A modeling method for predicting multi-index component content of day lily

Abstract

The present application relates to the technical field of pharmaceutical analysis, and more particularly to a modeling method for predicting the content of multiple index components of day lily, comprising the following steps: S1, collecting day lily samples from multiple producing areas, preparing the samples into powders and collecting near-infrared spectra; S2, pre-processing the data collected by near-infrared spectroscopy; S3, determining the content of components including quercetin, kaempferol, isorhamnetin and beta-sitosterol in the day lily samples by high performance liquid chromatography as reference values; S4, dividing the near-infrared spectral data into a training set and a test set. The present application is a rapid prediction model for the content of multiple index components of day lily based on near-infrared spectroscopy technology and gradient boosting decision tree algorithm, which is simple in sample pre-treatment, fast in analysis, non-destructive and environmentally friendly, and has high model prediction accuracy, providing an efficient and reliable new technical means for the quality control of day lily medicinal materials.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical analysis technology, and in particular to a modeling method for predicting the content of multiple components in daylily. Background Technology

[0002] Quality control of Chinese medicinal materials is a core element in ensuring their safety and efficacy. Currently, the identification of the origin and quality evaluation of Chinese medicinal materials mainly rely on botanists or experienced identification experts to observe and identify the appearance, color, texture, and microstructure of the materials. This method has significant limitations: on the one hand, it is highly dependent on the personal experience and subjective judgment of experts, making standardization and popularization difficult; on the other hand, the identification process is time-consuming and laborious, failing to meet the market's demand for rapid, batch identification of medicinal materials. The Pharmacopoeia of the People's Republic of China typically controls the quality of medicinal materials by controlling the content of active ingredients and extracts. However, daylily is not included in the Pharmacopoeia of the People's Republic of China (2025 edition). The previous chapter discussed the shape and microscopic identification, thin-layer chromatography identification, and determination of the content of active ingredients in daylily. However, conventional testing methods cannot provide information on the origin of medicinal materials, nor can they fully control the internal quality of medicinal materials. Therefore, it is crucial to establish a method that can quickly and effectively identify the origin of medicinal materials and reflect the overall information of the medicinal materials, whether for identifying the origin of medicinal materials or for evaluating the quality of medicinal materials. In recent years, fingerprinting technology has been widely used in the identification of the origin of medicinal materials and quality control. However, its sample pretreatment procedures are relatively cumbersome. Near-infrared spectroscopy detection technology does not require destroying the sample, is fast and simple, and has broad application prospects.

[0003] In summary, current quality control methods for daylily medicinal materials lack a rapid, non-destructive, and efficient approach that can simultaneously identify the origin and perform quantitative analysis of multiple components. Existing technologies either rely too heavily on subjective experience and cumbersome chemical analysis, or face challenges in establishing specialized rapid quantitative models. Summary of the Invention

[0004] The purpose of this invention is to provide a modeling method for predicting the content of multiple index components in daylilies, in order to solve the problem mentioned in the background art of lacking a fast, non-destructive and efficient method that can take into account both origin identification and quantitative analysis of multiple index components.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a modeling method for predicting the content of multiple index components in daylilies, comprising the following steps:

[0006] S1. Collect daylily samples from multiple production areas, prepare them into powder, and perform near-infrared spectroscopy.

[0007] S2. Preprocess the data acquired from the near-infrared spectrum;

[0008] S3. The contents of components including quercetin, kaempferol, isorhamnetin and β-sitosterol in daylily samples were determined by high performance liquid chromatography and used as reference values.

[0009] S4. Divide the near-infrared spectral data into a training set and a test set;

[0010] S5. Using the near-infrared spectral data of the training set and its corresponding component content reference values, a multi-index component content prediction model is established by combining the gradient boosting decision tree (GBDT) algorithm.

[0011] S6. Validate the prediction model using a test set and evaluate the model's prediction accuracy.

[0012] Preferably, the near-infrared spectral acquisition in step S1 uses an integrating sphere diffuse reflectance method, with a scanning range of 12000-4000 cm⁻¹ under normal temperature conditions. -1 8cm resolution -1 64 scans.

[0013] Preferably, the preparation of the powder in step S1 includes drying and crushing the collected daylily samples, making them into powder, and then passing them through a 120-mesh sieve.

[0014] Preferably, the preprocessing in step S2 includes importing the near-infrared spectral data into the sklearn module of the machine learning algorithm to be used, and performing preprocessing of the spectral data using linear regression algorithm, ensemble algorithm, error prediction algorithm and multiple linear regression algorithm.

