Drug classification method based on colloid AuNPs (at) AgCl and soft voting model

Abstract

The invention discloses a drug classification method based on colloid AuNPs (at) AgCl and a soft voting model, and the method comprises the steps: preparing colloidal gold nanoparticles AuNPs based on a seed growth method; a colloidal AuNPs (at) AgCl solution with an AgCl activation layer is prepared; the method comprises the following steps: taking colloid AuNPs (at) AgCl as an SERS (Surface Enhanced Raman Scattering) substrate, and obtaining Raman spectrum data of a unitary drug and a binary drug; performing normalization processing and feature screening on the Raman spectrum data in sequence to obtain a feature set; and assigning a weight to each base classifier through a soft voting method, obtaining an integrated classifier, and classifying the unitary drugs and the binary drugs. According to the soft voting ensemble learning classifier, four base classifiers including an extreme random tree, a random forest, a support vector machine and XGBoost are integrated, the classifier can quickly classify and identify unitary drugs and binary drugs, and the classification accuracy is greatly improved.

Description

Technical Field

[0001] The present application belongs to the technical field of drug classification, and specifically relates to a drug classification method based on colloidal AuNPs@AgCl and a soft voting model. Background Art

[0002] Monodrugs are drugs containing only one chemical substance, such as methamphetamine, morphine, or methadone. Binary drugs, on the other hand, are drugs containing two chemical substances. These drugs are highly addictive and harmful, with serious consequences for families, society, and the nation. Conventional methods such as gas chromatography, liquid chromatography, and liquid chromatography-tandem mass spectrometry are not suitable for rapid on-site testing due to their high requirements for instrumentation and specialized technicians.

[0003] In recent years, surface enhanced Raman spectroscopy (SERS) technology has been widely used in the field of drug analysis due to its high sensitivity, high stability and fingerprint recognition properties. When detecting drugs, we can determine whether the sample contains a certain drug based on whether the sample contains the "key" characteristic peak Raman signal. However, in recent years, with the continuous advancement of science and technology, the manufacturing and circulation methods of drugs have become increasingly covert and complex, posing greater challenges to investigation and crackdown work. At the same time, binary drugs containing two harmful substances have also begun to circulate in the illegal market, further increasing the difficulty of drug detection and identification. At present, there are relatively more studies on the classification of monovalent drugs, but relatively few studies on the classification of monovalent and binary drugs.

[0004] Currently, researchers have developed a series of high-quality SERS substrates for drug analysis and detection. For example, CN102886933B discloses a noble metal nanocone array structure coated on a substrate as a SERS substrate. Although this method has a low detection limit for ketamine, the preparation process relies on large magnetron sputtering machines, which takes too long and is only effective for single-drug detection. Therefore, preparing a colloidal SERS substrate with strong Raman activity can significantly shorten the entire detection process. Moreover, this active substrate can be placed directly in a quartz tube for detection, without having to drip it onto a substrate and wait for the sample to dry before measurement. This is of great significance for on-site drug detection.

[0005] In addition, the application of machine learning methods or deep learning technology to automatically classify a large number of sample data of suspected drugs has become a current research hotspot. For example, CN117929351B discloses a method for detecting amphetamine drugs in aqueous solutions and solid powders by establishing a quantitative analysis model based on the partial least squares method. This type of method, with its wide adaptability and automated learning capabilities, can effectively utilize a large amount of high-quality Raman spectral data obtained from the SERS substrate to achieve efficient and accurate classification of monolithic drugs. However, if a binary mixture sample of the target object is detected, the quality of the acquired SERS spectral data may be reduced due to the uneven adsorption of multiple molecules in the hot spot area of the SERS substrate, thereby affecting the accuracy of classification model training and prediction, which poses a huge challenge to the use of traditional machine learning models for classification. Summary of the Invention

[0006] The present application provides a drug classification method based on colloidal AuNPs@AgCl and a soft voting model to solve the technical problem of reduced reliability caused by the low accuracy of a single classifier in existing machine learning technology.

[0007] To solve the above technical problems, the present application adopts a technical solution: a drug classification method based on colloidal AuNPs@AgCl and a soft voting model, comprising:

[0008] S1. Preparation of colloidal gold nanoparticles (AuNPs) based on the seed-based growth method.

