Method and model for predicting bile acid component content in bear gall powder, storage medium and modeling method

Abstract

The invention belongs to the field of model prediction, and particularly relates to a method for predicting the content of bile acid components in bear gall powder, a model, a storage medium and a modeling method. On the basis that TUDCA, TCDCA, UDCA, CDCA and TCA are screened out to distinguish significant difference components of bear gall powder and other animal-derived bile powder, near-infrared quantitative models of five components are respectively established by combining PLS and BP-ANN algorithms, and are compared to find that the prediction effect is best when the quantitative models of the TUDCA, UDCA and CDCA are established by applying the PLS method, the effect is best when the quantitative models of the TCDCA and TCA are established by applying the BP-ANN method, and the prediction effect is good. The relative error between the predicted value and the true value of external verification is less than 5%, the model prediction performance is good, and a novel simple and low-cost method can be provided for quality evaluation research of bear gall powder.

Description

Technical Field

[0001] The invention belongs to the field of model prediction, and specifically relates to a method for predicting the content of bile acid components in bear bile powder, a model and a storage medium, and a modeling method. Background Art

[0002] Bear bile powder Ursi Fellis Pulvis is a dried product obtained by bile drainage from the black bear Selenarctos thibetanus Cuvier or the brown bear Ursus arctos Linnaeus of the Ursidae family through gallbladder surgery. It has the effects of clearing heat, calming the liver, and improving eyesight. Modern pharmacological studies have shown that bear bile has pharmacological effects such as protecting the liver and promoting bile secretion, anti-tumor and anti-atherosclerosis. Currently, bear bile powder is contained in national medicine standard number drugs such as bear bile powder eye drops, bear bile Kaiming tablets, and compound bear bile mint lozenges. Because bear bile is expensive, there is often adulteration in the market, such as processing bile from other animals such as pigs, cattle, and sheep into powder, or mixing bile powder from pigs, cattle, and sheep with bear bile powder for sale, which seriously affects the effectiveness and safety of clinical drugs and restricts the healthy development and scale expansion of the bear bile powder industry chain. Therefore, it is necessary to establish a rapid and effective detection method to control the quality of bear bile powder, standardize the production and operation of bear bile powder, and promote the healthy development of the industry.

[0003] The main source of spectral information of Near Infrared Spectrum (NIRS) is the harmonic and sum frequency absorption of hydrogen-containing groups. The collected near-infrared spectra are used to reflect the comprehensive information of chemical components, and the data information is deeply mined and utilized with the help of chemometrics. It is a key technology for non-destructive quality evaluation of traditional Chinese medicine. Near-infrared spectroscopy technology not only has the ability of qualitative identification, but also can establish mathematical models by properly calibrating samples. The optimized mathematical model and the near-infrared spectra of unknown samples can be used to quickly determine the content of multi-component unknown samples. It has the characteristics of simple sample preparation, high sensitivity, rapid and non-destructive, and simultaneous determination of multiple components. It has been widely used in the authenticity identification and quality control of traditional Chinese medicine.

[0004] Li Wenlong et al. used near-infrared spectroscopy combined with partial least squares method to establish a quantitative model for ursodeoxycholic acid and chenodeoxycholic acid in bear bile powder in the article "Rapid determination of ursodeoxycholic acid and chenodeoxycholic acid in bear bile powder extract by near-infrared diffuse reflectance spectroscopy". However, when bile powders from different animal sources were detected by UPLC-MS / MS, it was found that the differential components of bear bile powder and other animal bile powders were not only ursodeoxycholic acid and chenodeoxycholic acid, but also tauroursodeoxycholic acid, taurochenodeoxycholic acid and taurocholic acid. Therefore, evaluating bear bile powder with the content values ​​of these differential components can more accurately judge the quality of bear bile powder. Summary of the invention

[0005] To achieve the above object, the present invention provides a method for constructing a prediction model for the content of bile acids in bear bile powder, comprising the following steps:

[0006] 1) Extraction of bear bile powder characteristics: Take bear bile powder, collect near-infrared spectrum data and pre-process, and then use HPLC-ELSD method to determine the content of bile acid components in bear bile powder;

[0007] 2) Model construction: SPA algorithm is used to extract the spectral characteristic wavelengths in the spectral data preprocessed in step 1), and the extracted spectral characteristic wavelengths are used as input variables, and the content of bile acid components is used as the target variable, and are imported into the PLS model for training, so as to obtain a prediction model for the content of bile acid components in bear bile powder;

[0008] Or: using Matlab software to extract the first 11 principal components in the preprocessed spectral data in step 1), using the extracted 11 principal components as input parameters and the content of bile acid components as output parameters, and importing them into the BP-ANN model for training, thus obtaining a prediction model for the content of bile acid components in bear bile powder;

[0009] The bile acid components are: tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and / or taurocholic acid.

[0010] Furthermore, the pretreatment method in step 1) includes MSC, SNV, 1 st D. 2 ed Any one or more of D, SG, and normalization.

