learning model can also be trained. The implementation includes: Step S101: Acquire near infrared spectrum data of a sample drug and the dissolution rate of an active ingredient in the sample drug at different times.

[0041] The above-mentioned sample drugs can be pre-prepared samples or drug samples, or they can be drugs produced in real time by a pharmaceutical factory, such as samples made from Yinqiao Jiedu tablets using different tableting processes and formulation conditions in the pharmaceutical factory. The active ingredients in Yinqiao Jiedu tablets are chlorogenic acid and forsythiaside.

[0042] See also Figure 2The schematic diagram of the acquisition process of the near-infrared spectral data of the sample drug provided by the embodiment of the present application is shown; the acquisition process of the near-infrared spectral data of the sample drug in the above step S101 is, for example: before collecting the near-infrared spectral data, in order to achieve a better collection effect, the collection environment can be first allowed to meet certain collection conditions, where the collection conditions are, for example, temperature 25°C, humidity 30%, etc. In the production environment of a pharmaceutical factory, the above-mentioned collection conditions are usually met. First, a plurality of sample drugs are produced by drug manufacturing equipment with different production processes and formulas, such as using a tablet press of a pharmaceutical factory to prepare the drug raw materials input into the feed port, and obtain a plurality of prepared Yinqiao Jiedu tablets, then, a plurality of Yinqiao Jiedu tablets fall onto the conveyor belt through the tablet outlet, and then are transmitted by the conveyor belt to the target scanning area of ​​the near-infrared spectrometer, and finally, the near-infrared spectrometer is used to scan the plurality of Yinqiao Jiedu tablets in the target scanning area multiple times at the same time, and the average spectrum of the near-infrared spectra of the multiple scans is calculated, and the average spectrum is determined as the near-infrared spectral data of the sample drug.

[0043] It is understandable that the reason for the above-mentioned multiple scans is to reduce random errors. Due to various random errors in the measurement process (such as instrument noise, environmental interference, etc.), the spectral data obtained by a single scan may not be stable enough. Therefore, by performing multiple scans and averaging the near-infrared spectral results of multiple scans, the noise caused by these random errors can be effectively reduced, thereby improving the stability, accuracy and reliability of the near-infrared spectral data.

[0044] The above-mentioned near-infrared spectrometer is mainly composed of a light source, a sample chamber, an optical system, a spectral analysis system and a data processing system. The process of using the near-infrared spectrometer to scan the Yinqiao Jiedu tablets is, for example: first, the light source emits near-infrared light, and the near-infrared light is focused on the Yinqiao Jiedu tablets through the optical system. After the Yinqiao Jiedu tablets absorb part of the light, the remaining light is reflected, scattered or transmitted to the spectral analysis system, and the spectral analysis system analyzes the remaining light to obtain the analysis results. Finally, the analysis results are processed by the data processing system to obtain the near-infrared spectrum data of the above-mentioned sample drugs. Among them, the near-infrared spectrum data of the sample drugs mainly records the frequency doubling and sum frequency absorption of the vibration of hydrogen-containing groups (such as CH, NH, OH), and the absorption wavelengths and intensities of different groups or the same group in different chemical environments are significantly different.

[0045] The process of obtaining the dissolution of the active ingredient in the sample drug at different times in the above step S101 is, for example: a hydrochloric acid solution with a pH of 1.2 can be used as a dissolution medium, and the in vitro dissolution test of Yinqiao Jiedu tablets under different tableting processes and formulation conditions is performed by the slurry method specified in the pharmacopoeia. Among them, the volume of the above dissolution medium can be 900ml, and its temperature can be 37.5°C, and it can be stored in a dissolution cup with a stirring speed of 100rpm. During the in vitro dissolution test, the dissolution of 5min, 15min, 30min, 45min, 60min, 90min, 120min, 150min, 210min, 270min, 330min, 420min, and 510min can be measured respectively, and then the dissolution solution of each time period is centrifuged, and the dissolution solution is filtered using a 0.22um filter membrane to prepare a test solution for Yinqiao Jiedu tablets. Finally, the test solution in the Yinqiao Jiedu tablets was measured using a high performance liquid chromatography method to obtain the dissolution of its active ingredients (chlorogenic acid and forsythiaside) at each time point. Optionally, the dissolution of the active ingredients in the Yinqiao Jiedu tablets at each time point can be plotted to obtain the dissolution curve of the active ingredients in the Yinqiao Jiedu tablets at different times.