[0015] Preferably, the training set and test set mentioned in step S4 specifically include cross-validation, using the train_test_split() function to randomly divide the data to be processed.

[0016] Preferably, in step S5, when establishing the prediction model, near-infrared spectral data are used as feature variables, and the contents of quercetin, kaempferol, isorhamnetin and β-sitosterol are used as target variables. Gradient boosting decision tree (GBDT) is used for regression modeling.

[0017] Preferably, in step S6, the prediction accuracy of the model is evaluated by calculating the correlation coefficient between the predicted value and the measured value.

[0018] A system for predicting the content of multiple components in daylilies includes:

[0019] The near-infrared spectroscopy acquisition module is used to acquire near-infrared spectral data of daylily samples;

[0020] The data processing module is used to preprocess the spectral data;

[0021] The modeling module is used to build a predictive model based on the Gradient Boosting Decision Tree (GBDT) algorithm, combined with spectral data and component content reference values;

[0022] The prediction module is used to predict the content of multiple components in an unknown sample using the model.

[0023] Preferably, the modeling module supports the simultaneous prediction of at least two of the components among quercetin, kaempferol, isorhamnetin, and β-sitosterol.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] 1. By combining near-infrared spectroscopy with machine learning algorithms, rapid and non-destructive detection of daylily samples has been achieved, reducing the analysis time from several hours in traditional chromatography to just a few minutes, greatly improving efficiency and reducing reliance on senior experts. By constructing a robust quantitative model, a fundamental shift in quality evaluation from "experience-based judgment" to "data-driven" has been realized, providing a replicable technical path for quality research of similar medicinal materials.

[0026] 2. The near-infrared spectroscopy technology employed offers significant advantages. Samples only need to be dried, pulverized, and sieved (e.g., 120 mesh) before direct detection, completely eliminating complex extraction, separation, and other chemical pretreatment steps. This method not only greatly simplifies the operation process, lowers the technical threshold and labor costs, but also conforms to the concept of green environmental protection because it does not require the use of chemical reagents. Therefore, this invention effectively overcomes the bottlenecks of existing rapid detection technologies, providing a practical solution for large-scale, high-throughput quality screening and rapid on-site identification of daylilies in market circulation.

[0027] 3. The established gradient boosting decision tree (GBDT) model can simultaneously and accurately predict the content of multiple key active ingredients in daylily, such as quercetin, kaempferol, isorhamnetin and β-sitosterol. The predicted values ​​are highly correlated with the measured values, thus providing a more comprehensive quality evaluation perspective than single-indicator detection. Attached Figure Description

[0028] Figure 1 The images show near-infrared spectra of daylily samples at different particle sizes, as presented in this invention.

[0029] Figure 2 This is a superimposed image of 50 scans of daylily samples from this invention;

[0030] Figure 3 This is a superimposed near-infrared spectrum of 51 batches of daylily samples from this invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1-3 This invention provides a technical solution: a modeling method for predicting the content of multiple components in daylilies, comprising the following steps:

[0033] S1. Collect daylily samples from multiple production areas, prepare them into powder, and perform near-infrared spectroscopy.

[0034] S2. Preprocess the data acquired from the near-infrared spectrum;

[0035] S3. The contents of components including quercetin, kaempferol, isorhamnetin and β-sitosterol in daylily samples were determined by high performance liquid chromatography and used as reference values.

[0036] S4. Divide the near-infrared spectral data into a training set and a test set;

[0037] S5. Using the near-infrared spectral data of the training set and its corresponding component content reference values, a multi-index component content prediction model is established by combining the gradient boosting decision tree (GBDT) algorithm.

[0038] S6. Validate the prediction model using the test set and evaluate the model's prediction accuracy.

[0039] Furthermore, in step S1, the near-infrared spectral acquisition employs an integrating sphere diffuse reflectance method, with a scanning range of 12000-4000 cm⁻¹ under normal temperature conditions. -1 8cm resolution -1 64 scans.

[0040] Preferably, the powder preparation in step S1 includes drying and crushing the collected daylily samples, making them into powder, and then passing them through a 120-mesh sieve.

[0041] Furthermore, the preprocessing in step S2 includes importing the near-infrared spectral data into the sklearn module of the machine learning algorithm to be used, and performing linear regression, ensemble, error prediction and multiple linear regression algorithms to preprocess the spectral data.