[0011] Furthermore, the pre-processing method is 1 st D, SG and ncl, HPLC-ELSD method was used to determine the content of tauroursodeoxycholic acid in bear bile powder; SNV and Ncl, HPLC-ELSD method was used to determine the content of ursodeoxycholic acid in bear bile powder; 1 st D and SG, HPLC-ELSD method was used to determine the content of chenodeoxycholic acid in bear bile powder; for normalization, HPLC-ELSD method was used to determine the content of taurochenodeoxycholic acid or taurocholic acid in bear bile powder.

[0012] Furthermore, the chromatographic conditions of the HPLC-ELSD method are: chromatographic column: Waters HSS T3 column, 4.6mm×250mm, 5μm; mobile phase: acetonitrile-0.1% formic acid aqueous solution; gradient elution: 0-3min, 40%-60% acetonitrile; 3-5min, 60%-66% acetonitrile; 5-10min, 66%-70% acetonitrile; 10-13min, 70%-90% acetonitrile; 13-15min, 90%-60% acetonitrile; 15-18min, 60%-40% acetonitrile; column temperature 40℃; flow rate 1.0mL / min; ELSD drift tube temperature 90℃; nitrogen flow rate 2.0mL / min; gain 8; splitless mode; injection volume 10μL;

[0013] And / or: the reference solution of the HPLC-ELSD method is prepared by dissolving tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and / or taurocholic acid reference substances in methanol; the test solution is prepared by ultrasonic extraction of bear bile powder with methanol.

[0014] Furthermore, in step 2), the PLS model predicts the content of taurochenodeoxycholic acid, ursodeoxycholic acid and / or chenodeoxycholic acid in bear bile powder after training; and the BP-ANN model predicts the content of taurochenodeoxycholic acid and / or taurocholic acid after training.

[0015] Furthermore, in step 2), the training function in the BP-ANN model is the Levenberg-Marquardt method, the transfer function is the Sigmoid function, and the output function is the Purelin formula;

[0016] The BP-ANN model uses RMSE as an indicator and determines the number of hidden layer nodes through training, where the calculation formula for the number of hidden layer nodes is:

[0017] L=(m+n) 1 / 2 +a, where L is the number of hidden layer nodes, n is the number of input layer nodes, m is the number of output layer nodes, and a is a natural number between 0 and 10.

[0018] The present invention also provides a prediction model for the content of bile acid components in bear bile powder constructed by the aforementioned method.

[0019] Finally, the present invention provides a method for detecting the content of bile acid components in bear bile powder, comprising the following steps:

[0020] The bear bile powder to be tested is collected for infrared spectrum and preprocessed, and then the SPA algorithm is used to extract the spectral characteristic wavelengths in the preprocessed spectral data, and the spectral characteristic wavelengths are input into the aforementioned prediction model to obtain the content of bile acid components in the bear bile powder;

[0021] The bile acid components are: tauroursodeoxycholic acid, ursodeoxycholic acid and / or chenodeoxycholic acid.

[0022] or:

[0023] Take the bear bile powder to be tested to collect infrared spectrum and preprocess it, then use Matlab software to extract the first 11 principal components in the preprocessed spectrum data in step 1), input the extracted 11 principal components into the aforementioned prediction model, and obtain the content of bile acid components in the bear bile powder;

[0024] The bile acid components are: taurochenodeoxycholic acid and / or taurocholic acid.

[0025] The present invention also provides a system for implementing the above-mentioned method for detecting the content of bile acid components in bear bile powder, comprising:

[0026] An input module, used for inputting the characteristics of Leonurus japonicus to be tested;

[0027] A computing module, including the aforementioned prediction model;

[0028] The output module is used to output the final prediction results of the content of bile acid components in the bear bile powder.

[0029] Finally, the present invention provides a computer-readable storage medium on which a computer program is stored, and the computer program is used to implement the aforementioned method for detecting the content of bile acid components in bear bile powder.

[0030] The invention screens out five bile acid components, namely, tauroursodeoxycholic acid (TUDCA), taurochenodeoxycholic acid (TCDCA), ursodeoxycholic acid (UDCA), chenodeoxycholic acid (CDCA) and taurocholic acid (TCA), as significantly different components for distinguishing bear bile powder from other animal-derived bile powders, combines PLS and BP-ANN algorithms to respectively establish near-infrared quantitative models of the five components, and compares the results to find that the quantitative models of TUDCA, UDCA and CDCA components established by the PLS method have the best prediction effect, and the quantitative models of TCDCA and TCA components established by the BP-ANN method have the best effect, the relative errors of the externally verified prediction value and the true value are both less than 5%, the model has good prediction performance, and can provide a new, simple and low-cost method for the quality evaluation research of bear bile powder.

[0031] Obviously, according to the above contents of the present invention, in accordance with common technical knowledge and customary means in the art, without departing from the above basic technical ideas of the present invention, other various forms of modification, replacement or change may be made.

[0032] The above contents of the present invention are further described in detail below through specific implementation methods in the form of embodiments. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following examples. All technologies realized based on the above contents of the present invention belong to the scope of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Near infrared overlay of bear bile powder sample

[0034] Figure 2 Chromatograms of mixed reference substance (A) and bear bile powder sample (B); Note 1: TCA; 2: TUDCA; 3: TCDCA; 4: UDCA; 5: CDCA.

