Prediction method of purification process of hydroxytyrosol, terminal device and computer readable storage medium

Abstract

The application relates to a prediction method of a purification process of hydroxytyrosol, a terminal device and a computer readable storage medium. The prediction method comprises the following steps: selecting column chromatography sample loading parameters and elution parameters of a hydroxytyrosol sample to be purified as characteristic parameters, forming a training set and a verification set, training an initial neural network model by using the training set to obtain a target neural network model, verifying the target neural network model by using the verification set, and determining the purification process of the hydroxytyrosol. The structure of a hidden layer in the target neural network model comprises at least two layers, and / or the number of neurons in the hidden layer comprises 6-12. The application realizes efficient and rapid prediction and establishment of the column chromatography purification process of the hydroxytyrosol by using the neural network model, greatly simplifies the test number and the purification process steps in the purification process, reduces the waste of consumables, has accurate prediction results, and has a high application prospect.

Description

Technical Field

[0001] This application relates to the field of component detection technology, and in particular to a method for predicting the purification process of hydroxytyrosol, a terminal device, and a computer-readable storage medium. Background Technology

[0002] Hydroxytyrosol is a naturally occurring phenolic compound, primarily found in olive oil, olive leaves, and some other plants. It possesses various biological activities, including antioxidant, anti-inflammatory, and antibacterial properties, making it a potential candidate for applications in cosmetics and food. In practical applications, the purity of hydroxytyrosol significantly impacts the activity of the product, and the purification process requires verification of the influence of multiple parameters in the purification process. Currently, response surface methodology (RSM) is commonly used to minimize the number of experiments during purification. However, RSM is more suitable for low-dimensional, relatively simple data processing. The purification process of hydroxytyrosol involves numerous variables, and various related factors exhibit complex nonlinear relationships. Therefore, RSM cannot efficiently and accurately determine the purification process parameters, resulting in low purification efficiency, a large optimization workload, and wasted resources. This is a key problem that urgently needs to be solved in the current hydroxytyrosol purification process. Summary of the Invention

[0003] Therefore, it is necessary to provide a method that can efficiently and accurately determine the purification process of hydroxytyrosol.

[0004] In a first aspect, this application provides a method for predicting the purification process of hydroxytyrosol, wherein the purification process includes purifying the hydroxytyrosol sample by column chromatography, and the parameters of the column chromatography include loading parameters and elution parameters.

[0005] The parameters of the column chromatography are predicted using a target neural network model, including the following steps:

[0006] The parameters of column chromatography were selected as feature parameters to form training and validation sets;

[0007] The initial neural network model is trained using the training set to obtain the target neural network model, and the target neural network model is validated using the validation set to determine the parameters of the column chromatography, thus forming a purification process for hydroxytyrosol.

[0008] The target neural network model comprises a hidden layer, and the structure of the hidden layer includes at least one of features i and ii:

[0009] i. The number of layers in the hidden layer includes at least two layers;

[0010] ii. The number of neurons in the hidden layer includes 6-12.

[0011] In some embodiments, the column chromatography uses macroporous adsorption resin packing; and / or, the elution parameters include an ethanol solution or a methanol solution, wherein the mass fraction of the ethanol solution and the methanol solution is independently 70%-80%; and / or, the loading parameters include one or more of the loading liquid volume and the loading liquid flow rate; and / or, the elution parameters include one or more of the elution liquid volume and the elution liquid flow rate; and / or, the training set has 15-20 samples; and / or, the validation set has 5-10 samples; and / or, the target neural network model further includes an input layer and an output layer, wherein the input layer, the hidden layer, and the output layer are sequentially connected.

[0012] In some embodiments, the input layer and the output layer each independently comprise one layer; and / or, the input layer comprises 2-4 neurons; and / or, the output layers in the target neural network model each independently comprise 1-2 neurons; and / or, the hidden layer comprises a first hidden layer and a second hidden layer connected sequentially to the input layer, the second hidden layer being connected to the output layer, the first hidden layer comprising 4-8 neurons, and the second hidden layer comprising 2-4 neurons.

[0013] In some embodiments, during training and validation, the evaluation metrics of the target neural network model independently include the coefficient of determination R. 2 One or more of the following: root mean square error (RMSE) and mean absolute percentage error (MAPE).

[0014] In some embodiments, the evaluation index is obtained by correlating the predicted value and the true value of the target neural network model. The predicted value includes the purity prediction value, and the true value includes the purity true value. The purity true value is the purity of the hydroxytyrosol purified sample after the hydroxytyrosol sample to be purified is purified by the column chromatography.

[0015] In some embodiments, the true purity value is obtained by one or more of chemical quantitative analysis methods and model prediction methods;

[0016] The model prediction method uses an infrared spectroscopy quantitative analysis model for prediction, and the prediction includes the following steps:

[0017] The purified hydroxytyrosol sample was subjected to infrared spectroscopy detection, infrared spectral data was collected, the infrared spectral data was preprocessed, and the preprocessed infrared spectral data was correlated with the prediction results to construct the infrared spectral quantitative analysis model; wherein, the correlation method includes one or more of partial least squares method and cross-validation method.

