Machine learning enhanced silver-based covalent organic framework nanozyme colorimetric detection method and application thereof

Abstract

The application discloses a machine learning enhanced silver-based covalent organic framework nanoscale enzyme colorimetric detection method and application thereof, and constructs a covalent organic framework loaded with silver nanoparticles to form a nanoscale enzyme with oxidase-like activity, which is used for rapid visual detection of mercury ions. When mercury ions exist, silver ions in the COF-AgNPs can form silver amalgam with the mercury ions, thereby catalyzing the colorless substrate 3,3',5,5'-tetramethylbenzidine to generate a blue oxidation product oxTMB. The content of mercury can be judged by detecting the absorbance and RGB value changes of the color development reaction. Meanwhile, a machine learning data analysis strategy of a support vector machine model is adopted to perform five-fold stratified cross-validation analysis on the obtained absorbance and RGB values, so that the detection range is expanded and the detection sensitivity is improved.

Description

TECHNICAL FIELD

[0001] The application relates to a machine learning enhanced silver-based covalent organic framework nanozyme colorimetric detection method, which can rapidly detect mercury ions (Hg 2+ ) in food and environmental samples, and belongs to the research fields of analytical chemistry, food safety and detection, and environmental analysis and monitoring. BACKGROUND

[0002] Mercury ions (Hg 2+ ) as a kind of toxic heavy metal pollutants, due to its high bioaccumulation and environmental persistence, has become a major threat to global aquatic environment and food safety. With the intensification of industrial wastewater discharge, mineral exploitation and agricultural activities, mercury ions enter aquatic products (such as fish, shellfish, shrimp, etc.) through surface runoff and biological enrichment, and finally cause irreversible damage to human nervous system, kidney and immune system. The World Health Organization (WHO) and the Minamata Convention have listed mercury pollution as a priority for control, and strictly limited the maximum residue limit of mercury in aquatic products, and it is urgent to develop efficient and accurate detection technology to protect food safety and public health. Traditional detection techniques for Hg 2+ include atomic fluorescence spectrometry (AFS) and inductively coupled plasma mass spectrometry (ICP-MS), which have high sensitivity but rely on large instruments, complex sample pretreatment and professional operation, and are difficult to meet the needs of on-site rapid screening. In addition, the complexity of aquatic product matrix can easily lead to signal deviation, and multiple purification steps are required to eliminate interference, which not only consumes time but also may introduce secondary pollution.

[0003] In recent years, nanozymes (Nanozymes) have shown great application potential in food safety, disease diagnosis and treatment, environmental monitoring and biological catalysis due to their convenient synthesis, low cost, excellent stability and easy storage and transportation. In particular, noble metal nanozymes, due to their high surface functionalization ability, excellent biocompatibility and biomimetic catalytic performance, have become the research focus in the field of colorimetric detection. Especially silver-based nanozymes, due to their lower cost and easier preparation, have attracted more and more attention in the field of catalysis. For example, Sun team embedded small size silver nanoparticles (AgNPs) into temperature-responsive gelatin matrix to construct a colorimetric sensor with specific catalytic activation of mercury ions (Hg 2+ ), which can realize the detection of Hg 2+High-throughput detection; Yin et al. innovatively utilized a carbon nitride-doped melamine-silver system to endow the material with dual-functional properties of peroxidase-like activity and visible light photocatalysis, simultaneously achieving total mercury detection and photocatalytic detoxification. However, these nanozymes inevitably suffer from poor catalytic stability, hindering their wider application. Therefore, to address the aforementioned catalytic stability issues, many scientists have attempted to prepare nanozymes with high catalytic stability by combining various organic compounds with noble metal nanoparticles (NPs) for the detection of mercury ions (Hg). 2+ The analysis and detection of Hg. In these studies, covalent organic frameworks (COFs) showed significant advantages due to their unique structural characteristics. Covalent organic frameworks (COFs), as a class of novel crystalline porous polymers formed by strong covalent bonds connecting elements such as carbon (C), nitrogen (N), and oxygen (O), have become ideal supports for loading noble metal nanoparticles due to their high specific surface area, tunable pore size, and excellent stability. The strong interaction between their nitrogen-containing groups and nanoparticles not only enhances the stability of the material but also controls the size of the nanoparticles through the pore confinement effect, thereby enhancing catalytic activity and sensitivity. For example, Wang's team developed a method for detecting Hg based on the Cu2O@Cu2S core-shell D-TA-COF heterostructure. 2+ A biosensor combining high sensitivity and dual-modal detection capabilities has been developed to detect Hg in complex environments (such as river water and serum). 2+ Precise analysis; Li's team developed a method for detecting trace amounts of Hg based on two-dimensional covalent organic framework nanosheets (COFPTAzo) loaded with gold nanoparticles (AuNPs). 2+ The colorimetric platform, which combines ultra-sensitive response and excellent selectivity, fully demonstrates the potential of functional nanocomposites to enhance sensing performance and environmental adaptability through organic-inorganic synergistic design.

