Multi-source data fusion LightGBM-SHAP feature weight explanation-based euryale ferox medicinal material origin tracing method

Abstract

The invention discloses a euryale ferox medicinal material production place tracing method based on multi-source data fusion LightGBM-SHAP feature weight interpretation, and particularly relates to the field of traditional Chinese medicinal material production place tracing. The method comprises the following steps: detecting stable isotopes, mineral elements and starch contents of gordon euryale seeds from different producing areas, and constructing a gordon euryale seed multi-producing area multivariate fusion data set; adopting a LightGBM algorithm to construct a production place classification model based on the multi-source fusion data set; and calculating the contribution of different characteristics to a model result through an SHAP interpreter, and screening multiple types of weight factors influencing the euryale ferox producing area distinguishing as characteristic indexes for euryale ferox producing area distinguishing. The euryale ferox medicinal material producing area tracing method based on the combination of the euryale ferox multi-source data and the LightGBM-SHAP feature weight interpretation has the advantages of being easy and convenient to operate, stable in model performance, high in judgment accuracy and the like, and can provide technical reference for producing area tracing of euryale ferox or other traditional Chinese medicinal materials.

Description

Technical Field

[0001] The present invention relates to the technical field of origin tracing of Chinese medicinal materials, and particularly relates to a method for tracing the origin of Euryale ferox medicinal materials based on multi-source data fusion and LightGBM-SHAP feature weight interpretation. Background Art

[0002] Euryale ferox is the dried and mature seed kernel of the plant Euryale ferox Salisb. of the Nymphaeaceae family. It was first recorded in "Shennong's Herbal Classic" and listed as a top-grade medicinal material, being a well-known tonic Chinese medicine. Euryale ferox mainly has the effects of tonifying the middle qi, tonifying the kidney and securing essence, and removing dampness and stopping leukorrhea. It is clinically used in traditional Chinese medicine for symptoms such as spermatorrhea, enuresis, chronic diarrhea due to spleen deficiency, leukorrhea, and vaginal discharge. As a traditional Chinese medicinal material with both food and medicine homology in China, Euryale ferox is known as the "ginseng in water" and "longan in water", and ancient books even praised it as an excellent food and vegetable product that "makes babies not grow old and makes the elderly live longer". Euryale ferox is an annual large aquatic herbaceous plant, mainly planted in Jiangsu, Anhui, Guangdong, Jiangxi and other places in China. It is pointed out in the "Compilation of Standards for Genuine Regional Medicinal Materials" that Jiangsu is the genuine producing area of Euryale ferox, with better quality, also known as Jiangsu Euryale ferox. There are differences in the content of various nutrients in Euryale ferox from different origins, so there are also differences in price. Now, with the expansion of planting areas, the phenomenon of mixing the origins of Euryale ferox medicinal materials occurs from time to time, which not only affects the quality and efficacy of Euryale ferox medicinal materials, but also increases the instability of the Euryale ferox market. Therefore, it is particularly important to carry out research on tracing the origin of Euryale ferox medicinal materials.

[0003] Currently, the reported techniques for differentiating the origin of Euryale ferox mainly include liquid chromatography analysis technology, near-infrared spectroscopy analysis technology, etc. Although these methods can achieve the origin identification of Euryale ferox medicinal materials to a certain extent, there are still deficiencies such as single data source, low discrimination accuracy, and poor model interpretability. Therefore, developing a method for identifying the origin of Euryale ferox medicinal materials based on multi-source data fusion and interpretable machine learning algorithms is of great significance for accurately differentiating Euryale ferox from different origins and screening key factors for origin tracing. Summary of the Invention

[0004] Object of the Invention: The object of the present invention is to provide a method for tracing the origin of Euryale ferox medicinal materials based on multi-source data fusion and LightGBM-SHAP feature weight interpretation. This method collects samples of Euryale ferox medicinal materials from different producing areas across the country, and collects multi-source traceability datasets of stable isotopes, mineral elements and starch content of Euryale ferox from different origins through technologies such as EA-IRMS and ICP-MS; constructs an origin classification model based on the multi-source fusion dataset using the LightGBM algorithm; calculates the contribution of different features to the model results through the SHAP interpreter, and screens multiple types of weight factors affecting the differentiation of Euryale ferox origins as characteristic indicators for differentiating Euryale ferox origins, so as to achieve accurate tracing of the origin of Euryale ferox.

