Underground water micro-pollution analysis method combining SERS-nanopore sequencing and TwinNet-LSTM

Abstract

The invention belongs to the technical field of environmental pollution monitoring, and particularly relates to an underground water pollution tracing method combining two detection means of surface enhanced Raman scattering (SERS) and nanopore sequencing and two intelligent algorithms of a twinNet and a long-short term memory (LSTM). Specifically, high-sensitivity detection is achieved through an SERS substrate, TwinNet improves the anti-interference capacity through two-channel input, nanopore sequencing and an improved LSTM algorithm are combined, a DNA library of microorganisms in underground water is obtained, time sequence data are processed through an LSTM model, the time change trend of microbial communities is predicted, and reverse prediction of micropollutants is achieved. The device and the method have the beneficial effects that the device and the method can be used for monitoring the influence of micropollutants in underground water on microbial communities, and can also give out early warning on the micro-pollution of the underground water through the change of microbial populations.

Description

1. Technical Field

[0001] The present invention belongs to the field of environmental pollution monitoring technology, and in particular, relates to a method for tracing the source of groundwater pollution that combines two detection methods, surface enhanced Raman scattering (SERS) and nanopore sequencing, with two intelligent algorithms, twin neural networks (TwinNet) and long short-term memory networks (LSTM). The main purpose is to solve the problem of monitoring the impact of micropollutants in groundwater on microbial communities, and to provide early warning of groundwater micropollutants through changes in microbial populations. 2. Background Technology

[0002] Scholars at home and abroad have developed a variety of source analysis techniques, including isotope method, fluorescence spectroscopy, geostatistics, principal component analysis (PCA) and positive definite matrix factorization (PMF) model. These technologies provide important basis for groundwater pollution prevention and control by analyzing the chemical characteristics and migration patterns of pollutants. However, due to the complexity of groundwater systems and the diversity of pollutants, existing technologies still face many challenges in practical application.

[0003] Existing source analysis technology shows certain limitations in groundwater pollution identification, mainly due to factors such as complex geological conditions, strong groundwater fluidity, low pollutant concentrations, and difficulty in sampling. First, the heterogeneity and anisotropy of the groundwater system make it difficult to accurately predict the migration path of pollutants. Secondly, the strong groundwater fluidity leads to a wide range of pollutant diffusion. In addition, the low concentration of micropollutants (some can reach the ng / L level) makes it difficult for traditional detection methods to accurately identify, and the concentration of some organic pollutants in water bodies is lower than the instrument detection limit, resulting in increased uncertainty in traceability results. Finally, the difficulty of sampling further limits the representativeness of the data and the accuracy of analysis. Under complex geological conditions, the layout of sampling points and the choice of sampling frequency have a significant impact on the results. These shortcomings show that existing technologies are in urgent need of innovative breakthroughs in the analysis of groundwater micro-pollution.

[0004] In recent years, significant progress has been made in the field of groundwater micropollutant source analysis technology at home and abroad, mainly focusing on high-sensitivity detection, multi-technology integration and intelligent algorithm application. In terms of source analysis methods, stable isotope tracing technology has become the mainstream. For example, the Environmental Stable Isotope Laboratory of the Chinese Academy of Agricultural Sciences successfully quantified NO in groundwater in the Chengdu Plain by combining the PMF model with stable isotope technology. 3 -The contribution rate of pollution sources, among which agricultural activities contributed 66.76% and domestic pollution contributed 16.47%. In addition, multivariate statistical analysis methods (such as principal component analysis and positive definite matrix factor decomposition) have been widely used in groundwater pollution source analysis. For example, Zhang Guanglu et al. found through principal component analysis that agricultural activities, ion exchange and nitrification are the main reasons for the formation of high concentrations of nitrates in groundwater in the piedmont plain of North China. In terms of intelligent algorithms, machine learning technology has gradually been introduced into source analysis models.