[0004] To further break through the mercury ion (Hg) 2+) detection, the linear response range of a single colorimetric sensor is narrow, and it is difficult to balance the detection of low and high concentrations. Machine learning provides an innovative solution for nanoscale enzyme sensing technology. As a powerful data analysis tool, machine learning is based on statistical, probabilistic and computational theory to build a data-driven algorithm system. It has significant advantages in processing complex data, classification and prediction. Through supervised learning (such as support vector machine, random forest), unsupervised learning (such as principal component analysis), etc., it can independently mine potential laws from multi-dimensional data and build prediction models. In the field of sensing technology and nanomaterials, machine learning technology effectively breaks through the response limitations and selectivity bottlenecks of traditional single sensors by integrating multi-source data such as ultraviolet spectra and RGB images. Taking the Prussian blue nanoscale enzyme multi-modal biosensor as an example, this platform innovatively integrates colorimetric and photothermal dual signal detection mechanisms, combines convolutional neural network (CNN) algorithm and smartphone terminal, not only realizes high-sensitivity portable detection of biomarkers, but also expands the detection dynamic range to continuous coverage from trace (pM level) to high concentration (μM level) through the unique non-linear modeling advantage of machine learning. This 'nanoscale enzyme + machine learning' collaborative strategy not only overcomes the interference problem in complex matrix of traditional methods, but also improves the prediction accuracy across concentration intervals through algorithm optimization, providing an intelligent detection new paradigm with high precision and practicality for environmental monitoring and clinical diagnosis. Therefore, according to the progress of the related research of predecessors, combining the advantages of COF-AgNPs nanoscale enzyme and machine learning, a kind of colorimetric sensor based on nanoscale enzyme for Hg 2+ detection is constructed. SUMMARY

[0005] In view of the technical problems existing in the detection of mercury ions (Hg 2+ ) in aquatic products in the background art, a machine learning enhanced nanoscale enzyme colorimetric sensor for rapid detection of mercury ions (Hg 2+ ) in aquatic products is proposed. Compared with traditional mercury ion (Hg 2+ ) detection methods such as atomic fluorescence spectrometry, cold atomic absorption spectrometry and inductively coupled plasma mass spectrometry, the method uses covalent organic framework (COF) loaded silver nanoparticles (AgNPs) as nanoscale enzyme (COF-AgNPs) probe to detect mercury ions (Hg 2+ ) in different aquatic products. Based on the strong selectivity of AgNPs to mercury ions (Hg 2+ ), the nanoscale enzyme colorimetric technology solves the problem of mercury ions (Hg 2+) Selectivity problems, and the formation of amalgam, so that the sensitivity and detection limit has certain advantages relative to traditional methods; then the system compares the classification performance of five machine learning models of random forest (RF), support vector machine (SVM), K nearest neighbor (KNN), logistic regression (LR) and linear discriminant analysis (LDA), and the classification ability under the single mode of spectrum, RGB single mode and fusion multi-mode. Not only greatly improve the high sensitivity and anti-interference of nanoscale enzyme colorimetric technology in different aquatic products for mercury ion (Hg 2+ ) Detection, but also realize the rapid detection of mercury ion (Hg 2+ ) In a wide concentration range.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A machine learning enhanced silver-based covalent organic framework nanoscale enzyme colorimetric detection method, the method is as follows:

[0008] S1: Mix COF-AgNPs composite material, acetate buffer, 3, 3', 5, 5' tetramethylbenzidine (TMB) according to the volume ratio of 0.02:10:2 to obtain sample liquid;