[0009] S2. Preparation of colloidal AuNPs@AgCl solution with AgCl activation layer;

[0010] S3. Obtaining Raman spectra of monovalent and binary drugs using colloidal AuNPs@AgCl as a SERS substrate.

[0011] S4. performing normalization processing and feature screening on the Raman spectral data in sequence to obtain a feature set;

[0012] S5. Based on extreme random trees, random forests, support vector machines, and XGBoost as base classifiers, a weight is assigned to each base classifier through the soft voting method to obtain an integrated classifier and classify monoid and binary drugs.

[0013] Furthermore, the method of step S1 includes:

[0014] S11. At room temperature, trisodium citrate (C6H5Na3O7) was mixed with ultrapure water in a mass-to-volume ratio of 100 mg:10 mL to obtain a trisodium citrate solution and placed in an ultrasonic bath for 15-20 minutes;

[0015] S12. In an ice bath, mix HAuCl4 with ultrapure water at a mass volume ratio of 100 mg:50 mL to obtain a HAuCl4 solution and sonicate for 15-20 minutes.

[0016] S13. The trisodium citrate solution obtained in step S1.1 and step S1.2 was mixed with chloroauric acid solution in a volume ratio of 1:25 and then placed in a preheated oil bath and boiled;

[0017] S14. After the solution has boiled for 2-5 minutes, an appropriate amount of low-concentration chloroauric acid solution (mass-to-volume ratio of chloroauric acid to ultrapure water: 33.979 mg:10 mL) is rapidly added to the solution under magnetic stirring (600 rpm). Stirring is continued for 15-20 minutes, followed by natural cooling to room temperature to obtain a gold seed solution.

[0018] S15. Mix ultrapure water and gold seed solution in a volume ratio of 4:1 and heat to boiling. Then add the trisodium citrate solution obtained in step S1.2. After 3 minutes, slowly add an appropriate amount of low-concentration chloroauric acid solution. Continue stirring for 15-20 minutes and then naturally cool to room temperature to obtain gold nanoparticles AuNPs.

[0019] Further, step S2 includes:

[0020] S21. Gold nanoparticles (AuNPs) were centrifuged at 5000 rpm for 15-20 minutes. The supernatant was removed and the solution was redissolved in deionized water. The solution was then ultrasonically dispersed in an ultrasonic bath to obtain colloidal AuNPs after centrifugation. The volume ratio of the colloidal AuNPs after centrifugation to the original solution was 1:5.

[0021] S22. Sodium chloride solution and silver nitrate solution are added to the centrifuged colloidal AuNPs successively. The colloidal AuNPs are kept fully stirred during the addition process. After the reaction is completed, a colloidal AuNPs@AgCl solution is obtained.

[0022] Furthermore, the volume ratio of the centrifuged colloidal AuNPs, sodium chloride solution, and silver nitrate solution was 50:5:1, the concentration of the sodium chloride solution was 1.5 M / L, the concentration of the silver nitrate solution was 10 mM / L, and the stirring reaction time was 5-10 min.

[0023] Furthermore, the method of step S3 includes:

[0024] S31. Dilute the drug solution to 100 ppm, thoroughly mix the drug and colloidal AuNPs@AgCl at a volume ratio of 1:9, and transfer the mixture into multiple quartz tubes to obtain multiple drug samples. The drug solution should include a single drug solution and a mixed solution of two drugs.

[0025] S32. Place the quartz tube containing the drug sample directly into the Raman spectrometer, adjust the laser position so that the laser hits the center of the quartz tube, set the Raman spectrometer integration time to 10s and the laser power to 15mW, repeat the measurement multiple times for each sample, and obtain Raman data including Raman signal intensity.

