A machine learning-enhanced fluorescence detection method for split nucleic acid aptamers silver nanoclusters

Abstract

This invention discloses a machine learning-enhanced fluorescence detection method for silver nanoclusters containing split nucleic acid aptamers, used for the detection of adenosine triphosphate (ATP) in aquatic products, belonging to the field of analytical chemistry. This detection method uses two split nucleic acid aptamers of ATP as templates to synthesize DNA-AgNCs. In the presence of the target ATP, a conformational change is induced, resulting in a decrease in fluorescence intensity. Quantitative analysis of ATP is achieved by detecting this fluorescence change. To further improve the detection effect, machine learning is introduced to preprocess the fluorescence spectrum. A recursive feature elimination algorithm is used to screen characteristic wavelengths, and a multi-dimensional feature set is constructed using maximum fluorescence intensity, peak area, and full width at half maximum (FWHM). The optimal classification model is obtained through cross-validation optimization. This invention combines aptamer recognition, DNA-AgNCs, and machine learning to construct a detection method where fluorescence dynamically decreases with increasing ATP concentration. It has advantages such as low cost, simple operation, and good biocompatibility, effectively improving detection sensitivity and stability, and has promising application prospects in the qualitative and quantitative analysis of ATP in aquatic products.

Description

Technical Field

[0001] This invention relates to a machine learning-enhanced fluorescence detection method for split nucleic acid aptamers silver nanoclusters, which can rapidly quantify ATP in biological, food, and environmental samples, and belongs to the research fields of analytical chemistry, food safety and detection, and environmental analysis and monitoring. Background Technology

[0002] Adenosine triphosphate (ATP) is the most fundamental energy carrier and metabolic intermediate in all known living organisms, often referred to as the "energy currency of life." It plays a central role in energy conversion, substance synthesis, signal transduction, and the maintenance of cellular homeostasis, forming the basis of metabolism and growth in organisms. Within organisms, ATP hydrolysis releases energy, directly driving almost all life activities, including muscle contraction, substance synthesis, and nerve conduction. This is why it is called the "energy currency." After aquatic products such as fish and shrimp die, cellular respiration ceases, and ATP begins to degrade. The degradation process typically follows this order: ATP → ADP → AMP → inosine acid → inosine → hypoxanthine. By detecting the content of ATP and its metabolites, the freshness of aquatic products can be accurately determined. Besides providing energy, ATP also acts as an extracellular signaling molecule (such as in purine energy signal transduction), participating in physiological processes such as pain perception and inflammatory responses, while simultaneously regulating the activity of various metabolic enzymes within cells. Therefore, achieving rapid and sensitive detection of ATP is of great significance for evaluating the freshness of aquatic products and ensuring food safety.

[0003] Currently, routine methods for ATP detection mainly include high-performance liquid chromatography (HPLC), liquid chromatography-mass spectrometry (LC-MS), and luciferase-based bioluminescence methods. Although these methods have high accuracy, they generally suffer from problems such as expensive equipment, complex operation, long detection time, and high technical requirements for operators, making it difficult to meet the needs of rapid on-site detection.

[0004] In recent years, nucleic acid aptamers, as a class of single-stranded oligonucleotides obtained through in vitro screening, have been widely used in the field of biosensing due to their advantages such as simple synthesis, good chemical stability, and broad target range. In particular, recognition strategies based on split aptamers, which involve splitting the aptamer into two or more segments and then specifically assembling them into ternary complexes in the presence of a target, can effectively reduce background signals. Meanwhile, silver nanoclusters (DNA-AgNCs) synthesized using DNA as a template, as a novel type of fluorescent nanoprobe, have advantages such as tunable fluorescence properties, good biocompatibility, and simple synthesis, and are increasingly being used to construct label-free fluorescent sensors.

[0005] However, existing fluorescence sensing methods based on DNA-AgNCs typically rely on only a single fluorescence intensity signal for analysis. In actual aquatic product sample testing, the complex matrix composition easily generates background fluorescence interference, limiting detection sensitivity and stability. Furthermore, traditional fluorescence analysis methods utilize only a single characteristic parameter (such as maximum fluorescence intensity) for quantification, making it difficult to fully extract effective information from the fluorescence spectrum, resulting in a limited signal variation range and requiring further improvement in detection performance.