[0009] S2: Add 100 μL of mercury ion Hg 2+ Standard solution with concentration range of 0.05-50 μmol / L into the sample liquid obtained in S1 respectively to obtain test liquid;

[0010] S3: The test liquid containing different concentrations of mercury ion Hg 2+ Standard solution obtained in S2 is reacted in a 30℃ environment for 5 minutes;

[0011] S4: Measure the absorbance value of the test liquid obtained in S3 in the range of 400-800 nm, and the difference of the absorbance value of the test liquid at 652 nm before and after adding mercury ion Hg 2+ Standard solution and mercury ion Hg 2+ Concentration to construct a linear model; use a smart phone to synchronously acquire RGB three-channel color values under LED light source, calculate the gray value by the formula Gray = 0.299R + 0.587G + 0.114B, and construct a linear regression model with the gray value and mercury ion Hg 2+ Concentration;

[0012] S5: The absorbance value of the test liquid containing different concentrations of mercury ion Hg 2+ Collected in the range of 400-800 nm is preprocessed by multiple scattering correction MSC to eliminate light scattering interference;

[0013] S6: The data pretreated by S5 is subjected to feature extraction by competitive adaptive reweighted sampling (CARS);

[0014] S7: The feature values extracted in S6 and the RGB three-channel color values extracted from the test liquid are used to construct a spectral single-mode, an RGB single-mode and a fusion double-mode data set;

[0015] S8: The collected spectral single-mode, RGB single-mode and fusion double-mode data sets are used to construct five kinds of classification and recognition models, and different classification and recognition models are evaluated to obtain the best model.

[0016] Further, the synthesis method of the COF-AgNPs composite material is as follows: COFs are dispersed in anhydrous ethanol at a ratio of 1:1 and ultrasonically mixed for 1 h, polyvinylpyrrolidone PVP with Mw≈5800 is added at a mass ratio of 3:1, and 2.5 mL of 0.1 mol / L silver nitrate, sodium citrate and glucose solutions are sequentially added under ultrasonic assistance, and stirring for 15 h enables AgNPs to be successfully loaded on the surface of the COFs. The yellow solid is collected by centrifugation and washed with a pure water and methanol mixture multiple times, and finally vacuum dried to obtain the COF-AgNPs composite material.

[0017] Further, the synthesis method of the COFs is as follows: 1,3,5-tris(4-aminophenyl)benzene and 2,5-divinyl terephthaldehyde DVA are placed in a conical flask at a molar mass ratio of 2:3, 50 mL of acetonitrile containing 2.9 mol / L acetic acid HAc is added, and after ultrasonic treatment, it is placed at room temperature for 72 h, centrifuged and washed multiple times with a mixture of tetrahydrofuran (THF) and anhydrous ethanol (EtOH), and finally vacuum dried to obtain yellow powdery COFs.

[0018] Further, the five kinds of classification and recognition models are: random forest (Random Forest, RF), support vector machine (Support Vector Machine, SVM), K-nearest neighbor (K - Nearest Neighbors, KNN), logistic regression (Logistic Regression, LR) and linear discriminant analysis (Linear Discriminant Analysis, LDA).

[0019] Further, the indicators for model preliminary screening evaluation include accuracy (Accuracy), recall (Recall), precision (Precision) and F1 score;

[0020]

[0021]

[0022]

[0023]

[0024] wherein Accuracy is the accuracy, Recall is the recall, Precision is the precision and F1 is the F1 score. F1 is the score, TP is the true positive, FN is the false negative, FP is the false positive and TN is the true negative.

[0025] Further, the rule of the competitive adaptive reweighted sampling (CARS) is that the Monte Carlo sampling number is 100 times, the cross-validation coefficient is 5, and the feature retention coefficient is 11.

[0026] Further, the support vector machine (SVM) model is the best model, the accuracy is 95.97%, the F1 score is 0.958, the detection range is expanded to a continuous linear interval of 0-50 µmol / L, and the detection limit of the nano-enzyme colorimetric sensor is 0.0107 μmol / L. F1

[0027] A machine learning enhanced silver-based covalent organic framework nano-enzyme colorimetric detection method is applied to the rapid detection of mercury ions (Hg 2+ ) in food and environmental samples.

