A multi-antibiotic rapid detection method based on high-throughput array spectrum and deep learning

Abstract

The application discloses a kind of based on high-flux array spectrum and deep learning multivariate antibiotic rapid detection method, belong to water environment new pollutant detection technical field.The application includes the following steps: antibiotic spectrum sensing system construction, combinatorial chemistry probe screening, high-flux spectrum experiment, deep learning chemometric model development, antibiotic intelligent detection system development.The high-flux combinatorial array spectrum sensing technology used in the application acquires fingerprint image data with holographic spectrum imaging as core, greatly improves experimental efficiency, expands sampling range by combining high-flux experiment and machine learning model, can cover more kinds of chemical mixture combination, deep learning model is good at extracting complex features from a large amount of data, can accurately identify and distinguish the fingerprint spectrum of different kinds of antibiotics, even in complex water environment also can maintain high sensitivity and high specificity, improve the accuracy and response speed of pollutant detection in complex chemical system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new pollutant detection in water environment, in particular to a multi-antibiotic rapid detection method based on high-throughput array spectrum and deep learning. BACKGROUND

[0002] Antibiotics, due to potential hazards such as inducing biological drug resistance genes, disrupting microbial community ecological balance, polluting water quality and threatening human health, have been listed as four typical new pollutants for global environmental pollution prevention and control together with endocrine disruptors, persistent organic pollutants and microplastics. The qualitative and quantitative information of antibiotics in environmental samples such as rivers, soils and groundwater is the basic data source for mastering the spatial and temporal distribution characteristics, migration and transformation rules and ecological risk threshold of antibiotic pollutants. The existing environmental monitoring technology has a mature conventional index detection system, but due to the concomitance of pollution behavior and the complexity of environmental medium, when facing complex antibiotic pollution samples with wide sources, complex composition and low concentration level, the traditional fine detection route of "separation first and then analysis" faces great challenges. In recent years, instrument analysis technologies such as high performance liquid chromatography, fluorescence spectrum, capillary electrophoresis and electrochemiluminescence immunoassay can detect multiple indexes of antibiotic pollution water samples. However, due to the low concentration of antibiotics in nature and the complexity of water matrix, traditional detection technologies face challenges such as complex pretreatment, tedious operation and high cost, which limits their application in actual water detection. Therefore, developing intelligent analysis technology with high sensitivity, high selectivity, cost-effectiveness and capable of simultaneous detection of multiple antibiotics is the main trend of antibiotic detection in environmental samples in the future.

[0003] At present, the existing common antibiotic detection methods still have the following problems: (1) The sources of antibiotic pollution in natural water bodies are wide, the composition is complex, and the concentration is low. Traditional antibiotic detection technology often relies on laboratory large-scale instrument detection and analysis, and these instruments have complex sample pretreatment process, long test period, high requirement for environmental conditions, high cost of detection equipment, tedious operation, high requirement for technical personnel, and large amount of manpower and material resources consumed for sampling analysis and detection; (2) The process of collecting chemical information of the sample to be detected by instrument equipment in the existing AI analysis method is tedious and expensive, and it is difficult to obtain large data required for training AI model in a short time; (3) When a small data set is used for model training, the obtained chemometric model is prone to overfitting, and its robustness and generalization ability are poor, which is difficult to meet the demand of actual application scene. Therefore, when facing antibiotic pollution water samples with multiple sources, complex composition and trace amount, how to develop intelligent analysis technology with high sensitivity, high selectivity and simultaneous detection of multiple antibiotics from the aspects of hardware sensing and software analysis is the main trend of antibiotic detection in environmental samples in the future.

[0004] Based on this, the present application combines high-throughput array spectrum technology with deep learning algorithm, and develops a multi-element antibiotic rapid detection method based on high-throughput array spectrum and deep learning. SUMMARY

[0005] The present application aims to provide a multi-element antibiotic rapid detection method based on high-throughput array spectrum and deep learning, and a significant advantage of the detection technology based on array spectrum deep learning is that the sample pretreatment step is simplified, and at the same time, the use of a large amount of consumables is avoided in the data acquisition stage, the secondary pollution is reduced, the accuracy and repeatability of the detection are ensured, and the complexity of the experimental operation is also significantly reduced, thereby reducing the detection cost and maintenance cost, and solving the problems of traditional antibiotic detection technology, such as complex pretreatment process, low data generation, high cost, and difficulty in rapid multi-index synchronous detection.