[0046] Step S102: performing standardization correction on the near-infrared spectrum data to obtain corrected spectrum data.

[0047] It is understandable that due to the complexity of the components of Chinese medicine tablets, in addition to having a wide range of absorption characteristics in the near-infrared region, multi-component Chinese medicine tablets may exhibit component spectral overlap and baseline drift in the near-infrared spectrum. Therefore, after obtaining the near-infrared spectral data of Yinqiao Jiedu Tablets, the near-infrared spectral data can be standardized and corrected first.

[0048] The implementation method of the above step S102 is, for example: using the spectral scattering correction processing methods such as Multiplicative Scatter Correction (MSC) and Standard Normal Variate (SNV) provided by TQ Analyst 9.7 software, the spectral background subtraction method of derivative calculation, and the spectral smoothing and noise filtering method of Savitzky-Golay convolution smoothing and norris derivative smoothing to perform first-order derivative correction, normalization correction, standardization correction, convolution smoothing correction and other pre-processing operations on the near-infrared spectral data, effectively removing the outliers and noise in the near-infrared spectral data. Among them, the normalization correction or standardization correction of the near-infrared spectral data can eliminate the influence of dimension and order of magnitude.

[0049] Step S103: extracting the spectral characteristics of the active ingredients in the sample medicine from the corrected spectral data.

[0050] The implementation method of the above step S103 is, for example: using a non-negative matrix factorization (NMF) algorithm to perform non-negative matrix factorization on the corrected spectral data to obtain the decomposed spectral data. Then, the self-modeling mixture analysis (SMMA) algorithm is used to extract features from the corrected spectral data to obtain the spectral characteristics of the active ingredients in the sample medicine, such as the spectral characteristics of chlorogenic acid and forsythiaside in Yinqiao Jiedu Tablets. In the standard NMF algorithm, it is usually started from the random initialization of the concentration matrix Hk×n. For binary mixtures, multiple iterations starting from random initialization can obtain the optimal decomposition. However, for ternary mixtures or more complex mixtures, unrestricted random initialization requires a large number of iterations, and it is difficult to obtain a unique spectral decomposition with a physical interpretation. In order to improve the random initialization problem of NMF, SMMA uses a pure variable method of primary processing to initialize NMF. Since the intensity of the pure variable is proportional to the content of the corresponding component, the initialization is more intuitive and can be reasonably explained. Therefore, the use of a pure variable method to initialize NMF can obtain a unique decomposition result.

[0051] Compared with the NMF algorithm, the SMMA algorithm mentioned above is more suitable for extracting component spectra from multi-component mixtures. After standardization and correction of baseline drift, the spectra extracted by SMMA are highly consistent with the measured spectra, which greatly facilitates the rapid identification of the components of the mixture. Obtaining the number of components k from an unknown mixture is a key prerequisite for further analysis. Since the ordered singular values ​​(SVs) of the original spectral matrix Vm×n provide a quantitative estimate of the representation in the data matrix, they are used to determine the number of component spectra (k). It can usually be considered that the first few significant SVs are related to the available components, while the subsequent smaller SVs are attributed to the contribution of noise. After the number of components k is determined by singular value decomposition (SVD) of the near-infrared spectra of Chinese medicine tablets, the first k pure variables are selected from the spectral matrix Vm×n. Then the initialization of the concentration matrix Hk×n is set to the normalized column value of Vm×n, and the pure variable method of SMMA primary processing is used to initialize NMF to obtain a unique decomposition result, that is, the spectra of different components in Chinese medicine tablets are extracted from the complex tablet mixture, and compared with the unique vibration characteristics of the substance in the near-infrared region, the components in the mixture can be accurately and quickly identified.