[0005] Technical solution: To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for tracing the origin of Euryale ferox medicinal materials based on multi-source data fusion LightGBM-SHAP feature weight interpretation, comprising the following steps:

[0007] (1) Collect Euryale ferox medicinal material samples from different origins, and detect the stable isotope, mineral element, and starch content data of the obtained Euryale ferox samples;

[0008] (2) Integrate the measured stable isotope, mineral element, and starch content data of Euryale ferox to construct a fusion dataset for tracing the origin of Euryale ferox; establish an origin tracing model for Euryale ferox based on the fusion dataset using the LightGBM algorithm; then use SHAP feature weight interpretation to calculate the contribution of different features to the model results, and screen out the optimal discriminant factors as the characteristic indicators for tracing the origin of Euryale ferox to achieve accurate identification of the origin of Euryale ferox.

[0009] As a preferred solution, for the method for tracing the origin of Euryale ferox medicinal materials described above, the stable isotopes detected in step (1) include δ 13 C, δ 15 N, δ 2 H, and δ 18 O; the mineral elements include Mg, Al, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Rb, Sr, Sb, Ba, Ca, K, P, Na, Mo; the multi-type starch components include total starch, amylose, amylopectin, and amylose / amylopectin.

[0010] As a preferred solution, for the method for tracing the origin of Euryale ferox medicinal materials described above, in step (1), the stable isotope content of Euryale ferox samples from different origins is determined by EA-IRMS; the mineral element content of Euryale ferox samples from different origins is determined by ICP-MS; the multi-type starch content of Euryale ferox from different origins is determined using a total starch and amylose kit.

[0011] As a preferred solution, for the method for tracing the origin of Euryale ferox medicinal materials described above, the steps for determining the stable isotope content by EA-IRMS are as follows: (1) δ 13 C and δ 15 N ratio: Weigh 5.0 - 6.0 mg of Euryale ferox powder into a special tin foil cup, convert the carbon and nitrogen elements of Euryale ferox into pure CO2 and N2 gases in the EA, and measure their isotope composition and the percentage content of C and N using IR-MS; EA conditions: The combustion tube filler is WO3, the temperature is 1000 - 1150 °C, and helium is used as the carrier gas; IR-MS conditions: The CO2 trap current is 100 μA, and the N2 trap current is 400 μA; (2) δ 2 H and δ 18O ratio: Weigh 0.4 - 0.6 mg of Euryale ferox powder and place it in the sample tray of the analyzer. Use EA to convert hydrogen and oxygen elements in the sample into pure H2 and CO gases, and use IR-MS to measure their isotope composition and the percentage content of H and O; EA conditions: the temperature of the high-temperature cracking furnace is 1400 - 1450 °C, and helium is used as the carrier gas; IR-MS conditions: the CO trap current is 200 μA, and the H2 trap current is 400 μA;

[0012] The method for determining the content of mineral elements by ICP-MS is as follows: Weigh 0.1 - 0.3 g of Euryale ferox powder into a polytetrafluoroethylene tank, add 4 - 6 mL of concentrated nitric acid and 1 - 3 mL of hydrogen peroxide to fully wet it, place it in a microwave digestion instrument. After the reaction is completed, cool the digestion solution to room temperature, dilute it with ultrapure water and make the final volume constant to 30 - 50 mL. The blank solution is prepared in the same way. Detect the content of each mineral element in all Euryale ferox digestion solutions by ICP-MS.

[0013] The method for detecting the content of multiple types of starch is as follows: Use the Megazyme K-TSTA-100A kit to determine the total starch content of Euryale ferox, use the Megazyme K-AMYL kit to determine the amylose content of Euryale ferox, and the amylopectin and amylose / amylopectin are calculated from the total starch and amylose.

[0014] As a more preferred scheme, for the method for tracing the origin of Euryale ferox medicinal materials described above, in the digestion method, weigh 0.2 g of Euryale ferox powder, add 5 mL of concentrated nitric acid, add 2 mL of hydrogen peroxide, and the temperature-rising conditions of the microwave digestion program are: raise the temperature to 120 °C within 10 minutes and keep it for 5 minutes, then raise the temperature to 160 °C within 7 minutes and keep it for 5 minutes, and then raise the temperature to 180 °C within 5 minutes and keep it for 15 minutes.