[0005] However, to date, a mature framework for successful source analysis of groundwater micropollutants has not yet been formed. The main reasons are:

[0006] 1) The detection of trace pollutants is easily interfered by the environmental background value and is not representative;

[0007] 2) The geological and groundwater conditions of the contaminated sites are complex, and the strategy of tracing the source of the pollution cannot solve the problem in a targeted manner;

[0008] 3) When conducting pollution source tracing, the interference effect of micropollutants on the groundwater ecological environment was not taken into account, nor was there any attempt to use this interference effect to reversely trace the micropollutants. III. Summary of the invention

[0009] The present invention is made in view of the problems existing in the prior art. The present invention adopts SERS and nanopore sequencing technology to achieve efficient detection of water samples. When processing the SERS signal of the water sample, a special high-sensitivity gold-copper alloy nanoparticle-modified silica substrate material is designed to improve the imaging quality and measurement accuracy, and an aspheric lens group is introduced to reduce aberrations; the nanopore sequencer has high-throughput single-molecule sequencing capabilities, and a large number of DNA fragment sequencing tasks can be completed in a few hours using a fast mode. These technical means ensure the effective detection of trace or even trace pollutants and DNA sequences, and overcome the problem that traditional methods are easily interfered by environmental background values.

[0010] Secondly, the present invention introduces TwinNet and improved bidirectional LSTM models to predict the population correlation between SERS spectra and DNA libraries and reversely predict the time correlation between the two. The TwinNet model processes the original SERS spectra and background signals through dual-channel input to improve anti-interference ability, introduces a residual connection structure to avoid the gradient vanishing problem, and speeds up training; batch normalization (BN) technology is applied to enhance model stability. The bidirectional LSTM architecture not only considers past information, but also captures future contextual information, which is suitable for trend analysis of microbial communities over time. In a simulated environment, the TwinNet model achieved a prediction accuracy of 95.67% for specific types of pollutants, while the bidirectional LSTM model increased the accuracy of capturing complex patterns in time series by 15.2%.

[0011] Finally, the present invention combines the pollutant characteristics obtained by SERS detection with the microbial DNA sequence obtained by nanopore sequencing, comprehensively analyzes the correlation between the two sets of data, and realizes effective monitoring of the effects of micropollutants in groundwater on microbial communities and timely early warning of groundwater micropollutants by changes in DNA libraries. By deduplicating the library, the repetitive sequences generated during the PCR amplification process are removed to ensure the authenticity and reliability of the sequencing results. In addition, the nanopore sequencing equipment is also equipped with error correction algorithms such as Racon and Medaka, which can further correct sequence errors after preliminary assembly and reduce the misjudgment rate. This method that combines physical and chemical detection with bioinformatics analysis can not only provide high-precision identification and traceability of micropollutants, but also predict their impact on groundwater ecosystems, achieving the unity of pertinence and comprehensiveness.

[0012] Specifically, the technical solution of the present invention is a method for analyzing groundwater micro-pollution by combining SERS-nanopore sequencing and TwinNet-LSTM, characterized in that the method comprises the following steps:

[0013] Step 1: Set up multiple sampling points in the suspected contaminated area to ensure that the samples are representative. Use sterile polyethylene bottles to collect water samples to avoid cross-samples. Use filter paper with a pore size of 1.2 microns to remove larger particles and impurities from the contaminated water samples. Then use a polyethersulfone filter membrane with a pore size of 0.22 microns to intercept and enrich bacteria, viruses and other microorganisms. The resulting water sample is diluted with ultrapure water to an appropriate concentration to meet the detection requirements of the surface enhanced Raman scattering spectrometer. In order to eliminate the interference effect of impurities in the water, the ultraviolet absorption correction method should be used to pre-treat the sample to ensure data accuracy. Use an ultrasonic crusher to place the polyethersulfone filter membrane loaded with bacteria, viruses and other microorganisms in a sterile beaker, perform ultrasonic treatment under ice bath conditions, and use the shock wave and shear force generated when the ultrasonic bubble ruptures to break the cells. The ultrasonic power is 200 to 400 watts, the ultrasonic time is 10 to 30 seconds each time, and the interval is 1 to 2 minutes, and it is repeated 3 to 5 times.