[0018] Infrared spectral data of the purified hydroxytyrosol sample to be tested are collected and imported into the infrared spectral quantitative analysis model for purity prediction. The output result is used as the true purity value.

[0019] In some embodiments, the evaluation metrics used in the preprocessing process include the coefficient of determination R. 2 and one or more of the following: root mean square error corrected by cross-validation (RMSECV);

[0020] The evaluation index is obtained by correlating the purity result output by the infrared spectroscopy quantitative analysis model with the baseline purity, wherein the baseline purity is obtained by detecting the purity of the hydroxytyrosol purified sample by liquid chromatography.

[0021] In some embodiments, the resolution of infrared spectroscopy detection is 2 cm⁻¹. -1 -4 cm -1 ; and / or, the number of scans for infrared spectroscopy detection is 8-16; and / or, the wavenumber range for infrared spectroscopy detection is 2000 cm⁻¹. -1 -500 cm -1 ; and / or, the preprocessing methods include one or more of vector normalization, multivariate scattering correction, first derivative and second derivative.

[0022] Secondly, this application also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the prediction method for the purification process of hydroxytyrosol provided in the first aspect.

[0023] Thirdly, this application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the steps of the predictive method for the purification process of hydroxytyrosol provided in the first aspect.

[0024] Compared with traditional technologies, the beneficial effects of the technical solution in this application are as follows:

[0025] This application uses a target neural network model to predict the purification process of hydroxytyrosol samples. The purification process includes column chromatography, specifically predicting one or more parameters of the column chromatography process, such as the loading flow rate, loading volume, elution flow rate, and elution volume. By training the initial neural network model, the target neural network model is optimized to quickly establish the parameters in the hydroxytyrosol purification process, greatly simplifying the purification process. Compared with the traditional response surface methodology, this method significantly reduces the number of experimental groups and experiments, significantly improves purification efficiency, and also ensures accurate prediction results. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the neural network model in the prediction method for the purification process of hydroxytyrosol in this application.

[0027] Figure 2 Infrared spectra of five batches of purified hydroxytyrosol samples.

[0028] Figure 3 This is the standard curve for hydroxytyrosol.

[0029] Figure 4 This is a flowchart of the cross-validation process for the infrared spectroscopy quantitative analysis model.

[0030] Figure 5 The hydroxytyrosol sample was purified at a wavenumber of 1500 cm⁻¹. -1 -1200 cm -1 Infrared spectral quantitative analysis model prediction spectrum with first derivative preprocessing.

[0031] Figure 6 The hydroxytyrosol sample was purified at a wavenumber of 1500 cm⁻¹. -1 -1200 cm -1 Infrared spectral quantitative analysis model predicts spectra after second-order derivative preprocessing.

[0032] Figure 7 The sample for hydroxytyrosol purification was carried out at 1500 cm⁻¹ -1 -1200 cm -1 The prediction results of the infrared spectroscopy quantitative analysis model after preprocessing with first and second derivatives within the wavenumber range.

[0033] Figure 8 The results show the accuracy of the infrared spectroscopy quantitative analysis model in predicting purified hydroxytyrosol samples. Detailed Implementation

[0034] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0036] As used herein, "optional," "optional," and "optional" refer to either "with" or "without" parallel options. If multiple "optional" entries appear in a technical solution, each "optional" entry is independent unless otherwise specified and there are no contradictions or mutual constraints. The term "and / or" as used herein includes any and all combinations of one or more related listed items. Unless otherwise specified, "multiple," "multiple," etc., as used herein refer to a quantity greater than 2 or equal to 2; for example, "one or more" indicates one, two, or more than two. In open-ended technical features or solutions described herein using words such as "containing," "including," and "comprising," unless otherwise specified, additional members beyond the listed members are not excluded. This can be considered as providing both a closed-ended feature or solution consisting of the listed members and an open-ended feature or solution that includes additional members beyond the listed members.

[0037] In this application, the terms "first aspect," "second aspect," "third aspect," "fourth aspect," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or quantity, nor should they be construed as implicitly indicating the importance or quantity of the indicated technical features. Moreover, "first," "second," "third," "fourth," etc., serve only as a non-exhaustive enumeration and should be understood not to constitute a closed limitation on quantity.

[0038] Unless otherwise specified, all embodiments and optional embodiments of this application can be combined to form new technical solutions.

[0039] Column chromatography is a commonly used purification process. It separates components in a sample based on their different partition coefficients in the stationary and mobile phases through repeated partitioning. Column chromatography process parameters include loading parameters, elution parameters, and column packing materials. These parameters are interrelated, and traditional column chromatography purification processes are typically obtained through orthogonal experimental design. Orthogonal experimental design is another design method for studying multiple factors and levels. It selects representative points from a comprehensive experiment based on orthogonality. These representative points possess the characteristics of "uniform dispersion and comparable data," making it a relatively efficient, rapid, and economical experimental design method. However, given the large number of process parameters in column chromatography and the complex nonlinear relationships between these factors, traditional orthogonal experimental design requires a large number of experiments, resulting in high time and economic costs and low efficiency.