[0035] Figure 3 Spectra after optimal preprocessing; A: TUDCA; B: TCDCA; C: UDCA; D: CDCA; E: TCA;

[0036] Figure 4 SPA screening modeling spectral variables; A-RMSE value; B-number of spectral variables;

[0037] Figure 5 The relationship between RMSECV and the number of main factors in PLS regression of five index components in bear bile powder; A: TUDCA; B: TCDCA; C: UDCA; D: CDCA; E: TCA;

[0038] Figure 6 Results of PLS ​​quantitative correction model; A: TUDCA; B: TCDCA; C: UDCA; D: CDCA; E: TCA;

[0039] Figure 7 External validation of the model; A: TUDCA; B: TCDCA; C: UDCA; D: CDCA; E: TCA;

[0040] Figure 8 Training process curve of BP-ANN; A: TUDCA; B: TCDCA; C: UDCA; D: CDCA; E: TCA;

[0041] Fig. 9 Quantitative results of the index components by BP-ANN model; A: TUDCA; B: TCDCA; C: UDCA; D: CDCA; E: TCA;

[0042] Fig.10 Fitting results of predicted values ​​and true values; A: TUDCA; B: TCDCA; C: UDCA; D: CDCA; E: TCA. DETAILED DESCRIPTION

[0043] Example 1 Nondestructive detection of differential components of bear bile powder by near infrared spectroscopy

[0044] In the early stage, the UPLC-MS / MS method was used to determine the content of bile acid components in bile powder from different animal sources, and a method that can simultaneously determine the content of 30 bile acid components in bile powder was established. The specific components of bear bile powder are UDCA and 3-O-DCA, which are only detected in bear bile powder. Trace amounts of ILCA, DHCA, and HDCA components are only detected in bile powder from other animals. Therefore, when judging whether the bear bile powder is adulterated, it is possible to detect whether the sample contains ILCA, DHCA, and HDCA components. The contents of TUDCA and CDCA components in bile powder from other animal sources are respectively below 0.1200% and 0.0495%, which are about 1 / 282 times and 1 / 21 times of the content of bear bile powder. The contents of TCDCA and TCA are different from those of bile powder from other animal sources. Combined with partial least squares discriminant analysis PLS-DA and variable weight VIP, TUDCA, TCDCA, UDCA, CDCA, and TCA are selected as differential components to distinguish bear bile powder from bile powder from other animal sources. In order to complete the detection of the differential components of bear bile powder without loss, a prediction model for the differential components of bear bile powder TUDCA, TCDCA, UDCA, CDCA, and TCA was further established.

[0045] 1. Experimental herbs

[0046] There were 93 batches of bear bile powder samples in total, and all of them were identified by Professor Huang Zehao of the Traditional Chinese Medicine Identification Teaching and Research Section of Fujian University of Traditional Chinese Medicine as dried products of Selenarctos thibetanus Cuvier bile powder.

[0047] 2. Methods and Results

[0048] 2.1 NIR data acquisition

[0049] Collect NIR by integrating sphere diffuse reflectance. Take appropriate amount of bear bile powder, 5 types of counterfeit powder and various adulterants through 80 mesh sieve, pour into finger vial and compact. Use built-in background as reference, scan range 4000~10000cm -1 , scanned 32 times, resolution 8cm -1 , temperature (25±0.5)℃, each batch of samples was collected 3 times, and the average spectrum was taken. The NIR overlay of 93 batches of bear bile powder samples is shown in Figure 1 .

[0050] 2.2 Simultaneous determination of five components in bear bile powder by HPLC-ELSD

[0051] 2.2.1 Chromatographic conditions and system suitability Waters HSS T3 column (4.6mm×250mm, 5μm), mobile phase acetonitrile (A)-0.1% formic acid aqueous solution (B), gradient elution: 0-3min, 40%-60% A; 3-5min, 60%-66% A; 5-10min, 66%-70% A; 10-13min, 70%-90% A; 13-15min, 90%-60% A; 15-18min, 60%-40% A. Column temperature 40℃; flow rate 1.0mL / min; ELSD drift tube temperature 90℃; nitrogen flow rate 2.0mL / min; gain 8; splitless mode; injection volume 10μL. Under the above chromatographic conditions, the resolution of the chromatographic peaks of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and taurocholic acid was >1.5. The chromatograms of the mixed control solution and samples are shown in Figure 2 .

[0052] 2.2.2 Preparation of mixed reference solution Take appropriate amount of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and taurocholic acid reference substances, weigh accurately, and add methanol to prepare reference substance stock solutions of 5.06, 4.04, 0.34, 0.44, and 0.45 mg / mL, respectively. Accurately take appropriate amount of each reference substance stock solution, add methanol to prepare mixed reference substance solutions containing the above five components at concentrations of 0.506, 0.404, 0.034, 0.044, and 0.045 mg / mL, respectively, and shake well to obtain.

[0053] 2.2.3 Preparation of test solution Take about 25 mg of bear bile powder, weigh it accurately, put it in a 50 mL conical flask with a stopper, accurately add 25 mL of methanol, weigh it, treat it with ultrasound (power 300 W, frequency 50 kHz) for 20 min, cool it, make up the weight with methanol, shake it well, filter it through a 0.45 μm microporous filter membrane, and take the filtrate to obtain the solution.