[0026] Further, the method of step S4 includes:

[0027] S41. Based on Raman data, in the range of 400-2000 cm -1 Find the peak I with the largest Raman signal intensity within the beam range max and the lowest value I min ;

[0028] S42. Based on the maximum-minimum normalization method and formula (1), the Raman signal intensity I is normalized to between 0 and 1; wherein formula (1) is:

[0029] I * =(II min ) / (I max -I min ) (1);

[0030] Among them, I * Represents the normalized Raman signal intensity, I max Indicates 400-2000cm -1 The maximum Raman signal intensity within the beam range, I min Indicates 400-2000cm -1 The lowest value of the Raman signal within the beam range;

[0031] S43. Name the normalized Raman signals x1, x2, ..., x 1601 , obtain the normalized first data set;

[0032] S44. Perform dimensionality reduction on the first dataset using the t-SNE algorithm to obtain a second dataset of drugs after dimensionality reduction. The number of new feature vectors after dimensionality reduction is set to 4, the learning rate is set to 100, and the maximum number of iterations is set to 300.

[0033] Further, the method of step S5 includes:

[0034] S51. Split the second dataset into a training set and a test set in a ratio of 7:3, and train them using four machine learning classifiers: Extreme Random Trees, Random Forest, Support Vector Machine, and XGBoost, to obtain the accuracy of drug classification.

[0035] S52. Build a soft voting classification model that integrates four base classifiers: extreme random trees, random forests, support vector machines, and XGBoost.

[0036] S53. Based on the soft voting classification model, the drugs are classified. The final prediction result should be The largest category is:

[0037]

[0038] in, Represents the final prediction result, is the maximum index function.

[0039] Furthermore, the method for constructing the soft voting classification model in step S52 includes:

[0040] S521. Based on the accuracy of each base classifier on the test set, assign a weight to each base classifier; wherein the weight of the base classifier is:

[0041]

[0042] Among them, w e is the corresponding weight of the e-th classifier; e = 1, 2, 3, 4 represent four machine learning classifiers: extreme random tree, random forest, support vector machine and XGBoost respectively; AUC e is the AUC value of the e-th base classifier; the denominator Represents the sum of the AUCs of the four base classifiers, which is used to normalize the weights to ensure that the sum of the weights is 1;

[0043] S522. Based on the weight of each base classifier, the final prediction result is weighted averaged; wherein, the accuracy of the soft voting prediction is:

[0044]

[0045] in, Indicates that the integrated model is that the drug sample belongs to category C g The final weighted probability of g =1,2,3,…,6, corresponding to the six drugs: morphine, methadone, methamphetamine, a binary mixture of morphine and methadone, a binary mixture of morphine and methadone, and a binary mixture of methadone and methamphetamine; w e is the corresponding weight of the e-th classifier; Indicates that the e-th base classifier predicts that the drug belongs to category C g The accuracy rate.

[0046] The beneficial effects of this application are:

[0047] (1) The SERS substrate of the present application has lower production costs and simpler production steps than chip-type substrates, and can also effectively avoid the uneven distribution of substances adsorbed on chip-type substrates during the drying process;

[0048] (2) Based on the SERS spectra obtained from gold nanoparticles, this application proposes the use of the T-SNE algorithm for feature screening. This method can not only effectively screen out the effective information in the SERS spectrum data, but also can remove a large amount of redundant data to the greatest extent, which is beneficial for subsequent machine learning classification;

[0049] (3) The soft voting ensemble learning classifier of the present application integrates four base classifiers: extreme random trees, random forests, support vector machines, and XGBoost. The classifier can quickly classify and identify monolithic and binary drugs, and greatly improve the classification accuracy. It can also solve the problems of drug detection in liquid samples at low concentrations and automated analysis and classification of drugs. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 1 is a flow chart of an embodiment of a drug classification method based on colloidal AuNPs@AgCl and a soft voting model of the present application;

[0051] Figure 2 This is the morphology of the SERS substrate (gold nanoparticles of about 40 nm) proposed in this application;

[0052] Figure 3 This is the EDS elemental scanning image of the AuNPs@AgCl SERS substrate proposed in this application;

[0053] Figure 4 The SERS spectra of six drugs obtained using gold nanoparticles as the SERS substrate for this application;

[0054] Figure 5 This is a ROC curve diagram obtained by the extreme random tree-based base classifier in Example 1 of the present application;

[0055] Figure 6 This is the ROC curve obtained by the random forest-based base classifier in Example 1 of the present application;