[0052] Step S104: construct a machine learning model using the spectral characteristics of the active ingredients in the sample medicine and the dissolution rates of the active ingredients in the sample medicine at different times.

[0053] The above-mentioned machine learning models may include: Partial Least Squares Regression (PLSR) model, Back Propagation Neural Network (BPNN) model or Support Vector Regression (SVR) model. Therefore, the construction process of these three machine learning models is described below.

[0054] The first implementation of the above step S104 is to construct a PLSR model, for example: the PLSR model is an analysis model in statistics, which associates the dissolution (component content or properties) of the active ingredients in the sample medicine at different times with the spectral characteristics of the active ingredients in the sample medicine, and can obtain the maximum number of principal components related to the dissolution amount of each active ingredient in the sample medicine and the spectral characteristics of each active ingredient. The PLSR modeling algorithm is used to establish a PLSR model according to the maximum number of principal components here, and the PLSR model is determined as the above-mentioned dissolution prediction model. The dissolution content analysis model established here is used to predict the dissolution (component content or properties) of unknown drugs to be processed. Among them, the PLSR model uses the idea of ​​information decomposition to obtain the optimal combination factor to explain the dependent variable. When the variables are highly linearly correlated internally, PLSR is particularly effective, and the relevant confounding factors are further eliminated, thereby obtaining a regression result with strong robustness and high accuracy.

[0055] The second implementation of the above step S104 is to construct the SVR model. For example, support vector regression (SVR) is a machine learning regression prediction algorithm model. The spectral characteristics of the active ingredient in the sample drug can be determined as training data, and the true value of the dissolution of the active ingredient in the sample drug at different times is used as a training label. The SVR model is trained as a machine learning model to obtain a trained SVR model. The spectral characteristics of the active ingredient in the sample drug are nonlinearly mapped through the SVR model, thereby mapping it to a high-dimensional linear space, and then linear regression is performed in the space to obtain the dissolution prediction value of the active ingredient in the sample drug. Finally, the loss value between the dissolution prediction value and the true value of the dissolution in the training label is calculated, and the model parameters of the SVR model are updated according to the loss value. Among them, the core idea of ​​the SVR model is the principle of structural risk minimization. By mapping to a high-dimensional linear space to find a regression model from features to targets, SVR can also achieve accurate predictions in complex small and medium-scale nonlinear problems.

[0056] The third implementation of the above step S104, the process of constructing the BPNN model is, for example: BPNN is a typical multi-layer forward neural network, which is a neural network composed of an input layer, an implicit layer and an output layer. The spectral characteristics of the active ingredient in the sample drug are used as training data, and the true value of the dissolution of the active ingredient in the sample drug at different times is used as a training label, and the BPNN model is trained to obtain the trained BPNN model. Finally, the trained BPNN model is determined as the above-mentioned machine learning model. Among them, the specific process of training the BPNN model is, for example: the spectral characteristics of the active ingredient in the sample drug are input into the BPNN model to obtain the predicted value of the dissolution of the active ingredient in the sample drug at different times, and then the loss value between the predicted value of the dissolution and the true value of the dissolution in the training label is calculated, and the model parameters of the BPNN model are updated according to the loss value. Among them, the essence of BPNN is to attribute the error between the network output and the expected output to the problem of weights and thresholds, and to divide the error into the weights and thresholds of each neuron through back propagation. The guiding ideology of BPNN is to adjust the weights and thresholds in the direction where the error function decreases fastest (negative gradient direction).

[0057] As a fourth implementation of the above step S104, the construction process of three machine learning models is described above. In practice, the above first to third implementations can also be combined to construct three models, namely, a PLSR model, an SVR model and a BPNN model. Then, the model with the best performance is selected from the three models to be determined as the machine learning model for predicting drug dissolution. This implementation may include: Step S104a: Divide the spectral characteristics of the active ingredients in the sample medicine and the dissolution rates of the active ingredients in the sample medicine at different times into a verification set and a calibration set.