[0054] 2.2.4 Methodological investigation

[0055] 2.2.4.1 Investigation of linear relationship 1, 2, 4, 6, 8, 10 mL of the above mixed reference solution was accurately pipetted respectively, and each was placed in a 10 mL volumetric flask, and methanol was added to dilute to the scale, and shaken to prepare 5 component series linear reference solutions. Accurately pipet 10 μL of each series of reference solutions, inject, measure and record the chromatogram of each reference solution, inject into the liquid chromatograph, and use the logarithm of the peak area integral value (mAU) as the ordinate (Y) and the logarithm of the reference concentration (μg / mL) as the abscissa (X) for linear regression analysis to obtain the linear regression equations of the 5 bile acid components. The results are shown in Table 1.

[0056] Table 1 Linear regression equations of the components

[0057]

[0058] 2.2.4.2 Precision Take the bear bile powder sample (No. 1) and prepare the test solution according to the method under "2.2.3". Determine it according to the chromatographic conditions under "2.2.1". Inject the sample six times continuously, record the peak areas of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and taurocholic acid chromatographic peaks respectively, and calculate the RSD of each peak area, which are 1.0%, 0.9%, 2.3%, 1.9% and 2.1% (n=6), respectively, indicating that the precision of the instrument is good.

[0059] 2.2.4.3 Stability Take the bear bile powder sample (No. 1) and prepare the test solution according to the method under "2.2.3". The test solution was determined according to "2.2.1" at 0, 2, 4, 8, 12 and 24 hours. The RSDs of the peak areas of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and taurocholic acid were 1.3%, 2.6%, 2.9%, 2.2% and 1.8% (n=6), respectively, indicating that the test solution was stable within 24 hours.

[0060] 2.2.4.4 Repeatability Take the bear bile powder sample (No. 1) and prepare 6 test solutions in parallel according to the method under "2.2.3", and determine them according to the method under "2.2.1". Results The RSDs of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and taurocholic acid in the samples were 0.9%, 0.7%, 1.4%, 1.1% and 1.7% respectively (n=6). This shows that this method has good repeatability.

[0061] 2.2.4.5 Recovery rate of sample addition Take 6 portions of bear bile powder sample (No. 1), each portion is about 12.5 mg, accurately weigh, accurately add 1 times the amount of reference solution according to the known content of the sample, prepare in parallel according to the method under "2.2.3", and inject and measure according to the chromatographic conditions under "2.2.1", and calculate the recovery rate and RSD of each component. The recovery rate of sample addition is between 97% and 101%, and the RSD / % is between 3.0 and 4.2%.

[0062] 2.2.5 Sample content determination Take the bear bile powder sample and prepare the test solution according to the method in "2.2.3". Determine it according to the chromatographic conditions in "2.2.1". Calculate the content of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and taurocholic acid in the sample according to the regression equation and express it as mean ± standard deviation. The content ranges of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and taurocholic acid are 26.75% to 41.75%, 16.01% to 39.32%, 0.34% to 1.94%, 0.23% to 4.22%, and 0.21% to 2.27%, respectively. The content of tauroursodeoxycholic acid in bear bile powder is >23.0%, which meets the requirements of the Chinese Pharmacopoeia. [3] , and its content is the highest compared with the other four components, followed by taurochenodeoxycholic acid, and chenodeoxycholic acid, ursodeoxycholic acid and taurocholic acid have the lowest contents.

[0063] 2.3 Optimization and establishment of PLS ​​quantitative model

[0064] 2.3.1 Sample set division In order to avoid subset selection bias, the Kennard-Stone method was used to divide the 93 batches of samples in a ratio of 3:1. 70 batches of samples were used as calibration sets to optimize and establish quantitative calibration models; the remaining 23 batches of samples were used as validation sets to verify the predictive ability of the established calibration model. The sample set division and content distribution results are shown in Table 2.

[0065] Table 2 Sample set division and content distribution results (%, n = 3)

[0066]

[0067] Note: -mean value; s-standard deviation

[0068] 2.3.2 Selection of spectral preprocessing methods This study investigated the unprocessed spectra (None), MSC, SNV, 1 st D. 2 ed D and SG and other common spectral preprocessing methods and combination methods. The results are shown in Tables 3 and 4. For TUDCA, TCDCA, UDCA, CDCA, and TCA, 1 st D+SG+ncl, 1 st D+SG+ncl, SNV+ncl, 1 st D+SG, SG+SNV, Figure 3 This is the spectrum after optimal preprocessing of the five components.

[0069] Table 3 Effects of different preprocessing methods on the model①

[0070]

[0071] Table 4 The impact of different preprocessing methods on the model②

[0072]

[0073]

[0074] 2.3.3 Selection of spectral bands for modeling Since the full-band spectrum contains not only the information of the components to be measured but also a large amount of redundant information, this study uses the SPA algorithm to extract the spectral characteristic wavelengths of the samples after the best spectral pretreatment.