[0006] To address the aforementioned issues, this invention combines split nucleic acid aptamer recognition technology, DNA-silver nanocluster fluorescent probes, and machine learning methods. Through multi-dimensional feature extraction and model optimization, a fluorescence detection method for split nucleic acid aptamers and silver nanoclusters is constructed. Summary of the Invention

[0007] This invention aims to provide a machine learning-enhanced fluorescence detection method for split nucleic acid aptamers silver nanoclusters, which can be used for rapid, sensitive, and low-cost detection of ATP in aquatic products, enabling accurate evaluation of the freshness of aquatic products and providing effective technical support for the preservation and quality control of aquatic products.

[0008] To achieve the above objectives, the technical solution adopted in this experiment is as follows:

[0009] A machine learning-enhanced fluorescence detection method for split nucleic acid aptamers silver nanoclusters is described below:

[0010] S1: Using the splitting nucleic acid aptamers Apt1 and Apt2 as recognition elements, DNA silver nanoclusters were prepared. The sequences of the splitting nucleic acid aptamers Apt1 and Apt2 are shown in SEQ ID NO:1 and SEQ ID NO:2, respectively. Specifically, the sequences SEQ ID NO:1 and SEQ ID NO:2 are: Apt1: 5'-ACCTGGGGGAGTAT-3'; Apt2: 5'-TGCGGAGGAAGGT-3';

[0011] S2: The DNA silver nanoclusters signal probes synthesized using split nucleic acid aptamers as templates are Apt1-AgNCs and Apt2-AgNCs, respectively.

[0012] S3: Mix Apt1 and Apt2 with buffer solution respectively, and after denaturation and annealing, add a certain amount of silver nitrate AgNO3 and sodium borohydride NaBH4 and mix well. Wrap the centrifuge tube with tin foil to avoid light and react to obtain Apt1-AgNCs and Apt2-AgNCs.

[0013] S4: Mix the Apt1-AgNCs and Apt2-AgNCs solutions obtained in step S3, add them to the sample to be tested, so that the split nucleic acid aptamers specifically bind to the target ATP, inducing a conformational change in the DNA template, resulting in a change in fluorescence intensity; then perform fluorescence spectral acquisition and record the fluorescence intensity values ​​at each wavelength;

[0014] S5: Preprocess the fluorescence spectrum obtained in step S4, extract spectral and statistical features, construct a target concentration classification model using machine learning algorithms, and achieve quantitative analysis of ATP content by detecting fluorescence spectral features.

[0015] Furthermore, the initial concentration of Apt1 is 1 μmol / L, the initial concentration of Apt2 is 1 μmol / L, the temperature for incubation after adding ATP is 37°C, and the incubation time is 60 min.

[0016] Further, in step S3, the buffer solution is Tris-HCl buffer, 50 mmol / L Tris, 100 mmol / L NaCl and 10 mmol / L MgCl2, pH 7.0. The denaturation conditions are heating at 95℃ for 5 min, and the annealing treatment is gradually cooling followed by incubation at 37℃ for 60 min. When synthesizing Apt1-AgNCs, the concentration ratio of Apt1, AgNO3, and NaBH4 is controlled at 1:18:18. After mixing Apt1 with AgNO3 solution, the reaction is carried out at 4℃ in the dark for 20 min. Then, freshly prepared NaBH4 solution is added, shaken thoroughly, and reacted at 4℃ in the dark for 90 min to obtain the Apt1-AgNCs solution. When synthesizing Apt2-AgNCs, the concentration ratio of Apt2, AgNO3, and NaBH4 is controlled at 1:12:12. After mixing Apt2 with AgNO3 solution, the reaction is carried out at 4℃ in the dark for 20 min. After min, add freshly prepared NaBH4 solution and shake thoroughly. React at 4℃ in the dark for 90 min to obtain Apt2-AgNCs solution.