[0006] The present application is realized by the following technical solutions:

[0007] The present application is a multi-element antibiotic rapid detection method based on high-throughput array spectrum and deep learning, comprising the following steps:

[0008] S1: Antibiotic spectrum sensing system construction

[0009] The antibiotic spectrum sensing system is composed of a reaction system and a data collection system, and the reaction system comprises a holographic spectrometer, which is used to complete the sample substance information sensing.

[0010] The data collection system is composed of a high-sensitivity computer connected with the reaction system, and the digital spectrum image collected by the reaction system is collected and recorded by the collection system.

[0011] S2: Combination of chemical probes

[0012] The chemical probes that have color difference reaction with antibiotics are selected, and they are divided into two categories of colorimetric probes and fluorescent probes, and they are prepared into standard solutions with antibiotics.

[0013] The solution preparation process comprises:

[0014] S2.1 Drop different chemical probes and antibiotic solutions

[0015] The addition ratio of the chemical probes and the antibiotic solution is 1:1.

[0016] S2.2 Scan colorimetric with enzyme label instrument

[0017] The absorbance of the colorimetric probe is scanned at 360-780nm, and the fluorescence intensity of the fluorescent probe is scanned at 230-490nm.

[0018] S2.3 Screening single probe

[0019] Among them, the single probe with high sensitivity and selectivity is screened.

[0020] S2.4 Making combined chemical probes

[0021] Among them, the single probe with high sensitivity and selectivity is mixed in a certain proportion to make a combined chemical probe.

[0022] S3: High-throughput spectral experiment

[0023] The complex chemical solution containing mixed multi-antibiotics is taken as the research object, the chemical information in the enhanced material is labeled by the combined chemical probe, and the array spectrum is captured by the CCD, then the high-throughput spectral experiment is carried out, and the experimental results are sorted, stored and calculated.

[0024] S4: Deep learning chemometrics model development

[0025] Among them, the model development includes:

[0026] Data set standardization, the data set includes image set and label value, after image acquisition, the image is uniformly standardized.

[0027] CNN network model building, based on the collected array spectral image set containing mixed antibiotics and its corresponding concentration label set, SqueezeNet CNN model is used to systematically supervise the training of the standardized array spectral image set and the corresponding label set.

[0028] CNN model evaluation, for the antibiotic concentration rapid prediction model, the linear fitting analysis is carried out between the model predicted antibiotic concentration and the real concentration in the test set, which is evaluated by the following formula.

[0029]

[0030] Among them, n is the sample size, is the predicted value, y i is the true value,

[0031] S5: Development of antibiotic intelligent detection system

[0032] By using Qt Designer tool on the basis of the trained CNNModel, the antibiotic rapid detection platform is built, and the rapid and direct end-to-end prediction of antibiotic species and its concentration in water sample is realized.

[0033] Further, in the step S2.1, the carrier used for dropping different chemical probes and antibiotic solutions is a 96-well plate.

[0034] Further, when different chemical probes and antibiotic solutions are dropped, a blank control group without antibiotics is set.

[0035] Further, in the step S3, a series of characteristic fingerprint spectrum image sets of multiple mixed antibiotics are collected in batches in a random sampling manner.

[0036] Further, in the step S4, the standardization processing method comprises:

[0037] S4.1 Rotate, crop, and deduct the blank chemical probes and background noise from the sorted spectrum images.

[0038] S4.2 Subtract the experimental images from the background blank images, and save the difference images.

[0039] S4.3 Check and process irrelevant data, repeated data, empty data, abnormal data, and error data.

[0040] S4.4 Correspond the image set and the label set one by one.

[0041] Further, in the step S4, when the model is trained, the fingerprint spectrum images containing mixed antibiotics of different concentrations are taken as inputs, and the concentrations of different antibiotics are taken as output indicators.

[0042] The present application has the following beneficial effects:

[0043] Firstly, the high-throughput combination array spectrum sensing technology adopted by the present application acquires fingerprint image data by taking holographic spectrum imaging as the core, greatly improving the experimental efficiency. This method breaks through the shortcomings of the traditional method, such as complicated experimental process and limited reaction dimension, realizes flexible experiment without input dimension limitation, significantly expands the scale and diversity of data, provides rich data support for the training of machine learning model, and overcomes the problem of data limitation in the traditional method. By combining high-throughput experiment and machine learning model, the sampling range is effectively expanded, so that it can cover more types of chemical mixture combinations. The deep learning model is good at extracting complex features from a large amount of data, and can accurately identify and distinguish the fingerprint spectra of different types of antibiotics. Even in a complex water environment, it can also maintain high sensitivity and high specificity, and improve the accuracy and response speed of pollutant detection in complex chemical systems.