[0028] The above technical solutions can achieve the following beneficial effects:

[0029] The innovative detection method based on the combination of the covalent organic framework loaded silver nanoparticles (COF-AgNPs) composite nano-enzyme and the multi-modal machine learning proposed by the application successfully realizes the full coverage detection and concentration partition optimization of mercury ions (Hg 2+ ) in a wide linear range of 0-50 µmol / L. The method uses 2,5-divinyl terephthaldehyde and 1,3,5-tris(4-aminophenyl) benzene as monomers, synthesizes a layered flower-shaped covalent organic framework (COFs) in an acetonitrile / glacial acetic acid catalytic system, and constructs a COF-AgNPs nano-enzyme probe by in-situ growth of silver nanoparticles (AgNPs). When there is mercury ion (Hg 2+ ) or mercury-containing compound, Hg 2+ forms an Ag-Hg alloy on the surface of AgNPs through a metal displacement reaction (Ag 0 + Hg 2+ → Ag + + Hg 0 ), and the synergistic effect of the electronic structure of the alloy significantly reduces the activation energy of the TMB oxidation reaction, and the bimetallic synergistic site jointly promotes the generation of active oxygen (singlet oxygen ( 1 ​O2), superoxide anion (O2) •− This study achieves efficient oxidation of TMB to generate blue oxTMB. A dual-mode detection system combining absorbance signal and RGB color values ​​was successfully established. To meet the need for accurate detection over a wide concentration range, the classification performance of five machine learning models—random forest (RF), support vector machine (SVM), k-nearest neighbor (KNN), logistic regression (LR), and linear discriminant analysis (LDA)—was systematically evaluated under spectral single-modality, RGB single-modality, and multimodal fusion conditions. Experimental results show that the multimodal data fusion strategy significantly improves the model's ability to resolve complex concentration gradients by integrating the quantitative characteristics of colorimetry with the responsive advantages of RGB numerical analysis. This strategy not only overcomes the detection range limitations of traditional single-modality methods but also significantly improves data processing efficiency and result accuracy through multi-source signal collaborative analysis, providing an innovative technical solution for high-throughput monitoring of heavy metal pollutants in the food and environmental fields.

[0030] Nanozyme probes based on in-situ growth of AgNPs from layered flower-like COFs utilize a porous structure that provides a high specific surface area and nitrogen-rich functional groups, enhancing the interaction between AgNPs and mercury ions (Hg). 2+ The method demonstrates strong binding ability to RGB image data. Experimental verification shows that the spiked recoveries in actual fish and shrimp samples are 90.18%-103.219%, with relative standard deviations (RSD) of only 0.9%-8.87%, indicating significant resistance to interference from complex matrices. Furthermore, the RGB image data acquisition is compatible with low-cost portable devices (smartphone cameras), overcoming the dependence of traditional instruments such as ICP-MS and AAS on laboratory environments, and providing feasibility for rapid on-site screening.

[0031] By employing a dual-signal acquisition mechanism combining colorimetry (UV absorption spectroscopy) and RGB numerical analysis, and integrating machine learning algorithms such as KNN, SVM, and RF to construct a multimodal fusion model, the detection of mercury ions (Hg) was achieved. 2+ This method offers highly sensitive quantitative detection of trace heavy metals. Its average accuracy reaches 91.98%, with a detection limit as low as 0.01 μmol / L, representing a sensitivity improvement of over 10% compared to traditional spectrophotometry. This breakthrough is attributed to the synergistic effect of spectral characteristics and color channel values, effectively overcoming the limitations of single-modality detection for complex samples. Furthermore, a differentiated model optimization strategy for different concentration ranges (low, medium, and high) achieves a detection accuracy of up to 95.08% in the lowest concentration range (0.01-1 μmol / L), addressing the industry challenge of low-concentration signals being susceptible to noise interference in trace heavy metal detection. Attached Figure Description

[0032] Figure 1 The synthesis roadmap for nanomaterials (1A), mercury ions (Hg) 2+) detection flow chart (1B) and machine learning processing flow chart (1C);

[0033] Figure 2 The concentration of different mercury ions (Hg 2+ ) and the relationship between the characteristic peak intensity at 500-800 nm in the ultraviolet absorption spectrum (2A) and the linear equation graph (2B, 2C) and the linear equation relationship graph (2D) of different mercury ions (Hg 2+ ) concentration and RGB gray value;