[0042] Furthermore, in step S4, the training set and test set specifically include cross-validation, using the train_test_split() function to randomly divide the data to be processed.

[0043] Furthermore, in step S5, when establishing the prediction model, near-infrared spectral data are used as feature variables, and the contents of quercetin, kaempferol, isorhamnetin and β-sitosterol are used as target variables. Gradient boosting decision tree (GBDT) is used for regression modeling.

[0044] Furthermore, in step S6, the prediction accuracy of the model is evaluated by calculating the correlation coefficient between the predicted value and the measured value.

[0045] A system for predicting the content of multiple components in daylilies includes:

[0046] The near-infrared spectroscopy acquisition module is used to acquire near-infrared spectral data of daylily samples;

[0047] The data processing module is used to preprocess the spectral data;

[0048] The modeling module is used to build a predictive model based on the Gradient Boosting Decision Tree (GBDT) algorithm, combined with spectral data and component content reference values;

[0049] The prediction module is used to predict the content of multiple components in unknown samples using a model.

[0050] Furthermore, the modeling module supports the simultaneous prediction of at least two of the components in quercetin, kaempferol, isorhamnetin, and β-sitosterol.

[0051] Example:

[0052] Materials and Instruments Preparation

[0053] Sample collection:

[0054] A total of 51 batches of representative daylily medicinal material samples were collected from Gansu, Sichuan, Hunan, Shaanxi and Shanxi provinces, and the sample information was fully recorded.

[0055] Main instruments and equipment:

[0056] Accessories include: ANTARIS II Fourier Transform Near Infrared Spectrometer (Thermo Fisher Scientific, USA), High Performance Liquid Chromatograph (brand and model), grinder, standard sieve sets (60 mesh, 70 mesh, 80 mesh, 100 mesh, 120 mesh, 140 mesh), quartz sample cups, sample press, etc.

[0057] Sample pretreatment and spectral acquisition

[0058] Sample preparation:

[0059] Daylily medicinal materials from various producing areas were dried, pulverized, and then made into powder. The near-infrared spectra of the powders sieved at different mesh sizes (60, 70, 80, 100, 120, 140 mesh) were compared (e.g., Figure 1 As shown in the figure, it was found that the absorbance increased with the increase of mesh size. Finally, a 120-mesh sieve was selected for uniform sieving to ensure the consistency of sample particle size.

[0060] Instrument parameter settings:

[0061] Integrating sphere diffuse reflectance measurement was used, with a resolution of 8 cm. -1 64 scans were performed at room temperature, with a scanning range of 12000-4000 cm. -1 .

[0062] Instrument performance verification:

[0063] The sample from origin No. 31 was measured 50 times consecutively, and the spectral overlay diagram (as shown) was obtained. Figure 2 As shown in the image, the instrument exhibits good precision, stability, and repeatability, and the RSD value meets the analytical requirements.

[0064] Spectral acquisition process:

[0065] Preheat the instrument for 30 minutes;

[0066] Open the OPUS 7.5 software, check the signal and save the peak values;

[0067] Scan the background single-channel spectrum (scan the background once for each sample);

[0068] Take an appropriate amount of 120-mesh daylily powder and put it into a quartz sample cup. Gently shake it to distribute it evenly.

[0069] Compact the sample using a sample compactor to ensure consistent sample thickness;

[0070] Collect sample spectra, and repeat the sample loading and scanning 6 times for each batch of samples, and calculate the average spectrum;

[0071] Near-infrared spectral overlays of 51 batches of samples (e.g.) Figure 3 (As shown).

[0072] Reference value determination

[0073] The contents of four target components in all samples were determined by high performance liquid chromatography (HPLC): quercetin, kaempferol, isorhamnetin, and β-sitosterol. The content ranges of each component are shown in Table 1.

[0074]

[0075] Data processing and modeling

[0076] Near-infrared spectral data were imported into the sklearn module for machine learning algorithms, including linear regression, ensemble algorithms, error prediction algorithms, and multiple linear regression, to preprocess the spectral data. Cross-validation was employed, and the data was randomly divided into 34 training sets and 17 test sets using the train_test_split() function. Using the contents of quercetin, kaempferol, isorhamnetin, and β-sitosterol as indicators, the near-infrared spectra of the training set samples were combined with a quantitative model for these four components in daylilies established using GBDT to predict their contents. Other indicator data were used as feature data, and a target Y was calculated for the four columns of data to be processed. Decision tree regression analysis was then performed. The prediction results of the decision tree model for quercetin, kaempferol, isorhamnetin, and β-sitosterol are shown in Tables 2 to 5.