[0056] Figure 7 This is the ROC curve obtained by the base classifier based on the support vector machine in Example 1 of the present application;

[0057] Figure 8 The ROC curve diagram obtained by the XGBoost-based base classifier in Example 1 of the present application;

[0058] Figure 9This is a ROC curve diagram of the soft voting integration algorithm for classifying six types of drugs in an embodiment of the drug classification method based on colloidal AuNPs@AgCl and a soft voting model of this application. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to specific embodiments.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] See Figure 1 , Figure 1 The figure is a flow chart of an embodiment of a drug classification method based on colloidal AuNPs@AgCl and a soft voting model. The method includes:

[0062] S1. Preparation of colloidal gold nanoparticles AuNPs based on the seed growth method.

[0063] Specifically, step S1 includes:

[0064] S11. At room temperature, trisodium citrate (C6H5Na3O7) was mixed with ultrapure water in a mass-to-volume ratio of 100 mg:10 mL to obtain a trisodium citrate solution and placed in an ultrasonic bath for 15-20 minutes;

[0065] S12. In an ice bath, mix HAuCl4 with ultrapure water at a mass volume ratio of 100 mg:50 mL to obtain a HAuCl4 solution and sonicate for 15-20 minutes.

[0066] S13. Mix the trisodium citrate and chloroauric acid solutions obtained in steps S1.1 and S1.2 in a volume ratio of 1:25 and place in a preheated oil bath and boil.

[0067] S14. After the solution has boiled for 2-5 minutes, quickly add an appropriate amount of low-concentration chloroauric acid solution (the mass volume ratio of chloroauric acid to ultrapure water is 33.979 mg:10 mL) under magnetic stirring (600 rpm). Continue stirring for 15-20 minutes and then naturally cool to room temperature to obtain a gold seed solution.

[0068] S15. Mix ultrapure water and gold seed solution in a volume ratio of 4:1 and heat to boiling. Then add the trisodium citrate solution obtained in step S12. After 3 minutes, slowly add an appropriate amount of low-concentration chloroauric acid solution. Continue stirring for 15-20 minutes and then naturally cool to room temperature to obtain gold nanoparticles AuNPs.

[0069] See Figure 2 The colloidal gold nanoparticles in this embodiment have a particle size of about 40 nm.

[0070] S2. Preparation of colloidal AuNPs@AgCl solution with AgCl activation layer.

[0071] Specifically, step S2 includes:

[0072] S21. Take the gold nanoparticles AuNPs prepared in S1, centrifuge them at 5000 r / min for 15-20 min, remove the supernatant and redissolve them with deionized water, then put them into an ultrasonic tank for ultrasonic dispersion to obtain colloidal AuNPs after centrifugation.

[0073] Among them, the volume ratio of colloidal AuNPs after centrifugation to the original solution should be 1:5.

[0074] S22. Sodium chloride solution and silver nitrate solution were added to the centrifuged colloidal AuNPs successively, and the colloidal AuNPs were kept fully stirred during the addition process. After the reaction was completed, a colloidal AuNPs@AgCl solution was obtained. The volume ratio of the centrifuged colloidal AuNPs, sodium chloride solution, and silver nitrate solution was 50:5:1, the concentration of the sodium chloride solution was 1.5 M / L, the concentration of the silver nitrate solution was 10 mM / L, and the stirring reaction time was 5-10 min.

[0075] See also Figure 3 As shown, the EDS element scanning image of the colloidal AuNPs@AgCl substrate obtained in this example.

[0076] See also Figure 4 As shown, in this embodiment, “SERS spectra of 6 drugs” can be obtained by using gold nanoparticles as the SERS substrate.

[0077] S3. Based on colloidal AuNPs@AgCl as the SERS substrate, Raman spectral data of monovalent and binary drugs were obtained.

[0078] Specifically, step S3 includes:

[0079] S31. Dilute the drug solution to 100 ppm, thoroughly mix the drug and colloidal AuNPs@AgCl at a volume ratio of 1:9, and transfer the mixture into multiple quartz tubes to obtain multiple drug samples. The drug solution should include a solution of a single drug and a mixed solution of two drugs.