[0015] For the method for tracing the origin of Euryale ferox medicinal materials described above, its characteristics are that the working parameters of ICP-MS detection are: the high-frequency plasma emission power is 1300 - 1350 W; the detector voltage is -1800 to -1850 V, and the carrier gas is high-purity argon; the plasma gas flow rate is 15 - 17 L·min -1 , the nebulizer gas flow rate is 0.8 - 0.92 mL·min -1 , the collision gas flow rate is 4 - 4.5 mL·min -1 , use the jump peak mode to repeat sampling 2 - 6 times, use the standard solution to construct an external calibration curve, and calculate the concentration of each element.

[0016] As an optimal solution, for the method for tracing the origin of the Euryale ferox medicinal materials described above, in step (2), the LightGBM algorithm is established using the Python programming language, and the LightGBM classification package is obtained using the scikit-learn open platform. The objective function formula (1) followed by LightGBM is:

[0017]

[0018] where represents the loss function, which measures the difference between the predicted value and the true value; Ω(f k ) represents the regularization term, which controls the model complexity and prevents overfitting; K represents the total number of trees.

[0019] The samples are divided into a training set and a validation set according to a 7:3 ratio using stratified sampling to construct a LightGBM classification model. The classification performance of the model is evaluated using accuracy, recall, precision, F1 score, and Matthews correlation coefficient. The calculation formulas are as shown in (2) to (6), where TP represents true positive, the model predicts as positive and is actually positive; FP represents false positive, the model predicts as positive but is actually negative; FN represents false negative, the model predicts as negative but is actually positive; TN represents true negative, the model predicts as negative and is actually negative.

[0020]

[0021] As an optimal solution, for the method for tracing the origin of the Euryale ferox medicinal materials described above, in step (2), the SHAP feature weight interpretation includes global feature interpretation and local feature interpretation. The core calculation method of SHAP is formula (7):

[0022]

[0023] where Φi represents the Shapley value of each i-th feature; K represents a subset of the input features; M represents the set of all inputs; N represents the total number of features; represents the feature subset that does not contain feature i, and the summation traverses all subsets K that do not contain feature i; g x (K) represents the value function or feature function, which assigns a value to any feature subset K, indicating the benefit that the subset K can achieve through cooperation; g x (K∪{i}) - g x (K) represents the marginal contribution of feature i to subset K, indicating how much the value of subset K increases when feature i is added to subset K.

[0024] As a preferred solution, for the origin tracing method of the Euryale ferox medicinal materials described above, the optimal discriminant factor is determined by the SHAP variable weight value. The top 20 variables in the SHAP interpretation ranking are selected as the best variables. The optimal discriminant factors selected in step (2) are: Na, RAA, V, Ba, Sb, AL, %N, Cu, Ti, Mn, δ 2 H, Co, Al, TS, Ni, δ 18 O, δ 13 C, %C, Fe, Ca can be used as characteristic indicators for tracing the origin of Euryale ferox

[0025] As a preferred solution, for the Euryale ferox medicinal material origin classification model constructed based on LightGBM described above, the accuracy of the training set is 100%, the accuracy of the validation set is 97.67%, the recall rate is 0.98, the precision rate is 0.98, the F1 score is 0.98, and the Matthews correlation coefficient is 0.97

[0026] The present invention utilizes the differences in stable isotopes, mineral elements, and starch content of Euryale ferox medicinal materials from different origins to construct a multi-source tracing dataset for the origin of Euryale ferox; uses the LightGBM algorithm to construct an origin classification model, achieving accurate differentiation of Euryale ferox from different origins; calculates the contribution of different features to the model results through the SHAP interpreter, and screens multiple types of weight factors affecting the differentiation of Euryale ferox origins as characteristic indicators for differentiating Euryale ferox origins

[0027] Beneficial effects: The origin tracing method of Euryale ferox medicinal materials based on multi-source data fusion LightGBM-SHAP feature weight interpretation provided by the present invention has the following advantages compared with the prior art

[0028] The present invention first proposes to use multi-source data of stable isotopes, mineral elements, and starch to trace the origin of Euryale ferox medicinal materials. The stable isotope, mineral element, and starch fingerprint information can comprehensively reflect the exogenous and endogenous component content information of Euryale ferox from different origins. This method has the advantages of high detection accuracy, strong sensitivity, and reliable method, avoiding the uncertainty of subjective identification based on Euryale ferox medicinal materials, and has good popularization and application value

[0029] The LightGBM-SHAP feature weight interpretation classification method adopted by the present invention can construct an excellent Euryale ferox origin classification model. The SHAP feature weight interpretation can screen multiple types of weight factors affecting origin differentiation, providing an important idea for the optimization of key indicators for tracing the origin of Euryale ferox Description of the Drawings

[0030] Figure 1 Radar chart (a) and confusion matrix chart (b) of LightGBM model evaluation parameters for differentiating Euryale ferox from different origins

[0031] Figure 2 Global importance explanation plot for SHAP (a) and feature contribution plot for predicting a single origin (b).