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085] Model Evaluation

[0086] A correlation analysis was performed on the predicted and measured values, and the results are shown in Table 6.

[0087]

[0088] The code for establishing a quantitative data model for sex is as follows:

[0089] Appendix 6:

[0090] Daylily quantitative modeling code

[0091] # coding: utf-8

[0092] # In[1]:

[0093] import pandas as pd

[0094] # Read in data from three Excel sheets, where the sheet names are sequentially changed to 1, 2, and 3.

[0095] # In

[44] :

[0096] df = pd.read_excel(".. / .. / extralJob / boss_wang / content determination.xls",sheetname=["1","2","3"],header=None)

[0097] # Merge data from three sheets

[0098] # In

[45] :

[0099] df = pd.concat([df['1'],df['2'],df['3']],axis=1)

[0100] # Transpose data rows and columns

[0101] # In

[53] :

[0102] df = df.T

[0103] # # Give the features names like f0 f1 ...

[0104] # In

[65] :

[0105] df.columns = map(lambda x:"f"+x, map(str,range(df.shape[1]-1))) + ['y']

[0106] # In

[67] :

[0107] df.head(2)

[0108] # In

[351] :

[0109] raw_df = df.iloc[range(1,df.shape[0]+1,2),:]

[0110] # # Sample size distribution by origin

[0111] # In

[352] :

[0112] raw_df.y.value_counts()

[0113] Because the data from different production areas is severely unbalanced, we attempted data augmentation by copying the Shanxi data three times.

[0114] # In

[353] :

[0115] raw_df = pd.concat([raw_df[raw_df.y.isin([u"Shanxi"])]]*3 + [raw_df[raw_df.y.isin([u"Hunan",u"Sichuan"])]]*1 +

[0116] [raw_df[raw_df.y==u"Gansu"]]+[raw_df[raw_df.y==u"Shaanxi"]])

[0117] # In

[354] :

[0118] raw_df.y.value_counts()

[0119] # Import Model Tool

[0120] # In

[344] :

[0121] # In

[355] :

[0122] le = LabelEncoder()

[0123] raw_df["label"] = le.fit_transform(raw_df.y)

[0124] # In

[356] :

[0125] for i,n in enumerate(le.classes_):

[0126] print i,n

[0127] # Split the training set and the test set

[0128] # In

[357] :

[0129] train_x,test_x,train_y,test_y = train_test_split(raw_df.iloc[:,:-2],raw_df.iloc[:,-1],train_size=0.8,random_state=41)

[0130] # coding: utf-8

[0131] # In[1]:

[0132] import pandas as pd

[0133] # In[2]:

[0134] df = pd.read_excel(".. / .. / extralJob / boss_wang / Quality Evaluation Data.xls",sheetname='2')

[0135] # In[8]:

[0136] raw_df = df.iloc[:16,:]

[0137] # In

[46] :

[0138] from sklearn.ensemble import GradientBoostingRegressor

[0139] from sklearn.linear_model import LinearRegression,MultiTaskElasticNet

[0140] from sklearn.metrics import mean_squared_error

[0141] from sklearn.model_selection import train_test_split

[0142] # In

[50] :

[0143] # train_x,test_x,train_y,test_y = train_test_split(raw_df.iloc[:,:-4],raw_df.iloc[:,-4],train_size=0.8,random_state=2)

[0144] # # Use other indicator data as feature data and the 4 columns of data to be processed as target Y for decision tree regression

[0145] # Use the trained model to predict values for filling

[0146] # In

[102] :

[0147] The above code implements the quantitative data model establishment process of this patent. It uses the GBDT algorithm to fit the spectral data of 34 training set samples with the HPLC measured values ​​and predicts the content of four components in 17 test set samples. Finally, the correlation coefficients between the predicted and measured values ​​of each component meet the significance correlation criteria shown in Table 6 (e.g., the correlation coefficient of quercetin is 0.998), proving that the model has high prediction accuracy and solves the technical problems of low efficiency and cumbersome operation of traditional detection methods.