[0034] Figure 3 The performance comparison chart of different models constructed by multi-modal and single-modal (3A) and the classification performance comparison chart of each model (3B);

[0035] Figure 4 The confusion matrix (4A) and the receiver operating characteristic curve (4B) generated under the support vector machine condition;

[0036] Figure 5 The SHAP feature contribution chart;

[0037] Figure 6 The interference experiment results. DETAILED DESCRIPTION

[0038] The present application will be further described below in conjunction with the accompanying Figures 1-5 and examples:

[0039] A machine learning enhanced silver-based covalent organic framework nanoscale enzyme colorimetric detection method, the synthesis method of the covalent organic framework is as follows: 1, 3, 5-tri(4-aminophenyl) benzene and 2, 5-divinyl terephthaldehyde (DVA) are placed in a conical flask according to the molar mass ratio of 2:3, 50 mL of acetonitrile containing 2.9 mol / L acetic acid (HAc) is added, and after ultrasonic treatment, it is placed at room temperature for 72 h. Centrifugation and washing with a mixture of tetrahydrofuran (THF) and anhydrous ethanol (EtOH) 3 times. Finally, vacuum drying obtains yellow powder COFs.

[0040] The synthesis method of COF-AgNPs composite material is as follows: the COFs obtained above are dispersed in anhydrous ethanol at a ratio of 1:1 and ultrasonic mixed for 1 h, polyvinylpyrrolidone (PVP) (Mw≈5800) is added at a mass ratio of 3:1, 2.5 mL of 0.1 mol / L silver nitrate, sodium citrate and glucose solution are added in turn under ultrasonic assistance, and stirring for 15 h, so that AgNPs are successfully loaded on the surface of COFs, centrifugal collection yellow solid and washing with pure water and methanol mixture 3 times. Finally, vacuum drying obtains COF-AgNPs composite material.

[0041] The machine learning model construction method in a machine learning-enhanced silver-based covalent organic framework nanozyme colorimetric detection method is as follows:

[0042] S1: Mix COF-AgNPs composite material, acetate buffer, and 3,3',5,5'-tetramethylbenzidine (TMB) at a volume ratio of 0.02:10:2 to obtain the sample solution;

[0043] S2: 100 μL of mercury ions (Hg) with a concentration ranging from 0.05 to 50 μmol / L were added to the sample solution obtained in S1. 2+ Standard solution, to obtain test solution;

[0044] S3: The mercury ions (Hg) obtained in S2 at different concentrations... 2+ The standard solution test solution reacts for 5 minutes at 30°C.

[0045] S4: Measure the absorbance of the test solution obtained in S3 within the range of 400-800 nm, and add mercury ions (Hg). 2+ The difference in absorbance at 652 nm between the standard solution and the test solution before and after is related to the mercury ion concentration (Hg). 2+ A linear model was constructed based on concentration. RGB three-channel color values ​​were simultaneously acquired using a smartphone under an LED light source. The grayscale value was calculated using the formula Gray = 0.299R + 0.587G + 0.114B, and then compared with the grayscale value to the mercury ion (Hg) concentration. 2+ Concentration was used to construct a linear regression model;

[0046] S5: Collected Hg containing different concentrations of mercury ions 2+ The absorbance values ​​of the test solution in the range of 400-800 nm were pretreated by multivariate scattering correction MSC to eliminate light scattering interference;

[0047] S6: The data preprocessed in S5 is used for feature extraction using competitive adaptive reweighted sampling CARS;

[0048] S7: The feature values ​​extracted from S6 and the RGB three-channel color values ​​extracted from the test solution are used to construct spectral single-mode, RGB single-mode and fused dual-mode datasets;

[0049] S8: Five classification and recognition models were constructed from the collected spectral single-mode, RGB single-mode, and fused dual-mode datasets. The models included Random Forest (RF), Support Vector Machine (SVM), K Nearest Neighbor (KNN), Logistic Regression (LR), and Linear Discriminant Analysis (LDA). The different classification and recognition models were evaluated to obtain the best model.

[0050] Example 1: Treatment of mercury ions (Hg) 2+ ) Detection and analysis results

[0051] Mercury ion (Hg 2+ ) standard solution configuration: according to the national standard GB / T602-2002 "preparation of standard solution for impurity determination of chemical reagents", the standard solution of mercury ion (Hg 2+ ) is prepared. 0.162 g of mercury nitrate is weighed and dissolved in 10 mL of nitric acid solution, transferred into a 1000 mL volumetric flask, and diluted to the calibration line with deionized water.