[0017] Furthermore, after fluorescence spectral preprocessing in step S5, the extracted statistical features include maximum fluorescence intensity, peak area, and full width at half maximum (FWHM). Feature extraction employs a recursive feature elimination algorithm, using random forest as the base classifier and combining it with 3-fold hierarchical cross-validation to select the feature wavelengths most relevant to ATP concentration from the full spectrum wavelength variables, thus forming a spectral feature set.

[0018] Furthermore, the machine learning algorithms include one or more of the following: Logistic Regression (LR), Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and Gradient Boosting Tree (GBDT); spectral features, statistical features, and their fusion features are used as the input set; the classification model is evaluated using accuracy, precision, recall, and macro-average F1 score; and the optimal model is determined through multiple random partitions and cross-validations.

[0019] Furthermore, the K-nearest neighbor model was determined to be the optimal classification model. The optimized model parameters were: n_neighbors=9, Euclidean distance as the distance metric, and uniform weighting. Under this parameter combination, the model showed the best classification performance for low, medium, and high concentrations of ATP, with the area under the ROC curve (AUC) reaching 0.9938, 0.8681, and 0.9318, respectively.

[0020] The above technical solution can achieve the following beneficial effects:

[0021] This invention presents a machine learning-enhanced fluorescence detection method using split nucleic acid aptamers and silver nanoclusters for the quantitative detection of ATP in aquatic products. The split aptamer described in this invention splits a 27-base ATP single-stranded aptamer into two fragments. These two separate fragments specifically bind to ATP, forming a folded related complex. Using the split aptamer as a recognition element, DNA-AgNCs are formed under the action of silver nitrate and sodium borohydride, respectively. Upon binding to the target ATP, fluorescence attenuation occurs, and the ATP content can be quantitatively analyzed by detecting changes in fluorescence intensity. This method overcomes the shortcomings of large background variations and limited signal changes, and has advantages such as low cost, simple operation, low toxicity, and good biocompatibility. Using this method to detect the ATP content in clam samples, the limit of detection reaches 33 µmol / L, demonstrating good sensitivity. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the detection principle;

[0023] Figure 2 This is a diagram showing the specific results of splitting nucleic acid aptamers;

[0024] Figure 3 This is a standard curve for ATP testing;

[0025] Figure 4 This is a fluorescence linear response diagram;

[0026] Figure 5 A graph showing the test set accuracy of different models on three types of feature sets;

[0027] Figure 6A comparison chart of F1 scores for different models on three feature sets;

[0028] Figure 7 Repeatedly partition the average confusion matrix diagram for the KNN model;

[0029] Figure 8 The ROC curve of the optimized KNN model;

[0030] Figure 9 This is a diagram showing the actual test results of the sample. Detailed Implementation

[0031] The following is in conjunction with the appendix Figure 1-9 The present invention will be further described as follows:

[0032] A machine learning-enhanced fluorescence detection method for silver nanoclusters of splitting nucleic acid aptamers is disclosed. This method uses specifically recognized ATP-based splitting aptamers as recognition elements and DNA-AgNCs as signal reporter molecules. The sequences of the splitting nucleic acid aptamers Apt1 and Apt2 are shown in SEQ ID NO:1 and SEQ ID NO:2, respectively: Apt1: 5'-ACCTGGGGGAGTAT-3'; Apt2: 5'-TGCGGAGGAAGGT-3'. The initial concentrations of Apt1 and Apt2 are 1 μmol / L, and the incubation with ATP is carried out at 37°C for 60 min.

[0033] The 27-base ATP single-stranded aptamer is split into two fragments, serving as split nucleic acid aptamers that specifically bind to ATP. These split nucleic acid aptamers, acting as recognition elements, can reassemble into a three-dimensional binding structure similar to the intact aptamer in the presence of ATP. This alters the secondary structure of the fluorescent probe, resulting in a significant decrease in the fluorescence intensity of the synthesized DNA-AgNCs probe. The ATP content can be determined by detecting this change in fluorescence intensity.