[0044] Meanwhile, by combining high-throughput experiments with machine learning models, reagent consumption, manpower investment, and equipment running time can be significantly reduced, thereby reducing detection costs. The automation and intelligent technology enables researchers to obtain and analyze data with higher efficiency, reduces errors caused by human intervention, and improves the repeatability and reliability of experiments. In summary, based on the prediction results of the deep learning model, decision makers can assist in formulating more scientific and reasonable environmental protection strategies, and improve the intelligence of environmental monitoring and management.

[0045] Of course, implementing any product of the present application does not necessarily require all the advantages described above to be achieved at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is a flowchart of the present application.

[0047] Figure 2 is a schematic diagram of a detection system.

[0048] Figure 3 is a schematic diagram of a high-throughput array spectral sensing system.

[0049] Figure 4 is a scanning result diagram of several probe enzyme labelers.

[0050] Figure 5 is a prediction result diagram of the SqueezeNet V1.1 model.

[0051] Figure 6 is a user interface diagram of the antibiotic intelligent detection system platform. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0053] Please refer to Figures 1-4 The present application provides a technical solution: a multi-element antibiotic rapid detection method based on high-throughput array spectrum and deep learning, comprising the following steps:

[0054] S1: Antibiotic spectral sensing system construction

[0055] The antibiotic spectral sensing system is composed of a reaction system and a data collection system, and the reaction system includes a holographic spectrometer, which completes sample substance information sensing through the holographic spectrometer.

[0056] The holographic spectrometer is composed of a light source, a light uniforming sheet, a filter, a sample reaction cell and a photosensitive coupling element.

[0057] In the embodiment, when the light source emits light, the light is projected onto the special filter. For each point on the color pattern filter, only light of a specific wavelength can pass through, and light of other wavelengths will be absorbed and cannot pass through. Thus, after the light source passes through the special filter, a specific fingerprint spectrum is formed.

[0058] In the embodiment, the sample reaction cell is where the sample to be tested and the combinatorial chemical probe undergo a color difference reaction. When the spectrum passes through the reaction cell, a specific fingerprint spectrum of the sample cell solution is formed. According to the change in the optical signal caused by the change in the type or concentration of the sample solution in the reaction cell, the array spectrum received by the CCD also changes accordingly. The spectrum image obtained by different types and concentrations of samples is like a fingerprint, which is one-to-one corresponding to the specific sample solution, i.e., a specific sample solution forms a specific fingerprint spectrum image.

[0059] The data collection system is composed of a high-sensitivity computer connected to the reaction system. The digital spectrum image collected by the reaction system is collected and recorded by the collection system.

[0060] S2: Combinatorial chemical probe screening

[0061] Taking levofloxacin hydrochloride, tetracycline hydrochloride and erythromycin as an example, 29 different types of single probes that can have specific color difference reactions with them were selected by consulting literature data. According to the different optical properties, they were divided into two categories: colorimetric probes and fluorescent probes. They were prepared into standard solutions with antibiotics, and the names and types are listed in Table 1:

[0062]

[0063]

[0064] Table 1. Single probe and its type

[0065] In each color block of the 96-well plate, different chemical probes and antibiotic solutions were added in the order of 1:1 by volume, i.e. 29 different chemical probes were added from top to bottom, and 3 antibiotics were added from left to right for specific color difference reaction, and each group was set as a blank control group without adding antibiotics. After adding, the colorimetric probe and fluorescent probe were placed in the enzyme marker for scanning the absorbance and fluorescence intensity, respectively. Specifically, the absorbance of the colorimetric probe at 360-780 nm and the fluorescence intensity of the fluorescent probe at 230-490 nm were scanned by the enzyme marker, and the absorption curve and fluorescence curve were drawn. According to the standards of color development sensitivity, large color difference, stable properties, and good specificity, six chemical probes, including cresol red, methyl red, bromothymol blue, fluorescein sodium, eosin y, and rhodamine 6G, were selected and configured into a combined chemical probe according to a certain proportion.