[0032] Figure 3 The local feature interpretation force diagram for SHAP. DETAILED DESCRIPTION

[0033] The present invention is further explained below in conjunction with specific examples. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent modifications of the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0034] Example 1 Analysis and evaluation of stable isotopes, mineral elements and multi-type starch content of Euryale ferox from different origins

[0035] 1. Collection and preparation of Euryale ferox samples

[0036] Fresh Gorgon fruit samples were collected from Huai'an, Jiangsu Province, Chuzhou, Anhui Province, Changde, Hunan Province, Jining, Shandong Province, Shangrao, Jiangxi Province, and Zhaoqing, Guangdong Province during the harvest period. Detailed information on sample collection is shown in Table 1. Fresh samples were collected directly from the field, then washed, peeled, and dried at the same temperature; the dried samples were accurately divided into several portions, sealed in bags, labeled with origin information, and transported back to the laboratory. Finally, Gorgon fruit samples from different origins were ground into fine powder using a high-speed grinder, passed through a 50-mesh sieve, and stored in a desiccator for later use.

[0037] Table 1 Information of Euryale ferox samples from different origins

[0038] Number Place of origin Batch Longitude Latitude S1 Huai'an City, Jiangsu Province 24 119.18～119.28 33.06～33.11 S2 Zhaoqing City, Guangdong Province 24 112.61～112.70 23.00～23.19 S3 Shangrao City, Jiangxi Province 24 116.54～116.78 28.58～28.77 S4 Changde City, Hunan Province 24 111.34～112.19 28.85～29.48 S5 Chuzhou City, Anhui Province 23 118.93～119.08 32.84～32.89 S6 Jining City, Shandong Province 24 116.71～117.14 34.79～35.18

[0039] 2. Determination of stable isotope ratios of Euryale ferox from different origins

[0040] (1)δ 13 C and δ 15 N ratio: Weigh 5.0-6.0 mg of Gorgon fruit powder into a special tin foil cup, convert the carbon and nitrogen elements of the Gorgon fruit into pure CO2 and N2 gases in EA, and use IR-MS to measure its isotopic composition and the percentage content of C and N; EA conditions: the combustion tube is filled with WO3, the temperature is 1150°C, and helium (99.999%) is used as the carrier gas; IR-MS conditions: CO2 trap current 100μA, N2 trap current 400μA.

[0041] (2)δ 2 H and δ 18O ratio: Weigh 0.5 mg of Euryale ferox powder and place it in the analyzer sample tray. Use EA to convert hydrogen and oxygen elements in the sample into pure H2 and CO gases, and use IR-MS to measure their isotope composition and the percentage content of H and O; EA conditions: the temperature of the high-temperature cracking furnace is 1450 °C, and helium (99.999%) is used as the carrier gas; IR-MS conditions: the CO trap current is 200 μA, and the H2 trap current is 400 μA.