[0075] Select m-min=2, m-max=33, Figure 4 (A) represents the curve of the number of feature variables and RMSE. The figure shows that RMSE fluctuates as the number of feature variables increases. Figure 4 (B) is the number of characteristic wavelength points selected by the SPA algorithm. From top to bottom, they are the characteristic bands corresponding to TUDCA, TCDCA, UDCA, CDCA and TCA. For example, the selection of the characteristic band of the TUDCA component in bear bile powder can be expressed as follows: when the number of characteristic variables is 21, the RMSE value is the smallest, which is 1.3362. The SPA algorithm selects 21 characteristic wavelength points of the TUDCA component of bear bile powder. Table 5 shows the characteristic wavelength points corresponding to the five index components. The extracted wavelengths are used as input variables, and the index component content is used as the target variable to import into the classification model. The accuracy of the characteristic wavelength model is significantly weaker than the full-band spectrum classification result. The reason may be that the full-band spectrum can obtain the spectral information of each wavelength of the sample within the measurement band. After extracting the characteristic information, although the model running speed is greatly improved, more information related to the quantitative components is eliminated at the same time. Only by extracting dozens of wave points, good quantitative results cannot be produced. In order to better select the bands, this study selected the bands containing as many characteristic wavelength points as possible as the optimal bands based on the characteristic wavelength points. Tables 6 and 7 show the five-component modeling parameters under different bands.

[0076] Table 5 Statistics of characteristic wavelength selection

[0077]

[0078]

[0079] Table 6 Parameters of five-component modeling in different bands

[0080]

[0081] Table 7 Parameters of five-component modeling in different bands②

[0082]

[0083] 2.3.4 Selection of the number of main factors in the model

[0084] The effect of the number of main factors on RMSECV was investigated by internal cross-validation. The results are as follows: Figure 5 As shown. The number of principal factors with smaller RMSECV value is selected as the optimal number of principal factors for modeling each component. It can be seen that the optimal number of principal factors for modeling TUDCA, TCDCA, UDCA, CDCA and TCA are 7, 6, 10, 6 and 10 respectively.

[0085] 2.3.5 Methodological Study of Quantitative Model

[0086] For precision test, collect near infrared spectrum according to the method in item "2.1" Figure 6 The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group. The results of the experiment were compared with those of the control group.

[0087] 2.3.6 Establishment and evaluation of PLS ​​quantitative model Through the above optimization, the PLS quantitative model of tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid, taurocholic acid and other components in bear bile powder was established. The optimal modeling parameters of each component are shown in Table 8, and the correlation of the calibration model is shown in Figure 6 From the correlation diagram, we can see that the samples of the calibration set and the validation set are evenly distributed on both sides of the regression line. The linear equations of the reference values ​​and predicted values ​​of the five components are Y = 0.7832X + 7.2138, r = 0.922 0;

[0088] Y=0.7846X+3.2943, r=0.963 9; Y=0.9338X+0.0249, r=0.979 8;

[0089] Y=0.9649X+0.0738, r=0.964 4; Y=1.0395X-0.0328, r=0.964 4; This indicates that there is a good correlation between the predicted value of the established model and the true value measured by the reference method.

[0090] Table 8 Optimal modeling parameters

[0091]

[0092] 2.3.7 External validation of the PLS quantitative calibration model

[0093] 10 batches of samples that did not participate in modeling were selected for external validation. The predicted recovery rate was the ratio of the model predicted value to the actual value of the external validation sample. The results are shown in Tables 9 and 10. The relative error between the predicted value and the true value of the TUDCA component quantitative model was 0.08% to 2.01%, and the average recovery rate was 99.95%; the relative error between the predicted value and the true value of the TCDCA component quantitative model was 0.90% to 3.01%, and the average recovery rate was 99.78%; the relative error between the predicted value and the true value of the UDCA component quantitative model was 0.00% to 11.54%, and the average recovery rate was 101.85%; the relative error between the predicted value and the true value of the CDCA component quantitative model was 0.68% to 5.57%, and the average recovery rate was 101.16%; the relative error between the predicted value and the true value of the TCA component quantitative model was 1.93% to 22.64%, and the average recovery rate was 105.16%. The predicted values ​​were compared with the measured values, and the correlation between the two was analyzed using SPSS23 statistical analysis software. Among the five differential components, the correlation between the two was greater than 0.991, and the correlation was good. There was no significant difference between the two (P>0.05). The distribution range of the model predicted values ​​was basically consistent with the true values, and the prediction results were relatively accurate ( Figure 7 ).

[0094] Table 9 Prediction results of external validation samples①

[0095]

[0096] Table 10 Prediction results of external validation samples②

[0097]

[0098]

[0099] 2.4BP-ANN Modeling

[0100] The improved BP algorithm in the neural network toolbox of Matlab software was used for modeling.

[0101] 2.4.1 Data preprocessing In order to eliminate the differences between the orders of magnitude of different data, the data needs to be normalized before modeling. The software "Matlab R2018a" was used to normalize the TUDCA, TCDCA, UDCA, CDCA and TCA data.

[0102] 2.4.2 Determination of the number of hidden layer nodes When the number of nodes in the hidden layer is large enough, the BP-ANN with one hidden layer can approximate a nonlinear continuous function, so an ANN with one hidden layer is selected. The number of hidden layer nodes has a great influence on the performance of the network. Too many hidden layer nodes can easily cause overfitting of the network and reduce the prediction ability. Too few hidden layer nodes will result in poor network fitting effect. Therefore, it is necessary to screen the number of hidden layer nodes according to the modeling object. The calculation formula for the number of hidden layer nodes is L = (m + n) 1 / 2 +a, where L is the number of hidden layer nodes, n is the number of input layer nodes, m is the number of output layer nodes, and a is a natural number between 0 and 10. This experiment uses the root mean square error (RMSE) between the true value and the predicted value of the training set as an indicator. By training neural networks with different hidden nodes, we find the minimum training error and prediction error, and then determine the number of hidden layer nodes. The optimal training number is shown in Figure 8 .