[0066] wherein, Figure 4 The three colorimetric probe absorbance curves and the fluorescein sodium fluorescence curve can be seen. Correspondingly, (a) is the absorbance curve, and (b) is the fluorescein sodium fluorescence curve.

[0067] S3: High-throughput spectral experiment

[0068] A complex chemical solution containing mixed multi-element antibiotics was used as the research object. The chemical information in the substance was labeled and enhanced by a combined chemical probe, and was captured by a CCD in the form of an array spectrum. A spectral experiment was carried out in an antibiotic spectral sensing system using a combined chemical probe, and the volume, concentration, and method of each sample addition were ensured to be consistent to reduce experimental errors. A set of 1500 three-element mixed antibiotic characteristic fingerprint images was collected in a random sampling manner, with a concentration range of 0-30 mg / L. A three-element characteristic fingerprint dataset was constructed by calculating the label set, and the image set and the label set corresponded one by one.

[0069] S4: Development of deep learning chemometric model

[0070] The model development includes:

[0071] The dataset includes an image set and a label value. After image acquisition, the images are uniformly standardized.

[0072] The standardization processing method includes rotating, cropping, and deducting the blank chemical probe and background noise of the sorted spectral images. The difference between the experimental image and the background blank image is saved as a fingerprint spectral image containing only the internal concentration change of the mixed antibiotic. At the same time, irrelevant data, repeated data, empty data, abnormal data, and error data are checked and processed to improve the data quality. Finally, the image set and the label set correspond one by one to ensure that the image set and the label set can meet the requirements of subsequent model training.

[0073] CNN network model building, on the basis of the collected mixed antibiotic array spectrum image set and its corresponding concentration label set, using SqueezeNet CNN model, the standardized array spectrum image set and the corresponding label set are systematically supervised training.

[0074] Among them, on the basis of the collected mixed antibiotic array spectrum image set and its corresponding concentration label set, using SqueezeNet CNN model, the standardized array spectrum image set and the corresponding label set are systematically supervised training. In the model training, the fingerprint spectrum image containing different concentration mixed antibiotics is taken as the input, and the concentration of different antibiotics is taken as the output index. 80% of the data set is used as the training set to train the model, and 20% is used as the test set to test the prediction result of the model. After the training of the deep learning model, the concentration level of multiple antibiotics can be accurately predicted.

[0075] CNN model evaluation, for the antibiotic concentration rapid prediction model, the linear fitting analysis is carried out between the antibiotic concentration predicted by the model in the test set and the true concentration. The evaluation is carried out through the following formula.

[0076]

[0077] Among them, n is the sample size, is the predicted value, y i is the true value,

[0078] In the above embodiment, based on the constructed data set, through data standardization, cleaning, difference making and other operations, the subsequent model training is prepared, and then they are input into the SqueezeNet V1.1 CNN network model for supervised training. The antibiotics of different concentrations are analyzed and predicted. For the deep learning algorithm concentration prediction model, R2, MAE and RMSE three indexes are used to evaluate the model prediction performance.

[0079] As Figure 5Figure 2 shows the linear fit results of the SqueezeNet V1.1 network model for the concentration prediction of a three-component mixture of levofloxacin hydrochloride, tetracycline hydrochloride, and dirithromycin. The predicted values ​​and the true values ​​are generally well matched. The R², MAE, and RMSE for levofloxacin hydrochloride are 0.926, 1.52, and 1.18, respectively; for tetracycline hydrochloride, they are 0.966, 0.972, and 0.783; and for dirithromycin, they are 0.874, 1.88, and 1.50, respectively. These results demonstrate that the model exhibits high accuracy and reliability when simultaneously predicting the concentrations of multiple antibiotics. Comparing the prediction results for the three antibiotics, tetracycline hydrochloride outperforms levofloxacin hydrochloride and dirithromycin in all three metrics, demonstrating that the model can meet the requirements for simultaneous detection of multiple antibiotic concentrations.

[0080] S5: Development of Intelligent Antibiotic Detection System

[0081] like Figure 6 The figure shows an intelligent antibiotic detection system platform built using Qt Designer. Through its intuitive user interface, it provides users with a convenient operation process. Users only need to click "Select Image" to upload the array spectrum image to be predicted, then click "Save Prediction Results" to select the location to output the prediction results. The system will automatically start the prediction model and accurately predict the antibiotic concentration. The prediction results are intuitively presented on the interface. At the same time, the interface supports adaptive resizing, ensuring that users can experience smooth and consistent operation on different devices. This design not only improves user convenience, but also optimizes the overall interaction process.