[0034] Examples of methods for preparing fluorescent probes are as follows:

[0035] (1) Dissolution of DNA sequence: Balance the centrifuge tube containing DNA dry powder and centrifuge it. After centrifugation, slowly open the cap and add a certain amount of double-distilled water according to the instructions and shake it thoroughly to obtain a DNA sample solution of 100 μmol / L. Store it at -20℃ for later use.

[0036] (2) Synthesis of DNA-AgNCs signal probes: Apt1 and Apt2 were mixed with Tris-HCl buffer and stirred until homogeneous. A quantitative amount of silver nitrate AgNO3 solution was added to the above solutions, and the mixture was reacted at 4°C in the dark for 20 min. Then, freshly prepared NaBH4 solution was added and shaken thoroughly. The mixture was reacted at 4°C in the dark for 90 min to obtain Apt1-AgNCs and Apt2-AgNCs solutions as signal probes. The concentration ratio of Apt1, AgNO3, and NaBH4 in the synthesis of Apt1-AgNCs signal probe was 1:18:18, and the final concentration ratio of Apt2, AgNO3, and NaBH4 in the synthesis of Apt2-AgNCs signal probe was 1:12:12.

[0037] (3) Recognition reaction of DNA-AgNCs fluorescent probe with ATP: Mix Apt1-AgNCs and Apt2-AgNCs solutions, then add diluted ATP solutions of different concentrations and shake to mix. The temperature for target incubation is set at 37℃, the incubation time is set at 60 min, and the concentration range of ATP is 0-3000 μmol / L.

[0038] This invention utilizes a silver nanocluster fluorescent probe constructed via a splitting aptamer to achieve sensitive detection of ATP. Specific examples of the detection method are as follows:

[0039] (1) Dissolution of DNA sequence: Balance the centrifuge tube containing DNA dry powder and centrifuge it. After centrifugation, slowly open the cap and add a certain amount of double-distilled water according to the instructions and shake it thoroughly to obtain a DNA sample solution of 100 μmol / L. Store it at -20℃ for later use.

[0040] (2) Preparation of Apt1 and Apt2 single chains: Take solutions of Apt1 and Apt2 with a concentration of 100 μmol / L, mix them evenly with an appropriate amount of Tris-HCl buffer, place them in a metal bath for high-temperature denaturation, gradually cool them down after the process, and then place them in a water bath for one hour after passing through an ice bath.

[0041] (3) Synthesis of DNA-AgNCs signal probes: Add a certain amount of silver nitrate AgNO3 solution to the two solutions mentioned above in proportion, react in an ice bath in the dark for 20 min, then add freshly prepared NaBH4 solution and shake thoroughly, react in an ice bath in the dark for 90 min, and Apt1-AgNCs and Apt2-AgNCs solutions are obtained as signal probes.

[0042] (4) Recognition reaction of DNA-AgNCs fluorescent probe with ATP: Mix Apt1-AgNCs and Apt2-AgNCs solutions, then add ATP solution, shake to mix, and incubate at 37℃ for 1 h. The construction of a fluorescent detection method for ATP using split nucleic acid aptamer silver nanoclusters is completed.

[0043] (5) Collect standard solutions of ATP at different concentrations and use DNA-AgNCs fluorescent probes synthesized with aptamers as templates for detection. Collect the fluorescence emission spectra of each sample in the range of 290–600 nm and record its core statistical characteristics such as peak wavelength, maximum height, half width at half maximum (FWHM) and peak area.

[0044] (6) Preprocessing of the original fluorescence spectra, including denoising and baseline correction. The recursive feature elimination (RFE) algorithm was used to screen the feature wavelengths most relevant to ATP concentration from the full spectrum of wavelength variables. The optimal number of features was determined to be 10 through cross-validation, and finally 10 key wavelength variables were obtained, forming a spectral feature set (10-dimensional). At the same time, the maximum height, half-width at half-maximum (FWHM), and peak area were used as statistical feature sets (3-dimensional). Spectral feature sets only; statistical feature sets only; and a combination of spectral and statistical feature sets were constructed.

[0045] (7) A strategy of multiple random partitioning and repeated stratified cross-validation was adopted to model, train and evaluate 10 typical classification algorithms. By comprehensively comparing the classification accuracy and stability of different feature combinations and models, the optimal model for ATP concentration classification was finally determined, so as to realize the intelligent and accurate identification of the freshness of aquatic products.