[0052] Detection process: mix different concentrations of mercury ion (Hg 2+ ) standard solution, acetate buffer, color developing substrate 3,3',5,5'tetramethylbenzidine according to the volume ratio of 1:10:2, and then add COF-AgNPs solution to obtain the test sample solution; after 5 minutes of reaction at 30℃, the test sample solution is detected for absorbance in the range of 400-800 nm and the synchronous RGB value of the smartphone (iPhone Xs) + LED light source (standard D65(5500 k)).

[0053] Constructing linear model: the relationship curve between different Hg 2+ concentrations and the characteristic peak intensity at 500-800 nm in the ultraviolet absorption spectrum is shown in Figure 2 A; through linear fitting, ΔA652 and c(Hg 2+ ) have a good linear relationship (R 2 >0.99) in the range of 0.05-0.5 μmol / L and 5-50 μmol / L, as shown in Figure 2 B, the linear regression equation in the range of 0.05-0.5 μmol / L is Y=0.4398 c(Hg 2+ )-0.03468, (R 2 =0.990). According to the 3σ criterion, the detection limit (LOD) calculated is 0.0107 μmol / L. As shown in Figure 2 C, the linear regression equation in the range of 5-50 μmol / L is Y=0.0217 c(Hg 2+ )+0.2858, (R 2 =0.9937), and the detection limit is 0.207 μmol / L. The RGB data acquisition needs to be carried out under the standard D65(5500 k) light source, and the gray value (Gray) is extracted through image processing to construct the linear model with different mercury ion (Hg 2+ ) concentrations, as shown in Figure 2 D, the linear fitting uses the gray value formula: Gray=0.2999R+0.587G+0.114B, the detection range is 0-50 µmol / L, through linear fitting, the gray value and c(Hg2+ ) showed a good linear relationship (R 2 > 0.99). The equation was Y = -3.29323 c(Hg 2+ )+153.1088, (R 2 =0.995). According to the 3σ criterion, the limit of detection (LOD) was calculated to be about 1.32 µmol / L. In summary, the detection limit of the nanosensor was 0.0107 μmol / L (about 0.002 mg / L), which was much lower than the Chinese national standard (GB 2762-2022) and the European Union's water product mercury residue limit, and the sensor had a wider linear range and lower detection limit compared with other sensors (as shown in Table 2).

[0054] Table 1 Detection standards of mercury in aquatic products

[0055]

[0056] Table 2 Comparison of other methods for detecting mercury ions

[0057]

[0058] The collected ultraviolet absorption spectrum data were corrected for baseline drift and scattering interference by multiple scattering correction (MSC), and the spectral features were optimized and screened by competitive adaptive reweighted sampling (CARS) algorithm (Monte Carlo iteration 100 times, 5-fold cross-validation, retaining 11 key wavelengths). The R, G, and B channel means of the color reaction solution were extracted to construct the visual feature set. Then the spectral features (10 dimensions), color features (3 dimensions), and their fusion features (13 dimensions) were respectively input into random forest (RF), support vector machine (SVM), K nearest neighbor (KNN), logistic regression (LR), and linear discriminant analysis (LDA) models for training. The accuracy, recall rate, precision, and F1 F1 score were evaluated. The classification performance of different models Figure 3 A) and different modalities Figure 3 B) was compared comprehensively. It was concluded that the spectral-color fusion dataset combined with the support vector machine (SVM) model showed the best performance, with an accuracy of 95.97%, F1 a value of 0.958, which was more than 10% higher than the performance of the single-spectrum model. We selected the support vector machine (SVM) model to generate the confusion matrix, receiver operating characteristic (ROC) curve, and area under the curve (AUC) to further analyze the performance of the sensor. As shown in Figure 4 A, the confusion matrix showed that the biosensor performed better at low and high concentrations, with an overall accuracy of 93.58%, but there was still room for improvement at medium concentrations. In the ROC curve Figure 4In B), the AUC values ​​corresponding to low concentration (0.99), medium concentration (0.96), and high concentration (0.97) were all at relatively high levels, demonstrating that the biosensor can effectively distinguish Hg-containing molecules. 2+ With Hg-free 2+ The samples were analyzed. Among them, the biosensor performed best at high concentration (black curve), followed by low concentration (yellow curve), and performed slightly weaker at medium concentration (green curve), but its overall performance was still good.