[0040] To address the aforementioned technical problems, this application constructs a neural network model that efficiently predicts the column chromatography process parameters for hydroxytyrosol, significantly reducing the time and cost of determining purification process parameters.

[0041] A neural network is a computer system that simulates the neural organization of the human brain, reflecting the information processing mechanism of the biological brain. It consists of a large number of interconnected neurons and possesses the properties of a biological nervous system. As a non-limiting example, the neural network model constructed in this application is a back-propagation (BP) neural network model. BP neural networks are the most widely used and best represent the fundamental advantages of neural networks. A BP neural network is a multi-layered pre-neural network that can backpropagate errors. The core of a BP neural network's operation is to calculate results through forward propagation and adjust parameters through backward propagation. It does not require manually setting complex mathematical formulas and can learn data patterns from large amounts of data.

[0042] Partial least squares regression (PLSR) is a multivariate statistical data analysis method that mainly studies regression modeling of multiple dependent variables on multiple independent variables.

[0043] The purpose of cross-validation is to obtain a reliable and stable model. In a given modeling sample, most of the samples are used to build the model, while a small portion of the samples are used to make predictions using the newly built model. The prediction errors of this small portion of samples are calculated and their squares are recorded and summed.

[0044] In a first aspect, this application provides a method for predicting the purification process of hydroxytyrosol, wherein the purification process includes purifying the hydroxytyrosol sample by column chromatography, and the parameters of the column chromatography include loading parameters and elution parameters.

[0045] The parameters of the column chromatography are predicted using a target neural network model, including the following steps:

[0046] The parameters of column chromatography were selected as feature parameters to form training and validation sets;

[0047] The initial neural network model is trained using the training set to obtain the target neural network model, and the target neural network model is validated using the validation set to determine the parameters of the column chromatography, thus forming a purification process for hydroxytyrosol.

[0048] The target neural network model comprises a hidden layer, and the structure of the hidden layer includes at least one of features i and ii:

[0049] i. The number of layers in the hidden layer includes at least two layers;

[0050] ii. The number of neurons in the hidden layer is 6-12.

[0051] This application constructs an initial neural network model and trains and optimizes it. The hidden layer structure of the target neural network model includes two hidden layers and / or hidden layer feature parameters with 6-12 neurons. The optimized target neural network model can efficiently predict the purification process of hydroxytyrosol.

[0052] In some embodiments, the loading parameters include one or more of the loading liquid volume and the loading liquid flow rate, and the elution parameters include one or more of the elution liquid volume and the elution liquid flow rate. Further, the prediction method provided in this application can simultaneously predict the loading liquid volume, loading liquid flow rate, elution liquid volume, and elution liquid flow rate, realizing a complex nonlinear relationship between four independent variables and the target output value. As a non-limiting example, the loading flow rate parameter range is 2 mL / min-6 mL / min, the loading liquid volume parameter range is 200 mL-400 mL, the elution flow rate is 1 mL / min-5 mL / min, and the elution liquid volume is 40 mL-60 mL. It can be understood that the parameter range includes the sample parameter range of the training set and validation set or the parameter range of the predicted purification process.

[0053] It is understood that the sample liquid in the volume and flow rate of the sample liquid is a liquid sample or a mixture of sample and solvent.

[0054] In some embodiments, the training set has 15-20 samples, including but not limited to 15, 16, 17, 18, 19, 20, or any combination thereof and values ​​within that range. Compared to traditional response surface methodology, the prediction method provided in this application requires only a smaller number of experimental groups to rapidly determine purification process parameters. For example, to examine the independent and interactive effects of loading flow rate, loading volume, elution flow rate, and elution volume, response surface methodology requires 3... 4 =81 sets of experiments, which require a lot of time to conduct the tests, as well as a lot of reagents and consumables.

[0055] In some embodiments, the number of samples in the validation set is 5-10, including but not limited to 5, 6, 7, 8, 9, 10 or any combination thereof and values ​​within that range.

[0056] In some embodiments, the initial neural network model and the target neural network model each independently include an input layer, a hidden layer, and an output layer, which are sequentially connected. It should be understood that the input layer is used to input the input parameters of the dataset, and the number of input layers and neurons can be adjusted according to parameters such as the number of input samples. In this application, the input layer is used to input parameters for column chromatography, including one or more of loading flow rate, loading liquid volume, elution flow rate, and elution liquid volume. The output layer is used to output the output parameters from the hidden layer, and the number of output layers and neurons can be adjusted according to the output detection index parameters. In this application, the output layer is used to output purification prediction values.