[0046] Based on the above embodiments: In step (1), the centrifuge tube containing DNA powder is balanced and placed in the centrifuge, and the centrifugation speed is set to 4000 rpm / min and the centrifugation time is 40s.

[0047] In step (2), the denaturation temperature of Apt1 and Apt2 sequences was set at 95 °C for 5 min. The system contained Tris-HCl buffer, specifically composed of 50 mmol / L Tris, 100 mmol / L NaCl, and 10 mmol / L MgCl2, with a pH of 7.0. The concentrations of Apt1 and Apt2 were both 1 µmol / L. The ice bath time was 5 min, the temperature was set at 37 °C, and the incubation time was 1 h.

[0048] In step (3), the concentration ratio of Apt1, AgNO3, and NaBH4 in the synthesis of the Apt1-AgNCs signal probe is 1:18:18, and the final concentration ratio of Apt2, AgNO3, and NaBH4 in the synthesis of the Apt2-AgNCs signal probe is 1:12:12. The ice bath time is 90 min.

[0049] In step (4), the temperature for target incubation is set to 37 °C and the incubation time is set to 60 min.

[0050] The rules for fluorescence spectral preprocessing in step (6) are as follows: Savitzky-Golay smoothing is used for noise reduction, and adaptive iterative reweighted penalized least squares method is used for baseline correction, with a processing wavelength range of 290–600 nm. The rules for the recursive feature elimination (RFE) algorithm used for feature selection are as follows: random forest is used as the base classifier, 3-fold hierarchical cross-validation is used, and the minimum number of features retained is 10.

[0051] The 10 classification and recognition models mentioned in step (7) are: Logistic Regression (LR), Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Gradient Boosting Tree (GBDT), Naive Bayes, and Multilayer Perceptron. Model evaluation uses accuracy, precision, recall, macro-average F1 score, and standard deviation as indicators. The average and standard deviation of the results from 10 experiments are used to obtain a robust estimate of the model performance.

[0052]

[0053]

[0054]

[0055]

[0056] In this context, TP represents a true example, FN represents a false negative, FP represents a false positive, and TN represents a true negative.

[0057] The K-Nearest Neighbors model is the optimal model, with the following optimal parameters: n_neighbors=9, metric=euclidean, weights=uniform. In the K-Nearest Neighbors algorithm, the model selects the 9 nearest neighbor samples to the test sample and determines the class of the test sample through a class vote among these 9 samples, ensuring classification accuracy while avoiding overfitting or underfitting.

[0058] The AUC for the low concentration category was 0.9946, for the medium concentration category it was 0.8688, and for the high concentration category it was 0.9318. All of these values ​​were significantly higher than the random guessing level (AUC=0.5), indicating that the model has the best ability to identify low and high concentrations of ATP, followed by the ability to identify medium concentrations of ATP, and has good overall classification performance.

[0059] The nucleic acid sequences described in the following examples were purchased from Shanghai Sangon Biotech Co., Ltd., and the sequences of the cleavage aptamers used in the embodiments of the present invention are shown in Table 1.

[0060] Table 1. Sequence List

[0061] name Sequence (5'-3') Apt1 5'- ACCTGGGGGAGTAT -3' Apt2 5'- TGCGGAGGAAGGT -3'

[0062] Example 1: Specificity Verification

[0063] The specificity of the method was verified under optimal detection conditions.

[0064] Specificity assay: To evaluate the specificity of this method, the fluorescence emission spectra of aptamers detecting 500 μmol / L ATP solution and other analogs at an excitation wavelength of 320 nm were measured. Plotting different target types on the x-axis and the ratio of fluorescence intensity changes on the y-axis, the differences in fluorescence response between other targets and ATP demonstrated that this method has good specificity in complex environments.