[0059] SHAP interpretability analysis Figure 5 As shown in Table 3, the synergistic effect of absorbance and RGB channel values ​​in dominating the prediction results (contribution 62% : 38%) provides a quantitative basis for the multimodal sensing data fusion mechanism. Furthermore, Table 3 shows that by calibrating the signal using the algorithm, the detection range is extended to a continuous linear range of 0–50 µmol / L, covering the entire concentration range of traditional methods and eliminating blind spots in the intermediate range, significantly improving the dynamic range and applicability of the detection.

[0060] Table 3 Comparison of methods

[0061]

[0062] High selectivity for mercury ions (Hg) in actual samples 2+ The detection of mercury ions (Hg) is crucial and is an important indicator of whether a method can be applied in actual sample testing. Because seafood still contains some metal ions after digestion, it may affect the detection of mercury ions (Hg). 2+ The selectivity of ions (Al) has an impact, therefore, the selection of common metal ions and anions (Al) in seafood is important. 3+ Pb 2+ Mg 2 + Cd 2+ Ca 2+ Mn 2+ SO4 2- CO3 2- Cl - PO4 3- K + Zn 2+ Na + Ni 4+ Cu 2+ The concentration of metal ions is mercury ions (Hg). 2 + The concentration of mercury ions (Hg) is 10 times that of mercury ions (Hg). 2+ Selectivity tests were conducted at 10 times the concentration of mercury ions (Hg). 2+) under the optimal conditions of detection, different metal ions were added into the same system, and the absorbance value at 652 nm was measured by using a microplate reader. The absorbance value of the blank control was subtracted from the absorbance value of the sample to obtain the final absorbance value. Figure 6 It can be seen that, except for mercury ions (Hg 2+ ), other ions have little response to the colorimetric detection method, indicating that the method has high specificity for the detection of mercury ions (Hg 2+ ) and is not interfered by other ions.

[0063] To prove the feasibility of the method in analytical applications, five kinds of marine products, shrimp, squid, yellow catfish, flower shell and kelp, were selected as representative samples to evaluate the analytical performance of the proposed method. The marine product samples need to be digested first. The specific steps are as follows: weigh 5 kinds of marine products (0.5 g) and put them into conical flasks, respectively, add 15 mL of HNO3 and 2.5 mL of H2SO4, and digest the resulting mixed solution at room temperature overnight. Then, heat the mixture in the five conical flasks to boiling until a large amount of white smoke is generated and the solution becomes clear. After cooling to room temperature, adjust the pH of the resulting clear solution to 5.0 with 1 mol / L NaOH solution, and dilute to 100 mL with deionized water. Under the optimal detection conditions, add different concentrations of Hg 2+ standard solution to the different marine product sample solutions for the standard addition recovery test. The addition amount of mercury ions (Hg 2+ ) standard solution is 100 μL, the absorbance value at 652 nm is measured by using a microplate reader, and the RGB value of the system is obtained by using a smartphone. The standard curve is established according to the established standard curve to calculate the standard addition recovery rate. All tests are repeated three times.

[0064] Table 4 actual sample detection table

[0065]

[0066] All experimental conditions are consistent with the aqueous solution conditions. As can be seen from Table 4, the standard addition recovery rate is between 90.18% and 103.219%, and the relative standard deviation (RSD) is between 0.9% and 8.87%. The above results show that the colorimetric detection method of mercury ions (Hg 2+ ) enhancing the COF-AgNPs oxidase activity is feasible and reliable, and the accuracy and precision are satisfactory, which has broad application prospects.

[0067] The above are preferred embodiments of the present application. For those skilled in the art, various equivalent forms of the present application can be modified without departing from the principles of the present application, and all modifications within the scope of the appended claims of the present application are within the scope of the present application.