[0057] In some embodiments, the input layer of the initial neural network model and the target neural network model each independently includes one layer, wherein the input layer of the initial neural network model and the target neural network model each independently includes 4 neurons.

[0058] In some embodiments, the hidden layers in the initial neural network model and the target neural network model each independently include at least two layers. It should be understood that all hidden layers are set as fully connected layers, meaning that neurons in each layer are connected to all neurons in the next layer, thereby achieving complete data transmission and full information sharing. The number of neurons in the hidden layers of the initial neural network model and the target neural network model are each independently 6-12. Within this range of the number of neurons in the hidden layers, the neural network model exhibits strong generalization ability; conversely, too few or too many neurons will lead to performance degradation. The former is due to insufficient model capacity, and the latter may be due to overfitting. Further, the hidden layers in the initial neural network model and the target neural network model each independently include a first hidden layer and a second hidden layer. The hidden layers include a first hidden layer and a second hidden layer sequentially connected to the input layer, and the second hidden layer is connected to the output layer. The first hidden layer has 4-8 neurons, and the second hidden layer has 2-4 neurons.

[0059] In some embodiments, the output layer of the initial neural network model and the target neural network model each independently includes one layer, wherein the output layer of the initial neural network model and the target neural network model each independently includes one neuron.

[0060] In some embodiments, the input layer, the hidden layer, and the output layer are all interconnected through neurons. This means that the input layer, the hidden layer, and the output layer are all interconnected through the nodes of neurons.

[0061] In some embodiments, the activation functions of the hidden layers in the initial neural network model and the target neural network model include one or more of the following: sigmoid function, softmax function, relu function, tanh function, and leaky relu function.

[0062] This application sets, adjusts, and optimizes the parameters of the input, hidden, and output layers of a neural network model, such as the number of layers, neurons, and activation functions. It provides sufficient complexity in capturing nonlinear relationships in the data while avoiding excessive computational burden and overfitting risks caused by unreasonable parameter settings. This application achieves a good balance between prediction accuracy and model efficiency in setting the relevant parameters of the neural network model.

[0063] In some embodiments, during training, an algorithm is used to optimize the hyperparameters in the initial neural network model, the algorithm including a forward feedback algorithm; the hyperparameters include one or more of the number of hidden layers and the number of neurons in the hidden layers.

[0064] In some embodiments, a termination condition is set for the training process, wherein the termination condition is 2≤RMSE≤3.

[0065] In some embodiments, the maximum number of iterations during training is set to 1000.

[0066] In some embodiments, the column chromatography packing material comprises a macroporous adsorption resin packing material, which includes one or more of styrene-based and divinylbenzene types. As a non-limiting example, the styrene-based macroporous adsorption resin packing material includes one or more of DA-201 and ADS-750.

[0067] In some embodiments, the elution solvent for the column chromatography comprises an ethanol solution or a methanol solution. As a non-limiting example, the mass fraction of the ethanol solution and the methanol solution are each independently 70%-80%.

[0068] In some embodiments, during training and validation, the evaluation metrics independently include the coefficient of determination R. 2 One or more of the following: root mean square error (RMSE) and mean absolute percentage error (MAPE). The coefficient of determination R... 2 The performance of a neural network model is evaluated using metrics such as Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). 2 The closer RMSE and MAPE are to 1, the better the fitting effect of the neural network model. When RMSE and MAPE are closer to 0, the smaller the error of the neural network model.

[0069] In some embodiments, the evaluation index is obtained by correlating the predicted value and the true value of the target neural network model. The predicted value includes the purity prediction value, and the true value includes the purity true value. The purity true value is the purity of the hydroxytyrosol purified sample after the hydroxytyrosol sample to be purified is purified by the column chromatography.

[0070] As a non-limiting example, the true purity value is determined by a chemical quantitative analysis method, wherein the chemical quantitative analysis method includes thin-layer chromatography, liquid chromatography, gas chromatography, mass spectrometry, or ultraviolet spectrophotometry.

[0071] As a non-limiting example, the true purity value is predicted using an infrared spectroscopy quantitative analysis model, wherein the prediction includes the following steps:

[0072] The purified hydroxytyrosol sample was subjected to infrared spectroscopy detection, infrared spectral data was collected, the infrared spectral data was preprocessed, and the preprocessed infrared spectral data was correlated with the prediction results to construct the infrared spectral quantitative analysis model; wherein, the correlation method includes one or more of partial least squares method and cross-validation method.

[0073] Infrared spectral data of the purified hydroxytyrosol sample to be tested are collected and imported into the infrared spectral quantitative analysis model for purity prediction. The output result is used as the true purity value.