[0014] Step 2: Collect the surface enhanced Raman scattering (SERS) signal of the diluted water sample. First, design a dedicated high-sensitivity SERS substrate material, select silica modified with gold-copper alloy nanoparticles, and use a spherical carrier with a diameter of 30 to 50 nanometers and a distribution density of 10 9 Up to 10 11 per square centimeter. The spectrometer supports an excitation wavelength range of 633 to 785 nanometers. It uses local fluorescence enhancement technology to add a nanoparticle coating on the sample surface to enhance the SERS signal intensity. It introduces an aspheric lens group to reduce aberrations and improve imaging quality. The built-in temperature control system has a stable working temperature within the range of ±0.1°C to prevent temperature fluctuations from interfering with measurement accuracy. It collects spectral information in batches and reduces the impact of noise interference through signal accumulation enhancement.

[0015] Step 3: Collect DNA information from water samples. The nanopore sequencer needs to have high-throughput single-molecule sequencing capabilities, read lengths of hundreds of thousands of base pairs, and a sequencing accuracy rate of ≥90%. It needs a built-in temperature control system to keep the sequencing environment constant at 37°C. It uses a fast mode to complete a large number of DNA fragment sequencing tasks within a few hours. It is also equipped with real-time data analysis software to monitor the sequencing progress and quality in real time.

[0016] Step 4: Prepare the library of the microbial DNA extracted from groundwater using the Nextera XT DNA Library Prep Kit to ensure that the library size is uniform and suitable for subsequent sequencing analysis. Before sequencing, the library is quantified and quality checked using the Qubit Fluorometer and Agilent Bioanalyzer instruments to ensure that the library concentration is between 0.2 and 2 nanograms per microliter and the fragment length is mainly distributed between 300 and 700 base pairs.

[0017] Step 5: Use the TwinNet model to predict the population correlation between the SERS spectrum and the DNA library. First, the original SERS spectrum and the background signal are processed separately through dual-channel input to improve the anti-interference ability. Each channel contains a convolution layer, a pooling layer, and a fully connected layer structure. The convolution kernel sizes are 3×3, 5×5, and 7×7 to extract multi-scale spatial features. The residual connection structure is introduced to avoid the gradient vanishing problem and speed up the training. Batch normalization (BN) technology is applied to enhance the stability of the model. Finally, the final result is output through the Softmax classifier to ensure the prediction accuracy.

[0018] Step 6, using the improved LSTM network to reversely predict the temporal correlation between the DNA library and the SERS spectrum, first construct an improved LSTM network architecture, each layer contains 128 to 256 hidden units, the activation function uses ReLU to accelerate the convergence process, the introduction of the Dropout mechanism is set to a ratio of 0.2 to 0.5 to prevent overfitting problems, the Adam optimization algorithm is used in the model training process with a learning rate of 0.001 to 0.01, and the batch size is set to 32 to 64 to balance the computational efficiency and model performance;

[0019] Step 7: Add a reverse transfer path to the standard LSTM to form a bidirectional LSTM structure, so that the model can consider both past and future contextual information, thereby better capturing complex patterns in time series and being suitable for the trend of microbial communities changing over time;

[0020] Step 8: Combine the pollutant characteristics obtained by SERS detection with the microbial DNA sequences obtained by nanopore sequencing, and use TwinNet and bidirectional LSTM models to comprehensively analyze the correlation between the two sets of data, thereby achieving effective monitoring of the effects of micropollutants in groundwater on microbial communities and timely early warning of groundwater micropollutants due to changes in the DNA library.

[0021] It is further defined that the improved surface enhanced Raman scattering spectrometer acquisition method described in step 2 also involves the introduction of an automatic focusing function, that is, using a laser rangefinder to monitor the liquid level in the sample pool in real time to ensure that the focus position is consistent during each measurement, thereby improving data consistency.

[0022] It is further defined that the nanopore sequencing device described in step 4 also includes error correction algorithms for improving sequencing accuracy, such as Racon and Medaka, which can further correct sequence errors after preliminary assembly to reduce the misjudgment rate.