[0042] (3) Use Microsoft Excel and IBM SPSS 26.0 software to analyze the stable isotope data of Euryale ferox from different origins. Through multiple comparison analysis, the analysis results of the differences of 4 stable isotopes among the origins of Euryale ferox are obtained. See Table 2 for details. The average value range of δ 13 C is from -24.33‰ to -23.82‰, with the highest in Shangrao area of Jiangxi Province and the lowest in Changde area of Hunan Province. The δ 13 C value in Shangrao area of Jiangxi Province is significantly different from that in other areas (p < 0.05). The δ 13 C value distribution of Euryale ferox indicates that it belongs to C3 type plants, which is consistent with aquatic crops such as rice. The δ 13 C value differences among different origins are obvious, indicating that the δ 13 C value plays an important role in the origin identification of Euryale ferox. The δ 15 N average value range is from 2.12‰ to 2.65‰, with the highest in Jining area of Shandong Province and the lowest in Shangrao area of Jiangxi Province (p < 0.05). The δ 15 N value difference is mainly attributed to the nitrogen component in the soil and the fertilizer application situation. The lower δ 15 N value indicates that a large amount of chemical fertilizers may have been used during the ES planting process. The δ 2 H and δ 18 O isotopes show similar geographical distribution trends in six regions. The δ 2 H average value range is from -45.66‰ to -56.16‰, and the δ 18 O average value range is from 18.88‰ to 19.55‰. The δ 2 H and δ 18 O values in Huai'an area of Jiangsu Province and Chuzhou area of Anhui Province are significantly higher than those in other areas (p < 0.05), while the lowest is in Changde area of Hunan Province. Research shows that the δ 18 O value is greatly affected by longitude, latitude, water source and transpiration, especially the water source plays a key role in the hydrogen and oxygen isotope fractionation of aquatic plants. The above-mentioned 4 stable isotopes of Euryale ferox mostly have significant differences among different origins, but it is obviously one-sided to identify the origin through the difference of one isotope. The accurate identification of the origin of Euryale ferox can be maximally achieved by the integration of 4 isotopes.

[0043] Table 2 Analysis results of differences in stable isotopes among origins of Euryale ferox medicinal materials

[0044]

[0045] Note: Different lowercase letters indicate significant differences (p < 0.05).

[0046] 3. Determination of Mineral Element Contents in Euryale ferox Salisb. from Different Origins

[0047] (1) Pretreatment of Euryale ferox Salisb. Samples by Microwave Digestion

[0048] Weigh 0.2 g of Euryale ferox Salisb. powder into a polytetrafluoroethylene tank, add 5 mL of concentrated nitric acid and 2 mL of hydrogen peroxide to fully soak it, place it in a microwave digestion instrument. The temperature-rising conditions of the microwave digestion program are as follows: raise the temperature to 120 °C within 10 minutes and keep it for 5 minutes, then raise the temperature to 160 °C within 7 minutes and keep it for 5 minutes, and then raise the temperature to 180 °C within 5 minutes and keep it for 15 minutes. Cool the digestion solution to room temperature. After the reaction is completed, dilute it with ultrapure water and make the final volume up to 50 mL, shake well, and let it stand to obtain the digestion solution to be measured. The blank solution is prepared in the same way.

[0049] (2) Determination of Mineral Element Contents Based on ICP-MS

[0050] Use ICP-MS technology to determine the elements of Mg, Al, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Rb, Sr, Sb, Ba, Ca, K, P, and Na in Euryale ferox Salisb. samples. The working parameters of ICP-MS detection are as follows: the high-frequency plasma emission power is 1350 W; the detector voltage is -1850 V, and the carrier gas is high-purity argon; the plasma gas flow rate is 17 L·min -1 , the nebulizer gas flow rate is 0.92 mL·min -1 , the collision gas flow rate is 4.5 mL·min -1 , repeat sampling 3 times using the jump peak mode, construct an external calibration curve using the standard solution, and calculate the concentration of each element. The mineral element data of all Euryale ferox Salisb. samples are subjected to multiple comparisons and variance analysis using SPSS software, and the results are shown in Table 3.

[0051] Table 3 Mineral Element Contents and Results of Multiple Comparison Analysis of Euryale ferox Salisb. from Different Origins

[0052]

[0053] Note: Different lowercase letters indicate significant differences (p < 0.05).

[0054] The results showed that P and K were the two elements with the highest accumulation, and the average contents in Jining, Shandong and Shangrao, Jiangxi were the highest, reaching 1547.83 mg / kg and 927.81 mg / kg respectively. The average concentrations of Mg, Na and Ca were relatively high in Chuzhou, Anhui, being 317.62, 116.94 and 43.10 mg / kg respectively. Multiple comparison analysis indicated that except for the P element, significant differences existed among the other mineral elements in Euryale ferox from different producing areas (p<0.05). The differences in mineral elements in Euryale ferox from different geographical sources could be attributed to ecological and agricultural factors. The elemental compositions of Euryale ferox from different producing areas had their respective regional characteristics, and a discriminant model needed to be further constructed for analysis.

[0055] 4. Determination of multi-type starch contents in Euryale ferox from different producing areas

[0056] The total starch content of Euryale ferox was determined using the Megazyme K-TSTA-100A kit, and the amylose content was determined using the Megazyme K-AMYL kit. The amylopectin and amylose / amylopectin ratios were calculated from the total starch and amylose. Multiple comparisons and variance analysis were performed using SPSS software, and the results are shown in Table 4.