[0065] Detection Procedure: DNA-AgNCs were synthesized according to the literature method. 60 μL of Apt1 (1 μmol / L) was diluted with Tris-HCl buffer. 10 μL of 108 μmol / L AgNO3 solution was added, vortexed for 1 min, and then incubated on ice for 20 min in the dark. 10 μL of 108 μmol / L NaBH4 solution was added to the mixture, vortexed for 1 min, and then incubated on ice for 90 min in the dark to obtain the Apt1-AgNCs signal probe. 60 μL of Apt2 (1 μmol / L) was diluted with Tris-HCl buffer. 10 μL of 72 μmol / L AgNO3 solution was added, vortexed for 1 min, and then incubated on ice for 20 min in the dark. 10 μL of 72 μmol / L NaBH4 solution was added to the mixture, vortexed for 1 min, and then incubated on ice for 90 min in the dark to obtain the Apt2-AgNCs signal probe. Mix 80 μL of Ap1-AgNCs solution with 80 μL of Apt2-AgNCs solution, and add 40 μL of double-distilled water, 40 μL of 500 μmol / L ATP solution, CTP solution, GTP solution, UTP solution, and ribose solution, respectively. Incubate at 37 ℃ for 60 min until the reaction is complete. Then, use a TECAN M1000Pro microplate reader to detect the fluorescence intensity of the reaction system at an excitation wavelength of 270 nm and an emission wavelength of 320 nm. The excitation slit and emission slit were both 10 nm. The fluorescence intensity was obtained by instrument detection.

[0066] Example 2: Plotting Standard Curves

[0067] The sensitivity of this method was evaluated under optimal detection conditions. Fluorescence emission spectra of ATP at excitation wavelengths of 320 nm were recorded at concentrations of 0 µmol / L, 10 µmol / L, 50 µmol / L, 100 µmol / L, 200 µmol / L, 400 µmol / L, 600 µmol / L, 800 µmol / L, 1200 µmol / L, 1600 µmol / L, 2000 µmol / L, and 3000 µmol / L. The dynamic detection range and the linear relationship between ATP concentration and fluorescence intensity were determined by plotting ATP concentration on the x-axis and fluorescence intensity on the y-axis. The linear equation and the limit of detection (LOD) of this method were calculated.

[0068] Detection Procedure: DNA-AgNCs were synthesized according to the literature method. 60 μL of Apt1 (1 μmol / L) was diluted with Tris-HCl buffer. 10 μL of 108 μmol / L AgNO3 solution was added, vortexed for 1 min, and then incubated on ice for 20 min in the dark. 10 μL of 108 μmol / L NaBH4 solution was added to the mixture, vortexed for 1 min, and then incubated on ice for 90 min in the dark to obtain the Apt1-AgNCs signal probe. 60 μL of Apt2 (1 μmol / L) was diluted with Tris-HCl buffer. 10 μL of 72 μmol / L AgNO3 solution was added, vortexed for 1 min, and then incubated on ice for 20 min in the dark. 10 μL of 72 μmol / L NaBH4 solution was added to the mixture, vortexed for 1 min, and then incubated on ice for 90 min in the dark to obtain the Apt2-AgNCs signal probe. Mix 80 μL of Lept1-AgNCs solution with 80 μL of Lept2-AgNCs solution, and add 40 μL of ATP solution of different concentrations. Incubate at 37 ℃ for 60 min. After the reaction is complete, use a TECANM 1000Pro microplate reader to detect the fluorescence intensity of the reaction system at an excitation wavelength of 270 nm and an emission wavelength of 320 nm, with an excitation slit of 10 nm and an emission slit of 10 nm. The linear equation between fluorescence intensity and ATP concentration is calculated.

[0069] Depend on Figure 3 It can be seen that as the ATP concentration increases, the fluorescence intensity in the system decreases in a gradient. Figure 4The results show that when the ATP concentration is between 50 μmol / L and 1200 μmol / L, the ATP concentration and fluorescence intensity have a linear relationship, and the linear equation is: Y = -27.98X + 53977.16, where R... 2 =0.997, and the calculated limit of detection (LOD) is 33 μmol / L, which indicates that the method has excellent sensitivity.