[0082] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A rapid detection method for multiple antibiotics based on high-throughput array spectroscopy and deep learning, characterized in that: The steps include: S1: Construction of antibiotic spectral sensing system Among them, the antibiotic spectral sensing system consists of a reaction system and a data collection system. The reaction system includes a holographic spectrometer, which completes the perception of sample material information through the holographic spectrometer; The data collection system consists of a highly sensitive computer connected to the reaction system. The digital spectrum images collected by the reaction system are collected and recorded by the collection system. S2: Combinatorial Chemical Probe Screening Select chemical probes that react with antibiotics to produce color differences and divide them into two categories: colorimetric probes and fluorescent probes. Prepare standard solutions with them and antibiotics. The solution preparation process includes: S2.1 Add different chemical probes and antibiotic solutions Among them, the addition ratio of chemical probe and antibiotic solution was 1:1; S2.2 microplate reader for colorimetry Among them, the absorbance of the colorimetric probe was scanned at 360-780nm, and the fluorescence intensity of the fluorescent probe was scanned at 230-490nm; S2.3 Screening of single probes Among them, screening for single probes with high sensitivity and selectivity; S2.4 Preparation of combinatorial chemical probes Among them, highly sensitive and selective single probes are mixed in a certain ratio to make a combination chemical probe; S3: High-throughput spectroscopy experiments Complex chemical solutions containing mixed multi-antibiotics are used as research objects. The chemical information in the substances is enhanced by combinatorial chemical probe labeling and captured in the form of array spectra using CCD. Then, high-throughput spectral experiments are carried out and the experimental results are organized, stored, and calculated. S4: Deep Learning Chemometric Model Development Model development includes: Dataset standardization: The data set includes image sets and label values. After image acquisition is completed, the images are uniformly standardized. The CNN network model was constructed. Based on the collected mixed antibiotic array spectral image set and the label set of their corresponding concentrations, the SqueezeNet CNN model was used to perform systematic supervised training on the standardized array spectral image set and the corresponding label set. CNN model evaluation: For the rapid antibiotic concentration prediction model, a linear fit analysis was performed between the antibiotic concentration predicted by the model in the test set and the actual concentration. The evaluation was performed using the following formula: Where n is the sample size, is the predicted value, y i is the true value, S5: Development of Intelligent Antibiotic Detection System By using the Qt Designer tool based on the trained CNNModel, a rapid antibiotic detection platform was created to achieve fast and direct end-to-end prediction of the types and concentrations of antibiotics in water samples.

2. The method for rapid detection of multiple antibiotics based on high-throughput array spectroscopy and deep learning according to claim 1, characterized in that: The holographic spectrometer in step S1 is composed of a light source, a light homogenizer, a filter, a sample reaction cell and a photosensitive coupling element.

3. The method for rapid detection of multiple antibiotics based on high-throughput array spectroscopy and deep learning according to claim 1, characterized in that: In step S2.1, the carrier used for adding different chemical probes and antibiotic solutions is a 96-well plate.

4. The method for rapid detection of multiple antibiotics based on high-throughput array spectroscopy and deep learning according to claim 3, characterized in that: When different chemical probes and antibiotic solutions were added dropwise, a blank control group without antibiotics was set up.

5. The method for rapid detection of multiple antibiotics based on high-throughput array spectroscopy and deep learning according to claim 1, characterized in that: In step S3, a series of characteristic fingerprint spectral image sets of multivariate mixed antibiotics are collected in batches by random sampling.

6. The method for rapid detection of multiple antibiotics based on high-throughput array spectroscopy and deep learning according to claim 1, characterized in that: In step S4, the standardization processing method includes: S4.1 rotates, crops, and subtracts blank chemical probes and background noise from the organized spectral image; S4.2 Make a difference between the experimental image and the background blank image and save the difference image; S4.3 Check and process irrelevant data, duplicate data, empty data, abnormal data, and erroneous data; S4.4 establishes a one-to-one correspondence between the image set and the label set.

7. The method for rapid detection of multiple antibiotics based on high-throughput array spectroscopy and deep learning according to claim 6, characterized in that: In step S4, when the model is trained, the fingerprint spectrum image containing mixed antibiotics of different concentrations is used as input, and the concentrations of different antibiotics are used as output indicators.

Images