[0058] The implementation method of the above step S104a is, for example: the spectral characteristics of the active ingredients in multiple sample drugs and the dissolution rates of the active ingredients in multiple sample drugs at different times are packaged as a training data set. For example, the spectral characteristics of the active ingredients in the first sample drug and the dissolution rates of the active ingredients in the first sample drug at different times are used as the first sample, and so on to obtain a packaged training data set. Then, the training data set is divided into a validation set and a calibration set according to a preset ratio. Among them, the preset ratio here can be set according to demand, for example, the preset ratio is set to 1:4. Specifically, assuming that there are 135 samples, the first sample can be used as the first sample in the prediction set, and then every 4 samples are separated, and the next sample is included in the prediction set. Finally, 27 samples are included in the prediction set, and the remaining samples are divided into the calibration set.

[0059] Step S104b: Use multiple machine learning algorithms to construct multiple correction models based on the correction set.

[0060] For example, the implementation method of the above step S104b is as follows: It can be understood that the calibration set includes: the spectral characteristics of the active ingredient in the sample drug extracted from the corrected spectral data, and the solubility of the active ingredient in the sample drug at different times. The spectral characteristics here and the solubility of the active ingredient in the sample drug at different times can be used to construct three calibration models, namely, the PLSR model, the SVR model and the BPNN model. The calibration model here refers to the machine learning model to be calibrated. The specific construction method of each calibration model refers to the first to third embodiments above, which will not be repeated here.

[0061] Step S104c: Use the validation set to evaluate the performance of multiple correction models, and select a correction model with the best performance from the multiple correction models to determine it as the machine learning model.

[0062] As an optional implementation of the above step S104c, the above implementation of determining a correction model with the best performance selected from multiple correction models as a machine learning model may include: using parameter indicators of the correction set, parameter indicators of the validation set and / or preset performance indicators to screen multiple correction models to obtain the correction model with the best performance. The following discusses the parameter indicators of the correction set, the parameter indicators of the validation set and / or the preset performance indicators.

[0063] The first implementation of the above step S104c is, for example, to use the parameter index of the calibration set to represent the prediction error of the three models for the calibration set, which can help reflect the degree of fit between the model and the spectral data. Here, the parameter index of the calibration set includes: the calibration correlation coefficient (Rc) index and the root mean square error (RMSEC) index. The model whose Rc index is greater than the first preset index threshold and / or the RMSEC index is less than the second preset index threshold can be screened out from the three models. Among them, the calibration correlation coefficient Rc index can be expressed as:

[0064] in, Represents the calibration correlation coefficient Rc index. The closer the Rc index is to 1, the more reasonable the calibration model is. is the number of samples in the calibration set, is the sample drug chemistry determination value in the calibration set, is the cross-validation prediction value of the model on the calibration set, express The average value of yes The average value of .

[0065] The above RMSEC index can be expressed as:

[0066] in, Represents the RMSEC index. The smaller the RMSEC index, the more reasonable the calibration model is. is the number of samples in the calibration set, is the sample drug chemistry determination value in the calibration set, is the cross-validation prediction value of the model on the calibration set.

[0067] The second implementation of the above step S104c is, for example, to use the parameter index of the validation set to represent the prediction error of the three models for the calibration set. The parameter index of the validation set here includes: the validation correlation coefficient (Rp) index and the root mean square error of the prediction (RMSEP) index. The model whose Rp index is greater than the third preset index threshold and / or whose RMSEP index is less than the fourth preset index threshold can be screened out from the three models. The above validation correlation coefficient Rp index can be expressed as:

[0068] in, Represents the verification correlation coefficient Rp index. The closer the Rp index is to 1, the more reasonable the verification model is. is the number of samples in the validation set, is the drug chemistry determination value of samples in the validation set, is the cross-validation prediction value of the model on the validation set, express The average value of yes The average value of .