[0080] S32. Place the quartz tube containing the drug sample directly into the Raman spectrometer, adjust the laser position so that the laser hits the center of the quartz tube, set the Raman spectrometer integration time to 10s and the laser power to 15mW, repeat the measurement multiple times for each sample, and obtain Raman data including the Raman signal intensity I.

[0081] S4. Normalize and feature screen the Raman spectral data in sequence to obtain a feature set.

[0082] Specifically, step S4 includes:

[0083] S41.400-2000cm -1 Find the peak with the largest Raman signal intensity within the beam range and name it I max At the same time, find the lowest value of the Raman signal and name it I min ;

[0084] S42. Normalize the intensity I of the Raman signal to between 0 and 1 according to the maximum and minimum normalization method; name the normalized Raman signals x1, x2, ...x 1601 , obtain the normalized first data set;

[0085] Among them, normalization is performed by formula (1):

[0086] I * =(II min ) / (I max -I min ) (1);

[0087] Among them, I * Represents the normalized Raman signal intensity, I max Indicates 400-2000cm -1 The maximum Raman signal intensity within the beam range, I min Indicates 400-2000cm -1 The lowest value of the Raman signal within the beam range.

[0088] S43. Use the T-SNE algorithm to analyze the 400-2000cm -1 The Raman signal within the beam range is reduced in dimension, and finally a second data set including 3-6 fusion features is obtained.

[0089] S5. Based on extreme random trees, random forests, support vector machines, and XGBoost as base classifiers, a weight is assigned to each base classifier through the soft voting method to obtain an integrated classifier and classify monoid and binary drugs.

[0090] Specifically, step S5 includes:

[0091] S51. Split the second dataset into a training set and a test set in a ratio of 7:3, and train them using four machine learning classifiers: Extreme Random Trees, Random Forest, Support Vector Machine, and XGBoost, to obtain the accuracy of drug classification.

[0092] S52. Construct a soft voting classification model, which integrates four base classifiers: extreme random trees, random forests, support vector machines, and XGBoost.

[0093] The method of constructing the soft voting classification model in step S52 includes:

[0094] S521. Based on the accuracy of each base classifier on the test set, assign a weight to each base classifier; wherein the weight of the base classifier is:

[0095]

[0096] Among them, w e is the corresponding weight of the e-th classifier (e=1, 2, 3, 4 and w1+w2+w3+w4=1); e=1, 2, 3, 4 represent the four machine learning classifiers of extreme random tree, random forest, support vector machine and XGBoost respectively; AUC e is the AUC value of the e-th base classifier; the denominator Represents the sum of the AUCs of the four base classifiers, which is used to normalize the weights to ensure that the sum of the weights is 1;

[0097] S522. Based on the weight of each base classifier, the final prediction result is weighted averaged; wherein, the accuracy of the soft voting prediction is:

[0098]

[0099] in, Indicates that the soft voting classification model believes that the drug sample belongs to category C g The final weighted probability of g =1,2,3,…6, corresponding to the six drugs: morphine, methadone, methamphetamine, a binary mixture of morphine and methadone, a binary mixture of morphine and methadone, and a binary mixture of methadone and methamphetamine; w e is the corresponding weight of the e-th classifier; Indicates that the e-th base classifier predicts that the drug belongs to category C g The accuracy rate.

[0100] S53. Based on the soft voting classification model, the drugs are classified. The final prediction result should be The largest category is:

[0101]

[0102] in, Represents the final prediction result, The soft voting classification model classifies drug samples as belonging to category C g The final weighted probability of is the maximum index function.

[0103] See Figure 5-8 , in this embodiment, the extreme random tree ( Figure 5 ), Random Forest( Figure 6 ), support vector machine ( Figure 7 ) and XGBoost( Figure 8 ) The classification performance of the four classifiers for each drug is poor, and most of the classification accuracy is less than 80%. Figure 6 Taking the optimal random forest as an example, its classification accuracy for morphine, methadone, and methamphetamine is 70%, 79%, and 58%, respectively. Its classification accuracy for the binary mixture of morphine and methadone, the binary mixture of morphine and methadone, and the binary mixture of methadone and methamphetamine is 55%, 86%, and 58%, respectively.