[0070] Example 3: Selection of Optimal Model and Feature Set

[0071] This invention acquires fluorescence spectra of ATP samples using a split nucleic acid aptamer silver nanocluster fluorescence detection method. After denoising and baseline correction, 10 key wavelengths are selected using the RFE algorithm to form a spectral feature set (10-dimensional). At the same time, the maximum height, half width at half maximum (FWHM), and peak area of ​​the fluorescence peak are extracted to form a statistical feature set (3-dimensional). Then, three types of input sets are constructed: spectral-only, statistical-only, and spectral + statistical fusion features (13-dimensional). These inputs are used to train and evaluate models such as logistic regression (LR), linear discriminant analysis (LDA), k-nearest neighbors (KNN), support vector machine (SVM), decision tree (DT), random forest (RF), and gradient boosting tree (GBDT).

[0072] As shown in Figures 5 and 6, the spectral + statistical fusion feature set exhibits the best performance. The LDA, random forest, and optimized KNN models achieve test set accuracy exceeding 98% with a macro-average F1 score of 0.97, representing a performance improvement of over 15% compared to the statistical feature set alone. The optimized KNN model (n_neighbors=9, metric=euclidean, weights=uniform) is selected for comprehensive performance analysis.

[0073] like Figure 7 As shown in the confusion matrix, the model did not misclassify low and high concentration ATP samples, and only misclassified a small number of medium concentration samples, with an overall average accuracy of 96.2%. Figure 8 The ROC curves show that the AUCs for low, medium, and high concentrations of ATP are 0.9938, 0.8681, and 0.9318, respectively, all of which are significantly better than random guessing. The low concentration has the best recognition ability, while the medium concentration has slightly weaker performance but still has good discrimination ability.

[0074] This invention achieves accurate classification of the entire ATP concentration range (low / medium / high) through feature fusion and machine learning calibration, providing reliable support for intelligent detection of aquatic product freshness.

[0075] Example 4: Detection in actual samples

[0076] The detection performance in actual samples was analyzed by detecting the ATP content in real clam samples at three storage temperatures. Clam samples were weighed and grouped, and stored at 25℃, 4℃, and -20℃ for 0 h, 12 h, 24 h, 48 h, and 72 h, respectively.

[0077] Before testing, 5 g of clam sample was homogenized, and then 10 mL of 5% trichloroacetic acid solution was added to the homogenized clam sample. After vortexing for 3 min, the sample was centrifuged (4℃, 10000 rpm). Potassium hydroxide was added to adjust the pH to 7.0, and the sample was centrifuged again at 10000 rpm for 10 min. The supernatant was collected and stored at -20℃ for subsequent fluorescence detection.

[0078] The total bacterial count in clam samples was analyzed according to the method provided in Chinese National Standard GB 4789.2-2016. 25 g of clam was homogenized in 225 mL of sterile physiological saline for 2 min, and the appropriately diluted sample was evenly spread onto agar medium. After incubation at 30 ℃ for 72 h, the total number of microorganisms on the agar medium was calculated.

[0079] Due to microbial growth, the trends of ATP content and TVC over time at different storage temperatures are as follows: Figure 9 As shown, both ATP content and TVC exhibit a continuous upward trend with prolonged storage time, with the fastest growth rate at 25°C, followed by 4°C, and the slowest at -20°C. This further verifies that higher storage temperatures lead to faster microbial proliferation and higher total ATP content in the system, with both trends being highly synchronized. The study also revealed a strong linear relationship between TVC and ATP content. Calculations based on the obtained linear equation showed that when the ATP content of clams stored at 25°C and 4°C exceeded 3.905 mg / 5g and 2.57 mg / 5g, respectively, the TVC exceeded 10 mg / 5g. 7 CFU g -1 It is no longer suitable for human consumption.

[0080] The above descriptions are all preferred embodiments of the present invention. For those skilled in the art, any modifications to the present invention in various equivalent forms without departing from the principle of the present invention shall fall within the protection scope of the appended claims.