[0069] The above RMSEP index can be expressed as:

[0070] in, Represents the RMSEP index. The smaller the RMSEP index, the better the prediction performance of the validation model. is the number of samples in the validation set, is the drug chemistry determination value of samples in the validation set, is the cross-validation prediction value of the model on the validation set.

[0071] The third implementation of the above step S104c is, for example: using a preset performance indicator to represent the reliability of the prediction performance of the three models, specifically using the mean relative error (MRE) indicator to evaluate the prediction performance (i.e., accuracy) of the three models, and selecting a model whose MRE indicator is less than a fifth preset indicator threshold from the three models. The above MRE indicator can be expressed as:

[0072] in, represents the MRE index, is the number of samples in the validation set, is the drug chemistry determination value of samples in the validation set, is the cross-validation prediction value of the model on the validation set.

[0073] The fourth implementation of the above step S104c is combined with at least two implementations from the first to the third implementations above, that is, any three or more of the above Rc index, RMSEC index, Rp index, RMSEP index and MRE index are combined to screen the three models to obtain the above machine learning model for predicting drug dissolution.

[0074] Optionally, after building and training the machine learning model, a cross-validation method (such as Holdout cross-validation) can be used to evaluate the generalization ability of the machine learning model to avoid overfitting.

[0075] As an optional implementation of the above step S120, the above machine learning model may include: a partial least squares regression (PLSR) model, a back propagation neural network (BPNN) model or a support vector regression (SVR) model. Therefore, the following describes the process of using one of the three machine learning models to predict the dissolution rate of the active ingredient in the drug to be processed.

[0076] As a first optional implementation of the above step S120, using the PLSR model to predict the dissolution rate of the active ingredient in the drug to be processed may include: Step S121: using the PLSR model to predict the dissolution rate of the active ingredient in the drug to be processed according to the spectral characteristics.

[0077] An implementation example of the above step S121 is: using the PLSR model to decompose the spectral features to obtain the optimal combination factor of the spectral features. If the optimal combination factor characterizes that the spectral features are highly linearly correlated, then the PLSR model is used to calculate the solubility of the active ingredients in the drug to be treated corresponding to the spectral features. Specifically, the solubility of chlorogenic acid and forsythiaside in Yinqiao Jiedu Tablets can be predicted to obtain the component content or properties of the solubility of chlorogenic acid and forsythiaside.

[0078] As a second optional implementation of the above step S120, using the SVR model to predict the dissolution rate of the active ingredient in the drug to be processed may include: Step S122: using the SVR model to predict the dissolution rate of the active ingredient in the drug to be processed according to the spectral characteristics.

[0079] The implementation of the above step S122 is, for example: using the SVR model to perform nonlinear mapping on the spectral characteristics of the active ingredient in the drug to be processed, thereby mapping it into a high-dimensional linear space, and then performing linear regression in the space to obtain the dissolution rate of the active ingredient in the drug to be processed. The kernel function of the SVR model can adopt a radial basis function (RBF) kernel function, and specifically can use the RBF function implemented in the LIBSVM library.

[0080] As a third optional implementation of the above step S120, using the BPNN model to predict the dissolution rate of the active ingredient in the drug to be processed may include: Step S123: using the BPNN model to predict the dissolution rate of the active ingredient in the drug to be processed according to the spectral characteristics.

[0081] The implementation method of the above step S123 is as follows: BPNN is a typical multi-layer feedforward neural network, which is a neural network composed of an input layer, an implicit layer and an output layer. First, the spectral features are input to the input layer to obtain the feature map calculated by the input layer, and then the feature map is passed to the implicit layer by the input layer. The neurons of the implicit layer can use an s-type transfer function to calculate the implicit features according to the feature map, and then pass the implicit features to the output layer. Finally, the output layer outputs the solubility of the active ingredient in the drug to be treated, and the neurons of the output layer can use a linear transfer function. Among them, See also Figure 3 The schematic diagram of the structure of the drug dissolution prediction device provided in the embodiment of the present application is shown; the embodiment of the present application provides a drug dissolution prediction device 200, comprising: The spectral feature acquisition module 210 is used to extract features from the near-infrared spectral data of the drug to be processed to obtain spectral features.