[0023] It is further defined that the acquisition and preprocessing of the microbial DNA sequence described in step 5 also includes deduplication processing of the library to remove the repeated sequences generated during the PCR amplification process to ensure the authenticity and reliability of the sequencing results.

[0024] It is further defined that the introduction of the bidirectional LSTM in step 7 to improve the long-term dependency modeling capability also includes the introduction of an attention mechanism to dynamically adjust the weights of each layer, highlight important feature areas, and suppress noise interference.

[0025] The present invention is beneficial in that:

[0026] 1) High-precision detection: Combining SERS and nanopore sequencing technology to achieve efficient detection of trace pollutants, avoid environmental background interference, and ensure data accuracy and representativeness.

[0027] 2) Intelligent algorithm analysis: Using TwinNet and bidirectional LSTM models, the relationship between pollutants and microorganisms is analyzed from the dimensions of time and space, providing accurate traceability and prediction of ecological impacts.

[0028] 3) Comprehensive monitoring and early warning: Integrate physical and chemical testing with bioinformatics analysis to monitor changes in groundwater micro-pollution in real time, issue early warnings in a timely manner, and ensure the safety of groundwater resources and ecological health. IV. Description of the drawings

[0029] In order to more clearly illustrate the specific embodiments of the present invention, the drawings used in the description of the specific embodiments are briefly described below.

[0030] Figure 1 This is a technical roadmap for a groundwater micro-pollution analysis method that combines SERS-nanopore sequencing with TwinNet-LSTM.

[0031] Figure 2 To predict the population correlation between SERS spectra and DNA libraries based on the TwinNet neural network model.

[0032] Figure 3 The improved LSTM network is used to reversely predict the temporal correlation between the DNA library and the SERS spectrum.

[0033] Figure 4 To comprehensively predict population and time correlation based on TwinNet and bidirectional LSTM models. V. Specific implementation methods

[0034] The present invention is described in detail below with reference to the accompanying drawings and embodiments:

[0035] Embodiment 1:

[0036] A method for analyzing groundwater micro-pollution by combining SERS-nanopore sequencing with TwinNet-LSTM, characterized in that the method specifically comprises the following steps: Step 1, setting multiple sampling points in a suspected polluted area to ensure that the samples are representative, using sterile polyethylene bottles to collect water samples to avoid crossover between samples; first using filter paper with a pore size of 1.2 microns to remove larger particles and impurities from the polluted water sample, and then using a polyethersulfone filter membrane with a pore size of 0.22 microns to intercept and enrich bacteria, viruses and other microorganisms, and the obtained water sample is diluted with ultrapure water to a suitable concentration to meet the detection requirements of a surface enhanced Raman scattering spectrometer, and in order to eliminate the interference effect of impurities in the water, an ultraviolet absorption correction method should be used to pre-treat the sample to ensure data accuracy; using an ultrasonic crusher, placing a polyethersulfone filter membrane loaded with bacteria, viruses and other microorganisms in a sterile beaker, and performing ultrasonic treatment under ice bath conditions, using the shock wave and shear force generated when the ultrasonic bubble ruptures to break the cells, the ultrasonic power is 200 to 400 watts, the ultrasonic time is 10 to 30 seconds each time, the interval is 1 to 2 minutes, and it is repeated 3 to 5 times;

[0037] Step 2: Collect the surface enhanced Raman scattering (SERS) signal of the diluted water sample. First, design a dedicated high-sensitivity SERS substrate material, select silica modified with gold-copper alloy nanoparticles, and use a spherical carrier with a diameter of 30 to 50 nanometers and a distribution density of 10 9 Up to 10 11 per square centimeter, the spectrometer supports an excitation wavelength range of 633 to 785 nanometers, uses local fluorescence enhancement technology, adds a nanoparticle coating on the sample surface to enhance the SERS signal intensity, introduces an aspheric lens group to reduce aberrations and improve imaging quality, and has a built-in temperature control system that stabilizes the operating temperature within the range of ±0.1°C to prevent temperature fluctuations from interfering with measurement accuracy. It collects spectral information in batches and reduces the impact of noise interference through signal accumulation enhancement; the surface enhanced Raman scattering spectrometer acquisition method also involves the introduction of an automatic focusing function, that is, using a laser rangefinder to monitor the liquid level in the sample pool in real time to ensure that the focus position is consistent for each measurement and improve data consistency