[0057] Table 4 Analysis results of multi-type starch contents in Euryale ferox from different producing areas

[0058]

[0059] Note: Different lowercase letters indicate significant differences (p<0.05)

[0060] The results of multi-type starch contents showed that the total starch (TS) contents in the regions of Jining, Shandong; Changde, Hunan; Shangrao, Jiangxi; and Zhaoqing, Guangdong were all higher than 70%, while the TS values in the regions of Huai'an, Jiangsu and Chuzhou, Anhui were lower than 70%. The TS content varied among different producing areas of Euryale ferox (p<0.05); the amylose (AL) content of Euryale ferox was the highest in Jining, Shandong and the lowest in Chuzhou, Anhui; while the amylopectin (AP) value was the highest in Shangrao, Jiangxi and the lowest in Huai'an, Jiangsu, indicating differences in the starch compositions of the six Euryale ferox producing areas. The results of the ratio of amylose to amylopectin (RAA) showed a more intuitive representation of the starch composition in different Euryale ferox regions, with high ratios of 4.91 and 4.72 shown in Chuzhou, Anhui and Shangrao, Jiangxi respectively.

[0061] Example 2 Construction of a discriminant model for the producing areas of Euryale ferox medicinal materials based on the LightGBM-SHAP feature weight interpretation method and screening of multi-type weight factors

[0062] (1) Establishment of the LightGBM classification model

[0063] Based on the above method, a LightGBM classification model was established based on the integrated dataset of Euryale ferox from different origins (stable isotopes, mineral elements, and multi-type starches). The model performance evaluation parameters (a) and confusion matrix (b) are shown in Figure 1 . The discrimination accuracy of the training set of the LightGBM model is 100%, and the discrimination accuracy of the validation set is 97.67%, indicating that the LightGBM algorithm combined with multi-source data of Euryale ferox achieved accurate differentiation of different origins. The precision, recall, F1-score, and Matthews index are 0.98, 0.98, 0.98, and 0.97 respectively. The confusion matrix diagram shows that only one sample was misclassified from Shangrao, Jiangxi to Zhaoqing, Guangdong, indicating that the origin traceability model of Euryale ferox constructed based on the LightGBM integrated model has excellent reliability.

[0064] (2) Global feature interpretation and key factor screening based on SHAP interpreter

[0065] Based on the LightGBM-SHAP strategy, a stacked bar Figure 2 a) was initially fitted to explain the global feature contributions of the model. The absolute SHAP values of different features show that the top 20 variables affecting classification are arranged in descending order (Na, RAA, V, Ba, Sb, AL, %N, Cu, Ti, Mn, δ 2 H, Co, Al, TS, Ni, δ 18 O, δ 13 C, %C, Fe, Ca), including 3 isotopes, 14 mineral elements, and 3 starch indexes. Obviously, the element Na was determined as the most critical index for origin differentiation, with the highest average absolute SHAP value of 8.0. The elements with higher scores also include V, Ba, Sr, %N, Cu, Ti, and Mn, indicating that multiple elements are important variables for tracing the origin of Euryale ferox. In addition to δ 15 N, δ 13 C, δ 2 H, and δ 18 O were also determined as key traceability indexes. In addition, multi-type starch indexes such as RAA, AL, and TS also play important roles in classification, indicating that there are also differences in the internal quality of Euryale ferox from different origins.

[0066] The beeswarm plot Figure 2 b) provides the ranking of the classification contribution values of all chemical fingerprint features to a single origin. The horizontal axis represents the impact on the model output using SHAP values, and the color gradient represents the change in the corresponding feature values. For example, RAA, Na, AL, Ba, V, δ 18 O are the most important features for the production area of Chuzhou, Anhui. Specifically, RAA, Na, V, δ 18The increase in the O content enhances the possibility of classifying Euryale ferox as being from the Chuzhou production area in Anhui, while the decrease in the contents of aluminum (AL) and Ba has a negative impact on the classification of Chuzhou in Anhui; the mineral elements Na, Al, and Ca have a positive effect on predicting Euryale ferox samples from Zhaoqing in Guangdong; Na is the variable with the greatest contribution to differentiating Euryale ferox samples from Changde in Hunan and Huai'an in Jiangsu. Specifically, it has a positive impact on the production area in Hunan and a negative impact on the samples from Huai'an in Jiangsu, which may be the key variable for differentiating the two production areas; the elements Sb and Na are the main contributing variables for classifying Euryale ferox into the Shangrao production area in Jiangxi, showing positive SHAP values when the abundances of N and V increase; in addition, RAA, Al, and TS play important roles in predicting Euryale ferox samples from Jining in Shandong.