[0082] The drug dissolution prediction module 220 is used to predict the dissolution rate of the active ingredient in the drug to be processed based on the spectral characteristics using a machine learning model. The machine learning model is constructed by extracting the spectral characteristics of the active ingredient based on the near-infrared spectral data of the sample drug, as well as the dissolution rate of the active ingredient in the sample drug at different times.

[0083] As an optional implementation of the above device, the spectral feature acquisition module includes: The spectral data decomposition submodule is used to perform non-negative matrix decomposition on the near-infrared spectral data of the drug to be treated to obtain the decomposed spectral data.

[0084] The data mixing analysis submodule is used to perform self-modeling mixing analysis on the near-infrared spectral data of the drug to be treated to obtain spectral characteristics.

[0085] As an optional embodiment of the above device, the drug dissolution prediction device further includes: The sample drug data acquisition module is used to acquire the near-infrared spectrum data of the sample drug and the dissolution rate of the active ingredient in the sample drug at different times.

[0086] The near-infrared spectrum data correction module is used to perform standardization correction on the near-infrared spectrum data to obtain corrected spectrum data.

[0087] The drug spectral feature extraction module is used to extract the spectral features of the active ingredients in the sample drug from the corrected spectral data.

[0088] The machine learning model building module is used to build a machine learning model using the spectral characteristics of the active ingredients in the sample medicine and the dissolution rate of the active ingredients in the sample medicine at different times.

[0089] As an optional implementation of the above device, the sample drug data acquisition module includes: The sample drug generation submodule is used to produce multiple sample drugs through drug manufacturing equipment with different production processes and formulations.

[0090] The sample drug scanning submodule is used to scan multiple sample drugs at the same time through a near-infrared spectrum scanner to obtain near-infrared spectrum data of the sample drugs.

[0091] As an optional implementation of the above device, the machine learning model building module includes: The verification and calibration set division submodule is used to divide the spectral characteristics of the active ingredients in the sample medicine and the dissolution rate of the active ingredients in the sample medicine at different times into a verification set and a calibration set.

[0092] The correction model construction submodule is used to construct multiple correction models according to the correction set using multiple machine learning algorithms.

[0093] The correction model screening submodule is used to evaluate the performance of multiple correction models using a validation set, and to determine a correction model with the best performance screened out from the multiple correction models as a machine learning model.

[0094] As an optional implementation of the above device, the correction model screening submodule includes: The calibration model screening unit is used to screen multiple calibration models using parameter indicators of the calibration set, parameter indicators of the validation set and / or preset performance indicators.

[0095] As an optional implementation of the above-mentioned device, the machine learning model includes: a partial least squares regression PLSR model, a back propagation neural network BPNN model or a support vector regression SVR model.

[0096] It should be understood that the device corresponds to the above-mentioned drug dissolution prediction method embodiment and can perform the various steps involved in the above-mentioned method embodiment. The specific functions of the device can be found in the above description, and the detailed description is appropriately omitted here. The device includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the operating system (OS) of the device.

[0097] See also Figure 4 The electronic device 300 provided in the embodiment of the present application includes: a processor 310 and a memory 320, wherein the memory 320 stores machine-readable instructions executable by the processor 310, and when the machine-readable instructions are executed by the processor 310, the above method is executed.

[0098] The embodiment of the present application also provides a computer-readable storage medium 330, on which a computer program is stored, and the computer program is executed by the processor 310 to execute the above method. The computer-readable storage medium 330 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk.

[0099] The embodiment of the present application also provides a computer program product, including: a computer program or a computer instruction, and the computer program or the computer instruction executes the method described above when executed by a processor.