[0038] Step 3: Collect DNA information from water samples. The nanopore sequencer needs to have high-throughput single-molecule sequencing capabilities, read lengths of hundreds of thousands of base pairs, and a sequencing accuracy rate of ≥90%. It needs a built-in temperature control system to keep the sequencing environment constant at 37°C. It uses a fast mode to complete a large number of DNA fragment sequencing tasks within a few hours. It is also equipped with real-time data analysis software to monitor the sequencing progress and quality in real time.

[0039] Step 4: Prepare the library of the microbial DNA extracted from the groundwater using the Nextera XT DNA Library Prep Kit to ensure that the library size is uniform and moderate, suitable for subsequent sequencing analysis. Before sequencing, the library is quantified and quality checked using the Qubit Fluorometer and Agilent Bioanalyzer instruments to ensure that the library concentration is between 0.2 and 2 nanograms per microliter, and the fragment length is mainly distributed between 300 and 700 base pairs. Among them, the nanopore sequencing equipment also includes error correction algorithms for improving sequencing accuracy, such as Racon and Medaka, which can further correct sequence errors after preliminary assembly to reduce the misjudgment rate;

[0040] Step 5: Use the TwinNet model to predict the population correlation between the SERS spectrum and the DNA library. First, the original SERS spectrum and the background signal are processed separately through dual-channel input to improve the anti-interference ability. Each channel contains a convolution layer, a pooling layer and a fully connected layer structure. The convolution kernel sizes are 3×3, 5×5 and 7×7 to extract multi-scale spatial features. The residual connection structure is introduced to avoid the gradient vanishing problem and speed up the training. Batch normalization (BN) technology is applied to enhance the model stability. Finally, the final result is output through the Softmax classifier to ensure the prediction accuracy. Among them, the acquisition and preprocessing of microbial DNA sequences also include deduplication of the library to remove the repeated sequences generated during PCR amplification to ensure the authenticity and reliability of the sequencing results.

[0041] Step 6, using the improved LSTM network to reversely predict the temporal correlation between the DNA library and the SERS spectrum, first construct an improved LSTM network architecture, each layer contains 128 to 256 hidden units, the activation function uses ReLU to accelerate the convergence process, the introduction of the Dropout mechanism is set to a ratio of 0.2 to 0.5 to prevent overfitting problems, the Adam optimization algorithm is used in the model training process with a learning rate of 0.001 to 0.01, and the batch size is set to 32 to 64 to balance the computational efficiency and model performance;

[0042] Step 7, adding a reverse transfer path to the standard LSTM to form a bidirectional LSTM structure, so that the model can consider past and future contextual information at the same time, so as to better capture the complex patterns in the time series, which is suitable for the trend of microbial communities changing over time; among them, the introduction of bidirectional LSTM improves the ability to model long-term dependencies, and also includes the introduction of an attention mechanism to dynamically adjust the weights of each layer, highlight important feature areas, and suppress noise interference.

[0043] Step 8: Combine the pollutant characteristics obtained by SERS detection with the microbial DNA sequences obtained by nanopore sequencing, and use TwinNet and bidirectional LSTM models to comprehensively analyze the correlation between the two sets of data, thereby achieving effective monitoring of the effects of micropollutants in groundwater on microbial communities and timely early warning of groundwater micropollutants due to changes in the DNA library.

[0044] The technical route described in Example 1 is as shown in the attached Figure 1 shown.