[0074] See appendix Figure 4 The cross-validation process in the infrared spectroscopy quantitative analysis model of this application includes:

[0075] Infrared spectral data of purified hydroxytyrosol samples were used as the calibration set spectral data, and the purity value of purified hydroxytyrosol samples determined by liquid chromatography was used as the baseline purity. Partial Least Squares (PLSR) modeling was selected, and cross-validation options were set (defining three types of parameters: validation method (K value / type), principal component number determination rules, and data processing rules). Modeling and validation were then initiated. One sample was taken from the calibration set as the "sample to be predicted" for the current round, and the remaining samples were used as the "training sample set." Then, using the spectral data and baseline purity of the "training sample set" as input values, a quantitative model was built using the PLSR method (this model is a temporary model for the current round). The established temporary PLSR model was then used to predict the purity of the "temporarily transferred validation sample" to obtain the predicted value of the sample. The "predicted purity" and "baseline purity" ("liquid chromatography reference value" in the figure) of the validation sample were paired and recorded for subsequent calculation of prediction error. The above steps were repeated until every sample in the calibration set had been tested as a "temporarily transferred validation sample" (i.e., all loops were completed). At this point, paired data of predicted and true values ​​for all calibration samples were obtained.

[0076] Based on the "purity prediction value - baseline purity" of all cycle records, the model's coefficient of determination R (reflecting the correlation between the predicted value and the true value) and the root mean square error of cross-validation RMSECV (reflecting the model's prediction accuracy) are calculated to determine whether the model's performance meets the standards.

[0077] In some embodiments, the evaluation metrics used in the preprocessing process include the coefficient of determination R. 2 The evaluation index is one or more of the following: root mean square error of cross-validation correction (RMSECV); the evaluation index is obtained by correlating the purity result output by the infrared spectroscopy quantitative analysis model with the baseline purity, wherein the baseline purity is obtained by detecting the purity of the hydroxytyrosol purified sample by liquid chromatography.

[0078] In some embodiments, the resolution of infrared spectroscopy detection is 2 cm⁻¹. -1 -4 cm -1 Furthermore, the resolution of infrared spectroscopy detection is 4 cm⁻¹. -1 .

[0079] In some embodiments, the number of scans for infrared spectroscopy detection is 8-16. Further, the number of scans for infrared spectroscopy detection is 8.

[0080] In some embodiments, the wavenumber range for acquiring the infrared spectral data of the purified hydroxytyrosol sample is 2000 cm⁻¹. -1 -500 cm -1 including but not limited to 2000 cm -1 1800 cm -1 1500 cm -1 1200 cm -1 1000 cm -1 800cm -1 500 cm -1 Or the range formed by either of the foregoing and the values ​​within that range. For example, the wavenumber range may be selected from 1800 cm⁻¹. -1 -1500 cm -1 1500 cm -1 -1200 cm -1 Or 1200 cm -1 -1000 cm -1 .

[0081] In some embodiments, the preprocessing method includes one or more of vector normalization, multivariate scattering correction, first derivative, and second derivative.

[0082] This application trains an initial neural network model using a collected training set, adjusts and optimizes the relevant architectural parameters of the model, and validates it using a validation set to form a target neural network model. This model can quickly and accurately predict the purity of hydroxytyrosol in samples to be purified, thereby enabling the rapid determination and establishment of the purification process for hydroxytyrosol. Furthermore, by constructing an infrared spectroscopy quantitative analysis model to predict the purity of purified hydroxytyrosol samples, the time required by traditional quantitative analysis methods is significantly reduced, thereby further accelerating the determination and establishment of the purification process for hydroxytyrosol.

[0083] The purification process parameters for hydroxytyrosol predicted in this application include: a loading volume of 350 mL, a loading flow rate of 4.8 mL / min, an elution volume of 45 mL-47 mL, and an elution flow rate of 4.0 mL / min-4.1 mL / min.

[0084] This application also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the predictive method for the purification process of hydroxytyrosol provided in the first aspect.

[0085] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the steps of the predictive method for the purification process of hydroxytyrosol provided in the first aspect.

[0086] For experimental parameters not specified in the following specific embodiments, please refer to the guidelines given in this application document first, or refer to experimental manuals or other experimental methods known in the art, or refer to the experimental conditions recommended by the manufacturer.

[0087] The raw materials and reagents involved in the following specific embodiments can be obtained commercially or prepared by those skilled in the art using known methods.

[0088] Example 1: Extraction of Hydroxytyrosol

[0089] Extraction steps of hydroxytyrosol from olive leaves: Hydroxytyrosol was extracted from olive leaves in a 2 mol / L hydrochloric acid solution. An orthogonal experimental design was used, as shown in Table 1.

[0090] Table 1: Orthogonal experimental design of extraction process

[0091]

[0092] The orthogonal experimental results showed that the optimal extraction process was: an extraction temperature of 44.58℃, a material-to-liquid ratio of 1:39.7, and an extraction time of 2.62 h, with a yield of 17.92 mg / g. Based on the actual production process, the optimal extraction process for hydroxytyrosol in this application was to mix olive leaves with 2 mol / L hydrochloric acid solution at a mass ratio of 1:40 and extract at 45℃ for 2.5 h. Under these conditions, the highest yield of hydroxytyrosol was 17.54 mg / g.