[0067] (3) Local feature interpretation based on the SHAP interpreter

[0068] Based on the LightGBM-SHAP strategy, the local feature prediction of the origin classification of Euryale ferox was further fitted. The pink bars indicate that the corresponding features increase the possibility of origin prediction, and the length of the bar represents the magnitude of its influence. As Figure 3 shown, the model output is to the right of the baseline, and the predicted value exceeds 1.0, indicating that the origin of the random sample is accurate. However, the feature contributions of each sample are different. For example, the baseline value for Jining in Shandong is 5.778, and the predicted value reaches 8.0, indicating that the variables Ba, V, Na, Co, Al, Ca, and Sb enhance the prediction of the samples from Jining in Shandong. In summary, local feature interpretation allows for observing the key variables and their directions of influence on individual samples in the LightGBM model, further demonstrating the reliability of LightGBM.

[0069] The above experimental results show that a method for tracing the origin of Euryale ferox medicinal materials based on multi-source data fusion LightGBM-SHAP feature weight interpretation provided by the present invention can accurately distinguish Euryale ferox from different origins. LightGBM-SHAP feature weights explain the reliability of the origin classification model and screen multiple types of weight factors that affect the differentiation of Euryale ferox origins. The selected variables can be used as regional characteristic indicators for tracing the origin of Euryale ferox. The proposed method has important guiding significance for the accurate tracing of the origin of Euryale ferox medicinal materials and is of great significance for protecting the authenticity and effectiveness of Euryale ferox in the market circulation.

[0070] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for tracing the origin of Gorgon fruit medicinal materials based on multi-source data fusion LightGBM-SHAP feature weight interpretation, characterized in that: The following steps are involved: (1) Collect samples of Euryale ferox from different origins and detect the stable isotopes, mineral elements and starch content of the samples; (2) The measured stable isotope, mineral elements and starch content data of Gorgon fruit were fused to construct a fusion dataset for Gorgon fruit origin traceability. The LightGBM algorithm was used to establish an origin traceability model based on the Gorgon fruit fusion dataset. Then, the SHAP feature weight interpretation was used to calculate the contribution of different features to the model results, and the optimal discriminant factor was selected as the characteristic indicator for Gorgon fruit origin traceability, so as to achieve accurate identification of the origin of Gorgon fruit.

2. The method for tracing the origin of Gorgon Fruit medicinal material according to claim 1, characterized in that: The stable isotopes detected in step (1) include δ 13 C.δ 15 N, δ 2 H and δ 18 O; mineral elements include Mg, Al, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Rb, Sr, Sb, Ba, Ca, K, P, Na, Mo, %C, %N; multiple types of starch components include total starch, amylose, amylopectin, and amylose / amylopectin.

3. The method for tracing the origin of Gorgon Fruit medicinal material according to claim 1, characterized in that: In the step (1), EA-IRMS is used to determine the stable isotope content of the Gorgon fruit samples from different origins; ICP-MS is used to determine the mineral element content of the Gorgon fruit samples from different origins; and a total starch and amylose kit is used to determine the content of multiple types of starch in Gorgon fruit from different origins.