[0045] Embodiment 2:

[0046] Using the technical route in Example 1, 10 types of antibiotic micropollutants, including β-lactams, macrolides, aminoglycosides, tetracyclines, fluoroquinolones, sulfonamides, chloramphenicol, lincosamides, polypeptides, and glycopeptides, were detected in groundwater in Xingtai, Hebei. The experimental process is as described in Example 1 and will not be repeated here.

[0047] First, the surface enhanced Raman scattering spectra of 10 types of antibiotic standards were established and normalized to form a cross matrix. Then, actual groundwater samples were collected and the surface enhanced Raman scattering spectra of the water samples were reconstructed; the correlation matrix between groundwater samples at different points and the spectra of 10 types of antibiotics is shown in the following table:

[0048] Correlation Matrix β-lactam Macrolide Aminoglycoside tetracycline Fluoroquinolones Sulfonamide Chloramphenicol Lincosamide Peptides Glycopeptides 1# 0.126 0.093 0.168 0.027 0.291 0.462 0.460 0.476 0.069 0.252 2# 0.159 0.398 0.108 0.275 0.281 0.414 0.451 0.274 0.324 0.324 3# 0.282 0.225 0.269 0.041 0.138 0.098 0.019 0.225 0.276 0.438 4# 0.366 0.139 0.490 0.437 0.387 0.361 0.073 0.300 0.268 0.266 5# 0.233 0.359 0.096 0.125 0.388 0.120 0.419 0.126 0.207 0.269 6# 0.484 0.383 0.058 0.207 0.055 0.258 0.119 0.228 0.103 0.198 7# 0.207 0.391 0.273 0.184 0.340 0.185 0.440 0.469 0.427 0.075 8# 0.473 0.489 0.376 0.039 0.275 0.223 0.106 0.490 0.121 0.397 9# 0.088 0.372 0.054 0.005 0.306 0.183 0.392 0.242 0.005 0.387 10# 0.318 0.445 0.187 0.378 0.404 0.259 0.175 0.395 0.121 0.077 11# 0.024 0.460 0.409 0.211 0.336 0.154 0.274 0.155 0.235 0.081 12# 0.340 0.307 0.462 0.261 0.445 0.281 0.322 0.408 0.012 0.100

[0049] Based on the above analysis, potential antibiotic micropollutants at different sampling points were identified. Furthermore, based on the DNA library information in the water samples, the population correlation between the SERS spectra and the DNA library was predicted based on the TwinNet model. The results are shown in the attached figure. Figure 2 As shown in Figure 2, it can be seen that the SERS spectrum is weakly correlated with the DNA library. Then, the improved LSTM network is used to reversely predict the temporal correlation between the DNA library and the SERS spectrum (see Figure 2). Figure 3 Finally, a reverse transfer path is added to the standard LSTM to form a bidirectional LSTM structure, which allows the model to consider both past and future context information at the same time and comprehensively analyze the correlation between the two sets of data. Figure 4 As shown, the results of combining TwinNet and bidirectional LSTM models significantly improved the prediction accuracy between surface enhanced Raman scattering spectra and DNA libraries, so that it can better meet the effective monitoring of the effects of micropollutants in groundwater on microbial communities and the timely early warning of groundwater micropollutants due to changes in DNA libraries.

[0050] Embodiment 3:

[0051] Using the technical route in Example 1, 10 types of antibiotic micropollutants such as β-lactams, macrolides, aminoglycosides, tetracyclines, fluoroquinolones, sulfonamides, chloramphenicol, lincosamides, polypeptides, and glycopeptides in groundwater in Dashiqiao City, Liaoning Province were detected. The experimental process is as described in Example 1 and will not be repeated here.