[0103] 2.4.3 Model establishment 70% of the samples were randomly selected from 93 samples as training sets, 15% of the samples were used as prediction sets, and 15% of the samples were used as validation sets. First, Matlab software was used to extract the principal components of the full-band spectral data of the bear bile powder sample, and the top 11 principal components of TUDCA, TCDCA, UDCA, CDCA and TCA with a cumulative contribution rate of more than 98% were found. The scores of the top 11 principal components of each indicator (x) were used as input parameters, and the index component content data (y) was used as the output parameter. The number of nodes in the hidden layer of the network was screened according to the method under "2.4.2", and the minimum RMSE of the training set and the prediction set was used as an indicator to determine the number of nodes in the hidden layer. A BP network with one hidden layer was established, and the entire network structure adopted the 11-6-1 type, where 11 is the number of input layer nodes, 6 is the number of hidden layer nodes, and 1 is the number of output layer nodes. By Figure 8 It can be seen that a better BP network training model can be obtained through a limited number of trainings. The best training times for TUDCA, TCDCA, UDCA, CDCA and TCA are 11, 8, 6, 10 and 14, and the performance of their BP-ANN models is optimal. The best quantitative models are as follows: Fig. 9 As shown, the linear equations of the reference and predicted values ​​of the five components are Y = 1.0535X-1.8039, r = 0.842 6; Y = 0.9745X + 0.7485, r = 0.984 9;

[0104] Y=0.9848X+0.0242, r=0.953 9; Y=0.7863X+0.3548, r=0.888 8;

[0105] Y=0.8635X+0.1489, r=0.994 4, the linearity of TCDCA and TCA is good.

[0106] 2.4.4 Simulation prediction of ANN model Using the trained network model, a program was written in Matlab software, and the principal components of the full-band spectral data of 10 batches of bear bile powder samples that were not involved in the modeling were input to verify the performance of the ANN model. The prediction results are shown in Fig.10 , the predicted recovery rate is the ratio of the model predicted value to the actual value of the external validation sample, and the results are shown in Tables 11 and 12. The relative error between the TUDCA component quantitative model predicted value and the true value is 0.25% to 8.25%, and the average recovery rate is 97.79%; the relative error between the TCDCA component quantitative model predicted value and the true value is 0.41% to 2.28%, and the average recovery rate is 99.31%; the relative error between the UDCA component quantitative model predicted value and the true value is 1.04% to 30.11%, and the average recovery rate is 100.19%; the relative error between the CDCA component quantitative model predicted value and the true value is 2.97% to 17.49%, and the average recovery rate is 103.23%; the relative error between the TCA component quantitative model predicted value and the true value is 0.85% to 4.26%, and the average recovery rate is 102.62%. The predicted values ​​were compared with the measured values, and the correlation between the two was analyzed using SPSS23 statistical analysis software. Among the five differential components, the correlation between the two was greater than 0.875, and the correlation was good. There was no significant difference between the two (P>0.05). The distribution range of the model predicted values ​​was basically consistent with the true values, and the prediction results were relatively accurate.

[0107] Table 11 Prediction results of external validation samples①

[0108]

[0109] Table 12 Prediction results of external validation samples②

[0110]

[0111]

[0112] 2.5 Quantitative results of PLS ​​and BP-ANN models

[0113] The results of the PLS and BP-ANN quantitative analysis models of TUDCA, TCDCA, UDCA, CDCA and TCA are shown in Table 13. Compared with the results of the PLS and BP-ANN quantitative models, the TUDCA component quantitative model established by combining the PLS algorithm has the best effect, with R 2The TCDCA component quantitative model established by combining BP-ANN algorithm has the best effect, with R 2 The UDCA component quantitative model established by combining the PLS algorithm had the best effect, and the R 2 The CDCA component quantitative model established with the PLS algorithm has the best effect, with R 2 The TCA component quantitative model established by combining BP-ANN algorithm has the best effect, with R 2 They are 0.9978 and 0.9430 respectively, and the root mean square errors of correction and prediction are 0.0215 and 0.0255 respectively.