Claims

1. A machine learning enhanced split aptamer silver nanocluster fluorescence detection method, characterized in that: The method is as follows: S1: DNA silver nanoclusters were prepared using split nucleic acid aptamers Apt1 and Apt2 as recognition elements. The sequences of the split nucleic acid aptamers Apt1 and Apt2 are shown in SEQ ID NO:1 and SEQ ID NO:2, respectively. S2: The DNA silver nanoclusters signal probes synthesized using split nucleic acid aptamers as templates are Apt1-AgNCs and Apt2-AgNCs, respectively. S3: Mix Apt1 and Apt2 with buffer solution respectively, and after denaturation and annealing, add a certain amount of silver nitrate AgNO3 and sodium borohydride NaBH4 and mix well. Wrap the centrifuge tube with tin foil to avoid light and react to obtain Apt1-AgNCs and Apt2-AgNCs. S4: Mix the Apt1-AgNCs and Apt2-AgNCs solutions obtained in step S3, add them to the sample to be tested, so that the split nucleic acid aptamers specifically bind to the target ATP, inducing a conformational change in the DNA template, resulting in a change in fluorescence intensity; then perform fluorescence spectral acquisition and record the fluorescence intensity values ​​at each wavelength; S5: Preprocess the fluorescence spectrum obtained in step S4, extract spectral and statistical features, construct a target concentration classification model using machine learning algorithms, and achieve quantitative analysis of ATP content by detecting fluorescence spectral features.

2. The machine learning enhanced split aptamer silver nanocluster fluorescence detection method of claim 1, wherein, The initial concentration of Apt1 was 1 μmol / L, the initial concentration of Apt2 was 1 μmol / L, and the temperature for incubation after adding ATP was 37°C and the incubation time was 60 min. 3.The machine learning enhanced split aptamer silver nanocluster fluorescence detection method of claim 1, wherein: In step S3, the buffer solution is Tris-HCl buffer, containing 50 mmol / L Tris, 100 mmol / L NaCl, and 10 mmol / L MgCl2, pH 7.

0. The denaturation conditions are heating at 95°C for 5 min, and the annealing treatment is gradually cooling followed by incubation at 37°C for 60 min. When synthesizing Apt1-AgNCs, the concentration ratio of Apt1, AgNO3, and NaBH4 is controlled at 1:18:

18. After mixing Apt1 with AgNO3 solution, the reaction is carried out at 4°C in the dark for 20 min. Then, freshly prepared NaBH4 solution is added, shaken thoroughly, and the reaction is carried out at 4°C in the dark for 90 min to obtain the Apt1-AgNCs solution. When synthesizing Apt2-AgNCs, the concentration ratio of Apt2, AgNO3, and NaBH4 is controlled at 1:12:

12. After mixing Apt2 with AgNO3 solution, the reaction is carried out at 4°C in the dark for 20 min. After min, add freshly prepared NaBH4 solution and shake thoroughly. React at 4℃ in the dark for 90 min to obtain Apt2-AgNCs solution.

4. The machine learning enhanced split aptamer silver nanocluster fluorescence detection method of claim 1, wherein: After fluorescence spectral preprocessing in step S5, the extracted statistical features include maximum fluorescence intensity, peak area, and full width at half maximum (FWHM). Feature extraction employs a recursive feature elimination algorithm with random forest as the base classifier, combined with 3-fold hierarchical cross-validation, to screen out the feature wavelengths most relevant to ATP concentration from the full spectrum wavelength variables, thus forming a spectral feature set.

5. The machine learning enhanced split aptamer silver nanocluster fluorescence detection method of claim 4, wherein: Machine learning algorithms include one or more of the following: Logistic Regression (LR), Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and Gradient Boosting Tree (GBDT). Spectral features, statistical features, and their fusion features are used as the input set. The classification model is evaluated using accuracy, precision, recall, and macro-average F1 score. The optimal model is determined through multiple random partitions and cross-validations.

6. The machine learning enhanced split aptamer silver nanocluster fluorescence detection method of claim 5, wherein: The K-nearest neighbor model was determined to be the optimal classification model. The optimized parameters were: n_neighbors=9, Euclidean distance as the distance metric, and uniform weighting. Under this parameter combination, the model showed the best classification performance for low, medium, and high concentrations of ATP, with the area under the ROC curve (AUC) reaching 0.9938, 0.8681, and 0.9318, respectively.

Images