Claims

1. A machine learning enhanced silver-based covalent organic framework nanzyme colorimetric detection method, characterized in that, The synthesis method of the covalent organic framework is as follows: first, 1, 3, 5-tris (4-aminophenyl) benzene and 2, 5-divinyl benzene formaldehyde (DVA) are added to an acetonitrile solution containing 2.9 mol / L acetic acid (HAc) in a 2:3 molar ratio, ultrasonic dispersion is performed, and then the reaction is allowed to stand for 72 hours; the yellow product is obtained by centrifugation, washed several times with tetrahydrofuran and anhydrous ethanol, and vacuum dried to obtain the COFs; then the obtained COFs are dispersed with anhydrous ethanol in a 1:1 ratio, 3 times the mass of polyvinylpyrrolidone PVP corresponding to the COFs is added as a stabilizer, 2.5 mL of 0.1 mol / L silver nitrate, sodium citrate and glucose solution are sequentially injected, ultrasonic-assisted stirring is performed for 15 hours to load silver nanoparticles AgNPs on the surface of the COFs, and finally the yellow composite is collected by centrifugation, washed with pure water and methanol, and vacuum dried to obtain the COF-AgNPs composite material; The method is as follows: S1: the COF-AgNPs composite material, acetate buffer, and 3, 3', 5, 5'-tetramethylbenzidine (TMB) are mixed in a volume ratio of 0.02:10:2 to obtain a sample solution; S2: In the sample solution obtained in S1, 100 μL of mercury ion Hg with a concentration range of 0.05-50 μmol / L was added respectively 2+ Standard solution, to obtain the test solution; S3: The solution containing different concentrations of mercury ions Hg2+ obtained in S2 was added into the solution of S1 to form a mixed solution. 2+ Standard solution test liquid reaction time 5 minutes in 30 °C environment; S4: The test solution obtained in S3 is subjected to absorbance value measurement in the range of 400-800 nm, and mercury ion Hg is added 2+ The difference between the absorbance of the test solution before and after the standard solution at 652 nm and the mercury ion Hg 2+ Concentration builds a linear model; use a smart phone to synchronously acquire RGB three-channel color values under an LED light source, calculate the gray value through the formula Gray = 0.299R + 0.587G + 0.114B, and take the gray value and mercury ion Hg 2+ Concentration builds a linear regression model; S5: The collected test solution 400-800 nm range containing different concentrations of mercury ions Hg 2+ The absorbance values of the test solution were pretreated by multiple scattering correction MSC to eliminate light scattering interference. S6: the data pretreated in S5 are subjected to competitive adaptive reweighted sampling (CARS) for feature extraction; S7: the feature values extracted in S6 and the RGB three-channel color values extracted from the test solution are used to construct spectral single-mode, RGB single-mode, and fused double-mode data sets; S8: the collected spectral single-mode, RGB single-mode, and fused double-mode data sets are used to construct five classification recognition models, and different classification recognition models are evaluated to obtain the best model.

2. The machine learning enhanced silver-based covalent organic framework nanozyme colorimetric detection method according to claim 1, characterized in that: The five classification recognition models are: random forest (RF), support vector machine (SVM), K-nearest neighbor (KNN), logistic regression (LR), and linear discriminant analysis (LDA).

3. The machine learning enhanced silver-based covalent organic framework nanozyme colorimetric detection method according to claim 1, characterized in that: The indicators of the model screening evaluation include accuracy, recall, precision, and F1 scores; ； ； ； ； wherein Accuracy is the accuracy, Recall is the recall, Precision is the precision, and F1 are scores, TP is true positive, FN is false negative, FP is false positive, and TN is true negative.

4. The machine learning enhanced silver-based covalent organic framework nanozyme colorimetric detection method according to claim 1, characterized in that: The rules of competitive adaptive reweighted sampling (CARS) are: 100 times of Monte Carlo sampling, a cross-validation coefficient of 5, and a feature retention coefficient of 11.

5. The machine learning enhanced silver-based covalent organic framework nanoszyme colorimetric detection method according to claim 1 or 3, characterized in that: Support vector machine (SVM) model is the best model, with an accuracy of 95.97%, F1 The value is 0.958, the detection range is expanded to a continuous linear interval of 0-50 μmol / L, and the detection limit of the nanoscale enzyme colorimetric sensor is 0.0107 μmol / L.

6. A machine learning enhanced silver-based covalent organic framework nanozyme colorimetric detection method for mercury ions Hg2+ in food and environmental samples 2+ for rapid detection applications.

Images