[0114] Table 13 Quantitative results of PLS ​​and BP-ANN models of index components

[0115]

[0116] By comparing the relative error results of external validation of PLS ​​and BP-ANN quantitative analysis models of TUDCA, TCDCA, UDCA, CDCA and TCA, it was found that the relative errors of the predicted value and the true value of the TUDCA component quantitative model were 0.08%-2.01% and 0.25%-8.25%, respectively; the relative errors of the predicted value and the true value of the TCDCA component quantitative model were 0.90%-3.01% and 0.41%-2.28%, respectively; the relative errors of the predicted value and the true value of the UDCA component quantitative model were 0.00%-11.54% and 1.04%-30.11%, respectively; the relative errors of the predicted value and the true value of the CDCA component quantitative model were 0.68%-5.57% and 2.97%-17.49%, respectively; the relative errors of the predicted value and the true value of the TCA component quantitative model were 1.93%-22.64% and 0.85%-4.26%, respectively. The average relative error of the prediction results of the NIR quantitative model of TUDCA, UDCA and CDCA components established by the PLS method was small, all maintained below 12%. The average relative error of the prediction results of the NIR quantitative model of TCDCA and TCA components established by the BP-ANN method was small, all maintained below 5%. Therefore, the prediction accuracy of the NIR quantitative model of TUDCA, UDCA and CDCA components established by the PLS method was higher, and the prediction accuracy of the NIR quantitative model of TCDCA and TCA components established by the BP-ANN method was higher. The comparison of the quantitative analysis models of the PLS and BP-ANN algorithms made the quantitative analysis results of the differential components TUDCA, TCDCA, UDCA, CDCA and TCA more accurate and reliable.

[0117] 3 Discussions

[0118] 3.1 Selection of quantitative algorithms

[0119] In the early stage of the experiment, the quantitative models of the five components, PLS, PCR, and MLR algorithms were selected, as shown in Table 14. Compared with the results of the PCR and MLR quantitative models, the model obtained by combining the PLS algorithm has the best performance. The PLS method combines the advantages of multiple linear regression, PCA and other analysis techniques to make the data structure simple and clear, while avoiding collinearity between multiple variables. Therefore, PLS (partial least squares regression) is selected as a multivariate data analysis algorithm as the correction model.

[0120] Table 14 Impact of regression method on the model

[0121]

[0122] Back propagating artificial neural net (BP-ANN) is a multi-layer feedforward neural network with reverse error propagation, and neurons are the basic structure of the neural network. The model includes an input layer, a hidden layer and an output layer. The number of neurons in the input layer must be the same as the dimension of the data, while the number of neurons in the output layer is equal to the number of spectral data to be fitted, and the number of neurons and the number of layers in the hidden layer are set according to a certain purpose. Generally, there is only one hidden layer, and the commonly used neural networks are three-layer. It has the advantages of strong fault tolerance, high reliability, and fast prediction speed. The present invention adopts the Levenberg-Marquardt (LM) method, the Sigmoid function and the Purelin formula as the training function, transfer function and output function of the BP-ANN model respectively.

[0123] 3.2 Preprocessing and characteristic wavelength selection

[0124] A variety of preprocessing algorithms and one characteristic wavelength selection algorithm were used. The model under each preprocessing algorithm has a certain improvement, which proves the necessity of spectral preprocessing before modeling. Due to the influence of external factors, the original spectrum will be affected. If the original spectrum and all wavelength points are directly used for modeling, the effect will be relatively poor. Therefore, it is necessary to preprocess the sample and select the wavelength before modeling. The preprocessing effects of MSC+SG smoothing and SNV+SG smoothing are similar, and both are better than single preprocessing. SG smoothing mainly plays a role in reducing data noise and effectively improves the signal-to-noise ratio of the data. Although the calculation methods of standard normal variation (SNV) and multivariate scattering correction (MSC) are inconsistent, they can both eliminate the influence of factors such as uneven particle size and optical path change. The continuous projection algorithm can effectively reduce the number of wavelengths, but some useful information is also eliminated when the wavelength is reduced, resulting in serious lack of spectral information. Therefore, the band range containing more characteristic wavelength points obtained by the continuous projection algorithm is selected as the optimal band.

[0125] 4 Summary

[0126] Combining PLS and BP-ANN algorithms, near-infrared quantitative models of the five components were established respectively. After comparison, it was found that the quantitative models of TUDCA, UDCA, and CDCA components established by PLS method had the best prediction effect. The pretreatment methods were 1 st D+SG+ncl, Ncl+SNV, SG+1 st D; After SPA characteristic band selection, 5000-9500cm -1 、5000-9000cm -1 、4000-7500cm -1is the modeling band; R of the calibration set and prediction set 2 The relative errors of the predicted value and the true value of the external validation were 0.08%-2.01%, 0.00%-11.54%, and 0.68%-5.57%, respectively. The BP-ANN method was the best to establish the quantitative model of TCDCA and TCA components, with R 2 The relative errors of the predicted value and the true value of the external verification were 0.41%-2.28% and 0.85%-4.26% respectively, with good prediction performance. The near-infrared spectroscopy technology was applied to the rapid determination of TUDCA, TCDCA, UDCA, CDCA and TCA contents in bear bile powder samples for the first time. The near-infrared content prediction model of bear bile powder with multiple components was established, which made up for the relatively single quality control index of bear bile powder medicinal materials and laid a foundation for further quality control of bear bile powder medicinal materials.

[0127] In summary, the present invention screens out TUDCA, TCDCA, UDCA, CDCA, and TCA as the significant different components for distinguishing bear bile powder from other animal bile powders; finally, near-infrared quantitative models of the five components are established respectively by combining PLS and BP-ANN algorithms, and a comparison is made. It is found that the quantitative models of TUDCA, UDCA, and CDCA components established by the PLS method have the best prediction effect, and the quantitative models of TCDCA and TCA components established by the BP-ANN method have the best effect. The relative errors of the externally verified predicted values ​​and true values ​​are both less than 5%, and the model has good prediction performance. The method can provide a new, simple and low-cost method for the quality evaluation of bear bile powder.