[0100] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0101] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed device and method can also be implemented in other ways. The device embodiments described above are only schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible implementation architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a part of a module, a program segment or a code, and a part of a module, a program segment or a code includes one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also be different from the order of occurrence marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which is mainly determined by the functions involved.

[0102] In addition, each functional module of each embodiment in the embodiment of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part. In addition, in the description of this specification, the description of reference terms "one embodiment", "some embodiments", "example", "specific example", "some examples", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present application. In this specification, the schematic representation of the above terms does not necessarily target the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples without contradiction.

[0103] The above description is only an optional implementation manner of the embodiments of the present application, but the protection scope of the embodiments of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the embodiments of the present application, which should be covered within the protection scope of the embodiments of the present application.

Claims

1. A method for predicting drug dissolution, characterized in that: include: Perform feature extraction on the near infrared spectrum data of the drug to be treated to obtain spectrum features; Predicting the dissolution rate of the active ingredient in the drug to be processed according to the spectral characteristics using a machine learning model, wherein the machine learning model is constructed by extracting the spectral characteristics of the active ingredient based on near-infrared spectral data of the sample drug and the dissolution rate of the active ingredient in the sample drug at different times; Among them, the feature extraction of the near-infrared spectral data of the drug to be treated includes: performing non-negative matrix decomposition on the near-infrared spectral data of the drug to be treated to obtain the decomposed spectral data; performing self-modeling hybrid analysis on the near-infrared spectral data of the drug to be treated to obtain the spectral features.

2. The method according to claim 1, characterized in that Before predicting the dissolution rate of the active ingredient in the drug to be processed according to the spectral features using the machine learning model, the method further includes: Obtaining near infrared spectral data of a sample drug and the dissolution rate of an active ingredient in the sample drug at different times; Performing standardization correction on the near-infrared spectral data to obtain corrected spectral data; Extracting the spectral characteristics of the active ingredient in the sample medicine from the corrected spectral data; The machine learning model is constructed using the spectral characteristics of the active ingredients in the sample medicine and the dissolution rates of the active ingredients in the sample medicine at different times.

3. The method according to claim 2, characterized in that The method of obtaining near infrared spectroscopy data of the sample medicine includes: Produce a plurality of sample drugs by drug manufacturing equipment with different production processes and formulations; A plurality of sample drugs are scanned at the same time by a near-infrared spectrum scanner to obtain near-infrared spectrum data of the sample drugs.

4. The method according to claim 2, characterized in that: The constructing of the machine learning model comprises: Dividing the spectral characteristics of the active ingredient in the sample medicine and the dissolution rate of the active ingredient in the sample medicine at different times into a verification set and a calibration set; Using multiple machine learning algorithms to construct multiple calibration models based on the calibration set; The validation set is used to evaluate the performance of the multiple correction models, and a correction model with the best performance screened out from the multiple correction models is determined as the machine learning model.

5. The method according to claim 4, characterized in that The step of determining a correction model with the best performance selected from the multiple correction models as the machine learning model comprises: The multiple calibration models are screened using the parameter index of the calibration set, the parameter index of the validation set and / or the preset performance index.

6. The method according to any one of claims 1 to 5, characterized in that: The machine learning model includes: a partial least squares regression PLSR model, a back propagation neural network BPNN model or a support vector regression SVR model.

7. A drug dissolution prediction device, characterized in that: include: A spectral feature acquisition module is used to extract features from near-infrared spectral data of the drug to be treated and obtain spectral features; The drug dissolution prediction module is used to predict the dissolution rate of the active ingredient in the drug to be processed based on the spectral characteristics using a machine learning model. The machine learning model is constructed by extracting the spectral characteristics of the active ingredient based on the near-infrared spectral data of the sample drug, as well as the dissolution rate of the active ingredient in the sample drug at different times.

8. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the machine-readable instructions are executed by the processor to perform any method according to claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is executed.

10. A computer program product, characterized in that include: A computer program or a computer instruction, wherein when the computer program or the computer instruction is executed by a processor, the method according to any one of claims 1 to 6 is executed.

Images