4. The method for tracing the origin of Gorgon Fruit medicinal material according to claim 3, characterized in that: The steps of EA-IRMS for determining the stable isotope content are as follows: (1) δ 13 C and δ 15 N ratio: Weigh 5.0-6.0 mg of Gorgon fruit powder into a special tin foil cup, convert the carbon and nitrogen elements of Gorgon fruit into pure CO2 and N2 gas in EA, and use IR-MS to measure its isotopic composition and the percentage content of C and N; EA conditions: the combustion tube is filled with WO3, the temperature is 1110-1150℃, and helium is used as the carrier gas; IR-MS conditions: CO2 trap current 100μA, N2 trap current 400μA; (2)δ 2 H and δ 18 O ratio: Weigh 0.4-0.6 mg of Euryale ferox powder and place it in the sample tray of the analyzer. Use EA to convert hydrogen and oxygen in the sample into pure H2 and CO gas. Use IR-MS to measure its isotopic composition and the percentage content of H and O. EA conditions: high temperature cracking furnace temperature 1400-1450℃, helium as carrier gas; IR-MS conditions: CO trap current 200μA, H2 trap current 400μA; The ICP-MS method for determining the content of mineral elements is as follows: weigh 0.1-0.3 g of Euryale ferox powder in a polytetrafluoroethylene tank, add 4-6 mL of concentrated nitric acid and 1-3 mL of hydrogen peroxide to fully infiltrate, place in a microwave digestion instrument, after the reaction is completed, cool the digestion solution to room temperature, dilute and fix the volume with ultrapure water, prepare a blank solution in the same way, and detect the content of each mineral element in all Euryale ferox digestion solutions in ICP-MS; The multi-type starch content detection method is as follows: the total starch content of Euryale ferox is determined by using a Megazyme K-TSTA-100A kit, the amylose content of Euryale ferox is determined by using a Megazyme K-AMYL kit, and amylopectin and amylose / amylopectin are calculated from total starch and amylose.

5. The method for tracing the origin of Gorgon Fruit medicinal material according to claim 1, characterized in that: The ICP-MS detection working parameters are as follows: high-frequency plasma emission power is 1300-1350W; detector voltage is -1800-1850V; carrier gas is high-purity argon; plasma gas flow rate is 15-17L·min -1 , the nebulizing gas flow rate is 0.8~0.92mL·min -1 The collision gas flow rate is 4-4.5 mL min -1 , repeated sampling was performed using the peak-hopping mode, and an external calibration curve was constructed using standard solutions to calculate the concentration of each element.

6. The method for tracing the origin of Gorgon Fruit medicinal material according to claim 1, characterized in that: In step (2), the LightGBM algorithm is established using the Python learning language, and the LightGBM classification package is obtained using the scikit-learn open platform. The objective function formula (1) followed by LightGBM is: in represents the loss function, which measures the difference between the predicted value and the actual value; Ω(f k ) represents the regularization term, which controls the model complexity and prevents overfitting; K represents the total number of trees; Stratified sampling was used to divide the samples into training set and validation set in proportion, and the LightGBM classification model was constructed. The classification performance of the model was evaluated by accuracy, recall, precision, F1 score and Matthews correlation coefficient. The calculation formulas are shown in (2) to (6), where TP represents true positive examples, which are predicted by the model as positive and are actually positive; FP represents false positive examples, which are predicted by the model as positive but are actually negative; FN represents false negative examples, which are predicted by the model as negative but are actually positive; TN represents true negative examples, which are predicted by the model as negative and are actually negative.

7. The method for tracing the origin of Gorgon Fruit medicinal material according to claim 1, characterized in that: The SHAP feature weight interpretation in step (2) includes global feature interpretation and local feature interpretation. The SHAP core calculation method is formula (7): Where Φi represents the Shapley value of each i feature; K represents a subset of input features; M represents the set of all inputs; N represents the total number of features; represents the feature subset that does not contain feature i, and the sum is traversed over all subsets K that do not contain feature i; g x (K) represents the value function or feature function, which assigns a value to any feature subset K, indicating the benefits that subset K can achieve through cooperation; g x (K∪{i})-g x (K) represents the marginal contribution of feature i to subset K, indicating how much the value of subset K increases when feature i is added to subset K.

8. The method for tracing the origin of Euryale ferox medicinal material according to claim 1, characterized in that: The optimal discriminant factors screened in step (2) are: Na, RAA, V, Ba, Sb, AL, %N, Cu, Ti, Mn, δ 2 H, Co, Al, TS, Ni, δ 18 O, δ 13 C, %C, Fe, and Ca can be used as characteristic indicators for tracing the origin of Gorgon fruit.

9. The origin classification model of Gorgon fruit medicinal material based on LightGBM according to claim 6 is characterized in that: The accuracy of the training set is 100%, the accuracy of the validation set is 97.67%, the recall rate is 0.98, the precision rate is 0.98, the F1 score is 0.98, and the Matthews correlation coefficient is 0.

97.

10. The method for tracing the origin of Gorgon Fruit medicinal material according to claim 1, characterized in that: The origins of the Euryale ferox samples in step (1) include: Jiangsu, Anhui, Jiangxi, Hunan, Guangdong and Shandong.

Images