[0104] See Figure 9 In this embodiment, the soft voting method is used as a combination strategy to integrate the four base learners, and the integrated classifier is used to classify drugs. The accuracy rates of morphine, methadone, methamphetamine, binary mixtures of morphine and methadone, binary mixtures of morphine and methadone, and binary mixtures of methadone and methamphetamine are 91%, 85%, 82%, 91%, 75% and 63% respectively. Figure 5-9 , the soft voting classifier ( Figure 9 ) outperforms any single classification model. Compared to the best performing random forest model ( Figure 6 ), the classification accuracy of morphine, methadone, and methamphetamine increased by 21%, 6%, and 24%, respectively. The classification accuracy of binary mixtures of morphine and methadone and binary mixtures of methadone and methamphetamine increased by 35% and 5%, respectively. This proves that the ensemble method plays a key role in improving the machine learning classification performance in unary and binary drug Raman datasets, and improves the classification accuracy.

[0105] This application proposes a drug classification method based on colloidal AuNPs@AgCl and a soft voting model. This method uses gold nanoparticles to acquire the SERS signals of three monovalent drugs and their binary mixtures, and then uses a soft voting classifier to distinguish the three drugs and their binary mixtures. The proposed soft voting ensemble learning classifier demonstrates higher classification accuracy and can achieve refined identification of complex drug samples to a certain extent.

[0106] To verify the effectiveness and reliability of the soft voting ensemble learning classifier, this application also conducted comparative experiments using four other traditional machine learning algorithms on datasets of identical sample sizes. The experimental results demonstrated that the soft voting ensemble learning classifier significantly outperformed the four traditional machine learning methods in the classification accuracy of various drug categories, further supporting its application in the accurate identification of drugs and their binary mixtures. Therefore, this embodiment has considerable practical application value.

[0107] The above description is merely an embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A drug classification method based on colloidal AuNPs@AgCl and a soft voting model, characterized in that: The following steps are involved: S1. Preparation of colloidal gold nanoparticles (AuNPs) based on the seed-based growth method. S2. Preparation of colloidal AuNPs@AgCl solution with AgCl activation layer; S3. Based on the colloidal AuNPs@AgCl as a SERS substrate, Raman spectral data of mono- and di-drugs were obtained; S4. The Raman spectral data are normalized and feature-screened in sequence to obtain a feature data set; S5. Based on extreme random trees, random forests, support vector machines, and XGBoost as base classifiers, a weight is assigned to each base classifier through the soft voting method to obtain an integrated classifier and classify monoid and binary drugs.

2. The method according to claim 1, characterized in that The method of step S1 includes: S11. At room temperature, trisodium citrate (C6H5Na3O7) was mixed with ultrapure water in a mass-to-volume ratio of 100 mg:10 mL to obtain a trisodium citrate solution and placed in an ultrasonic bath for 15-20 minutes; S12. In an ice bath, mix HAuCl4 with ultrapure water at a mass volume ratio of 100 mg:50 mL to obtain a HAuCl4 solution and sonicate for 15-20 minutes. S13. The trisodium citrate solution obtained in step S1.1 and step S1.2 was mixed with chloroauric acid solution in a volume ratio of 1:25 and then placed in a preheated oil bath and boiled; S14. After the solution has boiled for 2-5 minutes, an appropriate amount of low-concentration chloroauric acid solution (mass-to-volume ratio of chloroauric acid to ultrapure water: 33.979 mg:10 mL) is rapidly added to the solution under magnetic stirring (600 rpm). Stirring is continued for 15-20 minutes, followed by natural cooling to room temperature to obtain a gold seed solution. S15. Mix ultrapure water and gold seed solution in a volume ratio of 4:1 and heat to boiling, then add the trisodium citrate solution obtained in step S1.

2. After 3 minutes, slowly add an appropriate amount of low-concentration chloroauric acid solution. Continue stirring for 15-20 minutes and then naturally cool to room temperature to obtain the gold nanoparticles AuNPs.