Claims

1. A method for constructing a prediction model for the content of bile acids in bear bile powder, characterized in that: The following steps are involved: 1) Extraction of bear bile powder characteristics: Take bear bile powder, collect near-infrared spectrum data and pre-process, and then use HPLC-ELSD method to determine the content of bile acid components in bear bile powder; 2) Model construction: SPA algorithm is used to extract the spectral characteristic wavelengths in the spectral data preprocessed in step 1), and the extracted spectral characteristic wavelengths are used as input variables, and the content of bile acid components is used as the target variable, and are imported into the PLS model for training, so as to obtain a prediction model for the content of bile acid components in bear bile powder; Or: using Matlab software to extract the first 11 principal components in the preprocessed spectral data in step 1), using the extracted 11 principal components as input parameters and the content of bile acid components as output parameters, and importing them into the BP-ANN model for training, thus obtaining a prediction model for the content of bile acid components in bear bile powder; The bile acid components are: tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and / or taurocholic acid.

2. The construction method according to claim 1, characterized in that: Step 1) The pretreatment method includes MSC, SNV, 1 st D. 2 ed Any one or more of D, SG, and normalization.

3. The construction method according to claim 2, characterized in that: The pretreatment method is 1. st D, SG and ncl, HPLC-ELSD method was used to determine the content of tauroursodeoxycholic acid in bear bile powder; SNV and Ncl, HPLC-ELSD method was used to determine the content of ursodeoxycholic acid in bear bile powder; 1 st D and SG, HPLC-ELSD method was used to determine the content of chenodeoxycholic acid in bear bile powder; for normalization, HPLC-ELSD method was used to determine the content of taurochenodeoxycholic acid or taurocholic acid in bear bile powder.

4. The construction method according to claim 1 or 3, characterized in that: The chromatographic conditions of the HPLC-ELSD method are: chromatographic column: Waters HSS T3 column, 4.6 mm × 250 mm, 5 μm; Mobile phase: acetonitrile-0.1% formic acid aqueous solution; gradient elution: 0-3min, 40%-60% acetonitrile; 3-5min, 60%-66% acetonitrile; 5-10min, 66%-70% acetonitrile; 10-13min, 70%-90% acetonitrile; 13-15min, 90%-60% acetonitrile; 15-18min, 60%-40% acetonitrile; column temperature 40℃; flow rate 1.0mL / min; ELSD drift tube temperature 90℃; nitrogen flow rate 2.0mL / min; gain 8; splitless mode; injection volume 10μL; And / or: the reference solution of the HPLC-ELSD method is prepared by dissolving tauroursodeoxycholic acid, taurochenodeoxycholic acid, ursodeoxycholic acid, chenodeoxycholic acid and / or taurocholic acid reference substances in methanol; the test solution is prepared by ultrasonic extraction of bear bile powder with methanol.

5. The construction method according to claim 1, characterized in that: Step 2) After training, the PLS model predicts the content of tauroursodeoxycholic acid, ursodeoxycholic acid and / or chenodeoxycholic acid in bear bile powder; after training, the BP-ANN model predicts the content of taurochenodeoxycholic acid and / or taurocholic acid.

6. The construction method according to claim 1, characterized in that: Step 2) The training function in the BP-ANN model is the Levenberg-Marquardt method, the transfer function is the Sigmoid function, and the output function is the Purelin formula; The BP-ANN model uses RMSE as an indicator and determines the number of hidden layer nodes through training, wherein the calculation formula for the number of hidden layer nodes is: L=(m+n) 1 / 2 +a, where L is the number of hidden layer nodes, n is the number of input layer nodes, m is the number of output layer nodes, and a is a natural number between 0 and 10.

7. A prediction model for the content of bile acid components in bear bile powder constructed by the construction method according to any one of claims 1 to 6.

8. A method for detecting the content of bile acid components in bear bile powder, characterized in that: Take the bear bile powder to be tested to collect infrared spectrum and pre-process it, then use SPA algorithm to extract the spectral characteristic wavelength in the pre-processed spectral data, input the spectral characteristic wavelength into the prediction model described in claim 7, and obtain the content of bile acid components in the bear bile powder; The bile acid components are: tauroursodeoxycholic acid, ursodeoxycholic acid and / or chenodeoxycholic acid. or: Take the bear bile powder to be tested to collect infrared spectrum and preprocess it, then use Matlab software to extract the first 11 principal components in the preprocessed spectrum data in step 1), input the extracted 11 principal components into the prediction model described in claim 7, and obtain the content of bile acid components in the bear bile powder; The bile acid components are: taurochenodeoxycholic acid and / or taurocholic acid.

9. A system for implementing the method for detecting the content of bile acids in bear bile powder as claimed in claim 8, characterized in that: include An input module, used for inputting the characteristics of Leonurus japonicus to be tested; A computing module, comprising the prediction model according to claim 7; The output module is used to output the final prediction results of the content of bile acid components in the bear bile powder.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is used to implement the method for detecting the content of bile acid components in bear bile powder as described in claim 8.

Images