[0093] In this embodiment, hydroxytyrosol is obtained by filtering a solution of olive leaves after hydrolysis with hydrochloric acid, and then storing the filtrate at 4°C for later use.

[0094] This application uses the crude hydroxytyrosol extract obtained above as the initial raw material for the purification process, and uses a model prediction method to predict the purification process and parameters of the initial raw material, thereby efficiently constructing a purification process for hydroxytyrosol.

[0095] Example 2: Predictive method for the purification process of hydroxytyrosol

[0096] The specific steps of the method for predicting the purification process of hydroxytyrosol in this embodiment are as follows:

[0097] Sample to be purified: Crude extract of hydroxytyrosol extracted in Example 1.

[0098] Purification method: column chromatography, DA-201 macroporous adsorption resin packing, elution solvent is 75% ethanol solution.

[0099] Initial Neural Network Model Establishment: An initial neural network model was established using MATLAB 2014b. The model consists of an input layer, hidden layers, and an output layer connected sequentially. Each layer is fully connected, with the input, hidden, and output layers linked by neuron nodes, ensuring stable signal transmission. The input layer is used to input the column chromatography process parameters, including sample loading rate, sample volume, eluent flow rate, and eluent volume. The output layer outputs the purity.

[0100] Establishment of training and validation sets: The sample data for the training and validation sets were obtained using the loading flow rate, loading volume, eluent flow rate, and eluent volume from the column chromatography process parameters. Based on the experimental design in Table 2, 20 sets of column chromatography parameters were randomly selected as the training set and 5 sets as the validation set. See Table 3 for details.

[0101] Table 2: Parameter Design of Column Chromatography in Purification Process

[0102]

[0103] Table 3: Training and validation sets of parameters for column chromatography in the purification process

[0104]

[0105] Construction of the target neural network model (see Figure 1 The initial neural network model is trained using a selected training set. The activation function in the hidden layer is the sigmoid function. A forward feedback algorithm is used, and the termination condition is root mean square error 2 ≤ RMSE ≤ 3. The maximum number of iterations during training is set to 1000. The target neural network model is obtained and validated using a validation set. The structure of the target neural network model includes: an input layer containing 1 layer and 4 neurons, a hidden layer containing 2 layers and 9 neurons (the first hidden layer contains 6 neurons, and the second hidden layer contains 3 neurons), and an output layer containing 1 layer and 1 neuron. The coefficient of determination R in the target neural network model is... 2 The value is 0.964, and the root mean square error (RMSE) is 2.547.

[0106] The purification process of hydroxytyrosol was predicted in the target neural network model. The results showed that the optimal purification process parameters were 350 mL of sample loading liquid, 4.8 mL / min of flow rate, 47 mL of elution liquid and 4.1 mL / min of flow rate, which yielded a sample purity of 90.7%.

[0107] Further adjustments were made to the purification process parameters to facilitate actual experimental operation. Specifically, the sample loading volume was 350 mL, the sample loading flow rate was 5 mL / min, the elution buffer volume was 45 mL, and the elution flow rate was 4 mL / min. Three repeated experiments were conducted under these purification conditions, ultimately yielding a sample purity of 89.9%. This confirmed the purification process for the hydroxytyrosol sample.

[0108] When the target neural network model has one hidden layer and six neurons, the coefficient of determination R in the constructed target neural network model is... 2 The value is 0.912, and the root mean square error (RMSE) is 2.983.

[0109] When the target neural network model has three hidden layers, where the first hidden layer contains 3 neurons, the second hidden layer contains 3 neurons, and the third hidden layer contains 3 neurons, the determination coefficient R in the constructed target neural network model is... 2 The value is 0.944, and the root mean square error (RMSE) is 2.741.

[0110] It is evident that among the target neural network models constructed in this application, the model with two hidden layers exhibits the best fit. The architecture achieves a good balance between model complexity and prediction performance, effectively capturing data features while possessing excellent generalization ability. Furthermore, within the structure with two hidden layers, the structure with six neurons in the first hidden layer and three neurons in the second hidden layer is even more superior.

[0111] In the construction of the aforementioned target neural network model, the coefficient of determination R is used. 2 The root mean square error (RMSE) was used as the evaluation index. During the evaluation process, the predicted purity value of the model (the output value of the output layer) and the true purity value (the purity of the purified hydroxytyrosol sample obtained by purifying the crude hydroxytyrosol extract using the corresponding purification process parameters) were calculated.

[0112] This application constructs an infrared spectroscopy quantitative analysis model that enables rapid and accurate quantification of the true purity value. The construction and quantification method of the infrared spectroscopy quantitative analysis model are as follows:

[0113] Determination of wavenumber range: Five batches of purified hydroxytyrosol samples obtained from five purification processes were randomly selected from Table 3. Infrared spectroscopy was performed on purified hydroxytyrosol samples of the same mass. The average spectral peaks of the infrared spectra were found to be relatively consistent, with a range of approximately 2000 cm⁻¹. -1 -500 cm -1 There is a distinct absorption peak at this point (see [reference]). Figure 2 ).