3. The method according to claim 2, characterized in that The method of step S2 includes: S21. The gold nanoparticles AuNPs are centrifuged at 5000 rpm for 15-20 min, the supernatant is removed, and the mixture is redissolved in deionized water and then ultrasonically dispersed in an ultrasonic tank to obtain centrifuged colloidal AuNPs; wherein the volume ratio of the centrifuged colloidal AuNPs to the original solution is 1:5; S22. Adding sodium chloride solution and silver nitrate solution to the centrifuged colloidal AuNPs successively, keeping the colloidal AuNPs fully stirred during the addition process, and obtaining the colloidal AuNPs@AgCl solution after the reaction is completed.

4. The method according to claim 3, characterized in that The volume ratio of the centrifuged colloidal AuNPs, the sodium chloride solution, and the silver nitrate solution is 50:5:1, the concentration of the sodium chloride solution is 1.5M / L, the concentration of the silver nitrate solution is 10mM / L, and the stirring reaction time is 5-10min.

5. The method according to claim 3, characterized in that The method of step S3 includes: S31. Dilute the drug solution to 100 ppm, thoroughly mix the drug and the colloidal AuNPs@AgCl at a volume ratio of 1:9, and transfer the mixture into multiple quartz tubes to obtain multiple drug samples; wherein the drug solution should include a single drug solution and a mixed solution of two drugs; S32. Place the quartz tube containing the drug sample directly in the Raman spectrometer, adjust the laser position so that the laser hits the center of the quartz tube, set the Raman spectrometer integration time to 10s and the laser power to 15mW, repeat the measurement multiple times for each sample, and obtain Raman data including Raman signal intensity.

6. The method according to claim 5, characterized in that The method of step S4 includes: S41. Based on the Raman data, in the range of 400-2000 cm -1 Find the peak with the largest Raman signal intensity and the lowest value within the beam range; S42. Based on the maximum-minimum normalization method and formula (1), the Raman signal intensity I is normalized to between 0 and 1; wherein, the formula (1) is: I * =(I-I min ) / (I max -I min ) (1); Among them, I * Represents the normalized Raman signal intensity, I max Indicates 400-2000cm -1 The maximum Raman signal intensity within the beam range, I min Indicates 400-2000cm -1 The lowest value of the Raman signal within the beam range; S43. Name the normalized Raman signals x1, x2, ..., x 1601 , obtain the normalized first data set; S44. Based on the t-SNE algorithm, the first data set is subjected to dimensionality reduction processing to obtain a second data set of drugs after dimensionality reduction; wherein, the number of new feature vectors after dimensionality reduction is set to 4, the learning rate is set to 100, and the maximum number of iterations is set to 300 times.

7. The method according to claim 6, characterized in that The method of step S5 includes: S51. The second data set is divided into a training set and a test set in a ratio of 7:3, and four machine learning classifiers, namely, extreme random trees, random forests, support vector machines, and XGBoost, are used for training to obtain the accuracy of drug classification; S52. Construct a soft voting classification model, which integrates four base classifiers: extreme random trees, random forests, support vector machines, and XGBoost; S53. Based on the soft voting classification model, the drugs are classified. The final prediction result should be The largest category is: in, Represents the final prediction result, is the maximum index function.

8. The method according to claim 7, characterized in that The method for constructing the soft voting classification model in step S52 includes: S521. Based on the accuracy of each base classifier on the test set, assign a weight to each base classifier; wherein the weight of the base classifier is: Among them, w e is the corresponding weight of the e-th classifier; e = 1, 2, 3, 4 represent four machine learning classifiers: extreme random tree, random forest, support vector machine and XGBoost respectively; AUC e is the AUC value of the e-th base classifier; the denominator Represents the sum of the AUCs of the four base classifiers, which is used to normalize the weights to ensure that the sum of the weights is 1; S522. Based on the weight of each base classifier, the final prediction result is weighted averaged; wherein, the accuracy of the soft voting prediction is: in, Indicates that the integrated model is that the drug sample belongs to category C g The final weighted probability of g =1,2,3,…6, corresponding to the six drugs: morphine, methadone, methamphetamine, a binary mixture of morphine and methadone, a binary mixture of morphine and methadone, and a binary mixture of methadone and methamphetamine; w e is the corresponding weight of the e-th classifier; Indicates that the e-th base classifier predicts that the drug belongs to category C g The accuracy rate.

Images