[0148] In summary, this invention collected near-infrared spectra of daylily medicinal materials from 51 production areas. The near-infrared spectral data was preprocessed, and algorithms such as logistic regression (LR), support vector machine with RBF kernel, gradient boosting decision tree, Gaussian Naive Bayes, and multilayer perceptron were used to establish a classification model for the production areas of daylily medicinal materials. The results show that preprocessing the spectral data using linear regression, cross-validation, ensemble, error prediction, and neural network algorithms significantly improves the accuracy of the production area identification model. The classification results of the decision tree GBDT model demonstrate its excellent recognition ability. The decision tree GBDT model was applied to predict the production areas of actual daylily samples from 20 different production areas, achieving an accuracy rate of over 90%. These research results indicate that the decision tree GBDT algorithm combined with near-infrared spectroscopy can be applied to the identification and analysis of the production areas of daylily. The quantitative model also collected near-infrared spectra of daylily medicinal materials from 51 producing areas. The near-infrared spectral data were preprocessed, and the processed data were randomly divided into 34 training sets and 17 test sets. Using HPLC to measure the contents of quercetin, kaempferol, isorhamnetin, and β-sitosterol as indicators, the near-infrared spectra of the training set samples, combined with the quantitative model for quercetin, kaempferol, isorhamnetin, and β-sitosterol in daylilies established using GBDT, were used to predict the contents of these four components. Correlation analysis was performed between the predicted and measured values, showing a significant correlation between the predicted and measured values. The development of a non-destructive, rapid, and environmentally friendly quality evaluation system using a combination of near-infrared technology and chemometrics is of significant importance.

[0149] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A modeling method for predicting the content of multiple components in daylilies, characterized in that, Includes the following steps: S1. Collect daylily samples from multiple production areas, prepare them into powder, and perform near-infrared spectroscopy. S2. Preprocess the data acquired from the near-infrared spectrum; S3. The contents of components including quercetin, kaempferol, isorhamnetin and β-sitosterol in daylily samples were determined by high performance liquid chromatography and used as reference values. S4. Divide the near-infrared spectral data into a training set and a test set; S5. Using the near-infrared spectral data of the training set and its corresponding component content reference values, a multi-index component content prediction model is established by combining the gradient boosting decision tree (GBDT) algorithm. S6. Validate the prediction model using a test set and evaluate the model's prediction accuracy.

2. The modeling method for predicting the content of multiple components in daylilies according to claim 1, characterized in that, The near-infrared spectral acquisition in step S1 uses an integrating sphere diffuse reflectance method, with a scanning range of 12000-4000 cm⁻¹ under normal temperature conditions. -1 8cm resolution -1 64 scans.

3. The modeling method for predicting the content of multiple components in daylilies according to claim 1, characterized in that, The preparation of the powder in step S1 includes drying and crushing the collected daylily samples, making them into powder, and then passing them through a 120-mesh sieve.

4. The modeling method for predicting the content of multiple components in daylilies according to claim 1, characterized in that, The preprocessing in step S2 includes importing the near-infrared spectral data into the sklearn module of the machine learning algorithm to be used, and performing linear regression, ensemble, error prediction and multiple linear regression algorithms to preprocess the spectral data.

5. The modeling method for predicting the content of multiple components in daylilies according to claim 1, characterized in that, The training set and test set mentioned in step S4 specifically include cross-validation, using the train_test_split() function to randomly divide the data to be processed.

6. The modeling method for predicting the content of multiple components in daylilies according to claim 1, characterized in that, In step S5, when establishing the prediction model, near-infrared spectral data are used as feature variables, and the contents of quercetin, kaempferol, isorhamnetin and β-sitosterol are used as target variables. Gradient boosting decision tree (GBDT) is used for regression modeling.

7. The modeling method for predicting the content of multiple components in daylilies according to claim 1, characterized in that, In step S6, the accuracy of the model's prediction is evaluated by calculating the correlation coefficient between the predicted value and the measured value.

8. A system for predicting the content of multiple components in daylily, as described in any one of claims 1-7, and a modeling method for predicting the content of multiple components in daylily, characterized in that, include: The near-infrared spectroscopy acquisition module is used to acquire near-infrared spectral data of daylily samples; The data processing module is used to preprocess the spectral data; The modeling module is used to build a predictive model based on the Gradient Boosting Decision Tree (GBDT) algorithm, combined with spectral data and component content reference values; The prediction module is used to predict the content of multiple components in an unknown sample using the model.

9. A system for predicting the content of multiple components in daylilies according to claim 8, characterized in that, The modeling module supports the simultaneous prediction of at least two of the following components: quercetin, kaempferol, isorhamnetin, and β-sitosterol.

Images