[0114] Determination of reference purity: The purity of hydroxytyrosol in the above 5 batches of purified hydroxytyrosol samples was determined by liquid chromatography to form a reference purity. The liquid chromatography conditions were as follows: Waters Atlantis T3 column (4.6 x 150 mm, 3 µm), mobile phase A was 0.1 (v / v)% phosphoric acid aqueous solution, mobile phase B was acetonitrile, gradient elution (elution program is shown in Table 4), detection wavelength was 210 nm, injection volume was 10 μL, flow rate was 1 mL / min, and column temperature was 30 °C.

[0115] Table 4: Gradient elution procedure

[0116]

[0117] The content of hydroxytyrosol was quantitatively analyzed using the standard curve method. Hydroxytyrosol standard solutions of various concentrations were prepared by dissolving hydroxytyrosol standard in acetonitrile. These solutions were then detected using the aforementioned liquid chromatography conditions. The peak areas at different concentrations were measured. Linear regression fitting was performed with the concentration of the standard solution as the x-axis (X) and the peak area as the y-axis (Y). (See Appendix) Figure 3 ).

[0118] The five batches of purified hydroxytyrosol samples were diluted with acetonitrile to prepare sample solutions, which were then analyzed under the same liquid chromatography conditions. The peak area was used as the quantification, and the baseline purity of the five batches of purified hydroxytyrosol samples was obtained by means of a standard curve.

[0119] Construction of the infrared spectroscopy quantitative analysis model: Infrared spectral data of the above five batches of purified hydroxytyrosol samples were collected. Infrared spectral scanning was used, and the resolution of infrared spectral detection was 4 cm⁻¹. -1 The infrared spectroscopy detection involved 8 scans. The infrared spectral data and reference purity were imported into The Unscrambler 9.8 software. A calibration model was constructed using partial least squares method combined with cross-validation. Cross-validation was performed, and the calibration model was externally validated (see appendix). Figure 4 The predicted values ​​of hydroxytyrosol purified samples under different wavelength ranges and pretreatment methods were compared and optimized, with the internal cross-determination coefficient R0 being used as the criterion. 2 The root mean square error of internal cross-validation (RMSECV) is a comprehensive indicator, with the coefficient of determination R... 2 Within the range of 0-1, a value closer to 1 indicates a better model fit to the data. The root mean square error of internal cross-validation (RMSECV) represents the model's prediction error; a smaller value indicates a stronger model fit to the data. This is used to determine the optimal quantitative infrared spectroscopy model.

[0120] For infrared spectroscopy quantitative analysis models established using the same batch of purified hydroxytyrosol samples, different parameter settings significantly affect the model's fitting ability and performance, leading to differences in accuracy and applicability. Data preprocessing methods are used to eliminate spectral differences caused by variations in measurement conditions, thereby ensuring the correlation between infrared spectral data and purity values, resulting in more similar preprocessed spectra. Specific preprocessing schemes are detailed in Tables 5 and 6.

[0121] Table 5: Influence of different wavenumber ranges on the quantitative analysis model of infrared spectroscopy

[0122]

[0123] Table 6: Influence of different preprocessing methods on the quantitative analysis model of infrared spectroscopy

[0124]

[0125] As shown in Tables 5 and 6, the preferred wavenumber range is 1500 cm⁻¹. -1 -1200 cm -1The preferred preprocessing method is to use the first and second derivatives, since the first and second derivatives R... 2 The difference from the RMSECV value was not significant, so the purity was verified through prediction. Aqueous solutions of hydroxytyrosol purified samples with mass concentrations of 3.5%, 4%, 4.5%, 5%, 5.5%, and 6% were prepared, and equal amounts of each sample were subjected to infrared spectroscopy testing. The optimized wavenumber range of 1500 cm⁻¹ was selected. -1 -1200 cm -1 The preprocessing methods were compared using the first and second derivatives, and the processed infrared spectral data were input into the prediction model for analysis (see [link]). Figure 5 and Figure 6 See also Figure 7 As shown in the figure, the predicted values ​​of the model after preprocessing and optimization using the first derivative are all close to the standard curve (actual addition amount), while only 3 of the predicted values ​​of the model after preprocessing and optimization using the second derivative are close to the standard curve (actual addition amount), and the prediction effect is not as good as the optimized model after preprocessing with the first derivative.

[0126] Validation of the infrared spectroscopy quantitative analysis model: The purity of other batches of purified hydroxytyrosol samples was predicted using the optimized infrared spectroscopy quantitative analysis model described above (see [link to model]). Figure 8 The results fully validated the rationality and reliability of the model, and also demonstrated the accuracy of the constructed infrared spectroscopy quantitative analysis model in predicting the purity of hydroxytyrosol purified samples.

[0127] The infrared spectroscopy quantitative analysis model constructed above is used to predict the true purity value of hydroxytyrosol samples during the purification process prediction using the target neural network model. Compared with traditional chemical analysis methods, such as liquid chromatography, it is more efficient and simpler, further accelerating the determination and prediction of the purification process of hydroxytyrosol.

[0128] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting the purification process of hydroxytyrosol, characterized in that, The purification process includes purifying the hydroxytyrosol sample using column chromatography. The parameters of the column chromatography include loading parameters and elution parameters. The packing material used in the column chromatography includes macroporous adsorption resin packing material. The loading parameters include loading liquid volume and loading liquid flow rate. The elution parameters include elution liquid volume and elution liquid flow rate. The parameters of the column chromatography are predicted using a BP target neural network model, including the following steps: The parameters of column chromatography are selected as feature parameters to form training and validation sets; The initial BP neural network model is trained using the training set to obtain the BP target neural network model. The BP target neural network model is then validated using the validation set to determine the parameters of the column chromatography, thus forming a purification process for hydroxytyrosol. The structure of the BP target neural network model includes hidden layers, and the structure of the hidden layers includes i and ii features: i. The number of layers in the hidden layer includes at least two layers; ii. The number of neurons in the hidden layer is 6-12.

2. The method for predicting the purification process of hydroxytyrosol according to claim 1, characterized in that, The elution parameters include an ethanol solution or a methanol solution, wherein the mass fraction of the ethanol solution and the methanol solution is independently 70%-80%; and / or, the training set has 15-20 samples; and / or, the validation set has 5-10 samples.

3. The method for predicting the purification process of hydroxytyrosol according to claim 1, characterized in that, The structure of the BP target neural network model also includes an input layer and an output layer, wherein the input layer, the hidden layer, and the output layer are connected in sequence.

4. The method for predicting the purification process of hydroxytyrosol according to claim 3, characterized in that, The input layer and the output layer each consist of an independent layer; and / or, the number of neurons in the input layer is 2-4; and / or, the number of neurons in the output layer is 1-2; and / or, the hidden layer includes a first hidden layer and a second hidden layer connected in sequence to the input layer, the second hidden layer being connected to the output layer, the first hidden layer having 4-8 neurons, and the second hidden layer having 2-4 neurons.

5. The method for predicting the purification process of hydroxytyrosol according to any one of claims 1 to 4, characterized in that, In the process of training and verifying, the evaluation indexes of the BP target neural network model each independently include one or more of a coefficient of determination R 2 , a root mean square error RMSE, and a mean absolute percentage error MAPE.

6. The method for predicting the purification process of hydroxytyrosol according to claim 5, characterized in that, The evaluation index is obtained by correlating the predicted value and the actual value of the BP target neural network model. The predicted value includes the purity prediction value, and the actual value includes the purity actual value. The purity actual value is the purity of the hydroxytyrosol purified sample after the hydroxytyrosol sample to be purified is purified by the column chromatography.

7. The method for predicting the purification process of hydroxytyrosol according to claim 6, characterized in that, The true purity value is obtained through one or more of the following methods: quantitative chemical analysis and model prediction. The model prediction method uses an infrared spectroscopy quantitative analysis model for prediction, and the prediction includes the following steps: The purified hydroxytyrosol sample was subjected to infrared spectroscopy detection, infrared spectral data was collected, the infrared spectral data was preprocessed, and the preprocessed infrared spectral data was correlated with the prediction results to construct the infrared spectral quantitative analysis model; wherein, the correlation method includes one or more of partial least squares method and cross-validation method. Infrared spectral data of the purified hydroxytyrosol sample to be tested are collected and imported into the infrared spectral quantitative analysis model for purity prediction. The output result is used as the true purity value.

8. The method for predicting the purification process of hydroxytyrosol according to claim 7, characterized in that, The evaluation indicators used in the preprocessing process include the coefficient of determination R. 2 and one or more of the following: root mean square error corrected by cross-validation (RMSECV); The evaluation index is obtained by correlating the purity result output by the infrared spectroscopy quantitative analysis model with the baseline purity, wherein the baseline purity is obtained by detecting the purity of the hydroxytyrosol purified sample by liquid chromatography.

9. The method for predicting the purification process of hydroxytyrosol according to claim 7 or 8, characterized in that, The resolution of infrared spectroscopy detection is 2 cm. -1 -4 cm -1 ; and / or, the number of scans for infrared spectroscopy detection is 8-16; and / or, the wavenumber range for infrared spectroscopy detection is 2000 cm⁻¹. -1 -500 cm -1 ; and / or, the preprocessing methods include one or more of vector normalization, multivariate scattering correction, first derivative and second derivative.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor runs the computer program to implement the steps of the prediction method for the purification process of hydroxytyrosol according to any one of claims 1 to 9.

11. A computer-readable storage medium storing a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the predictive method for the purification process of hydroxytyrosol according to any one of claims 1 to 9.

Images