[0052] First, the surface enhanced Raman scattering spectra of 10 types of antibiotic standards were established and normalized to form a cross matrix. Then, actual groundwater samples were collected and the surface enhanced Raman scattering spectra of the water samples were reconstructed; the correlation matrix between groundwater samples at different points and the spectra of 10 types of antibiotics is shown in the following table:

[0053] Correlation Matrix β-lactam Macrolide Aminoglycoside tetracycline Fluoroquinolones Sulfonamide Chloramphenicol Lincosamide Peptides Glycopeptides 1# 0.623 0.562 0.002 0.411 0.618 0.114 0.349 0.373 0.499 0.100 2# 0.526 0.370 0.274 0.457 0.309 0.060 0.295 0.415 0.650 0.598 3# 0.641 0.072 0.122 0.674 0.248 0.442 0.492 0.238 0.052 0.546 4# 0.629 0.081 0.267 0.071 0.171 0.082 0.149 0.273 0.149 0.628 5# 0.580 0.123 0.552 0.478 0.471 0.263 0.308 0.258 0.314 0.430 6# 0.364 0.321 0.245 0.649 0.052 0.170 0.143 0.243 0.503 0.161 7# 0.628 0.225 0.194 0.694 0.459 0.166 0.073 0.181 0.489 0.028 8# 0.444 0.643 0.094 0.672 0.148 0.224 0.602 0.093 0.466 0.348 9# 0.343 0.200 0.181 0.288 0.363 0.217 0.194 0.519 0.426 0.537 10# 0.643 0.370 0.068 0.050 0.370 0.387 0.007 0.675 0.665 0.181 11# 0.593 0.162 0.201 0.573 0.076 0.519 0.019 0.466 0.540 0.659 12# 0.310 0.547 0.197 0.342 0.005 0.021 0.282 0.000 0.581 0.478

[0054] Based on the above analysis, potential antibiotic micropollutants at different sampling points were identified. Furthermore, based on the DNA library information in the water sample, the population correlation between the SERS spectrum and the DNA library was predicted based on the TwinNet model, and the SERS spectrum and the DNA library were weakly correlated; then, the improved LSTM network was used to reversely predict the time correlation between the DNA library and the SERS spectrum, and it can be seen that the time correlation between the two is weak; finally, a reverse transfer path was added on the basis of the standard LSTM to form a bidirectional LSTM structure, so that the model can simultaneously consider the past and future context information, and comprehensively analyze the association between the two sets of data, forming a strong correlation. It can be seen that the combination of TwinNet and bidirectional LSTM models significantly improves the prediction accuracy between surface enhanced Raman scattering spectra and DNA libraries, so that it can better meet the effective monitoring of the effects of micropollutants in groundwater on microbial communities and the timely warning of groundwater micropollutants due to changes in DNA libraries.

[0055] The specific implementation methods described above are only used to specifically illustrate the spirit of the present invention, and the protection scope of the present invention is not limited thereto. For those skilled in the art, other implementation methods can certainly be easily made by changing, replacing or modifying the technical contents disclosed in this specification, and these other implementation methods should all be included in the protection scope of the present invention.

Claims

1. A method for analyzing groundwater micro-pollution by combining SERS-nanopore sequencing and TwinNet-LSTM, characterized in that The method specifically comprises the following steps: Step 1: Set up multiple sampling points in the suspected contaminated area to ensure that the samples are representative. Use sterile polyethylene bottles to collect water samples to avoid cross-samples. Use filter paper with a pore size of 1.2 microns to remove larger particles and impurities from the contaminated water samples. Then use a polyethersulfone filter membrane with a pore size of 0.22 microns to intercept and enrich bacteria, viruses and other microorganisms. The resulting water sample is diluted with ultrapure water to an appropriate concentration to meet the detection requirements of the surface enhanced Raman scattering spectrometer. In order to eliminate the interference effect of impurities in the water, the ultraviolet absorption correction method should be used to pre-treat the sample to ensure data accuracy. Use an ultrasonic crusher to place the polyethersulfone filter membrane loaded with bacteria, viruses and other microorganisms in a sterile beaker, perform ultrasonic treatment under ice bath conditions, and use the shock wave and shear force generated when the ultrasonic bubble ruptures to break the cells. The ultrasonic power is 200 to 400 watts, the ultrasonic time is 10 to 30 seconds each time, and the interval is 1 to 2 minutes, and it is repeated 3 to 5 times. Step 2: Collect the surface enhanced Raman scattering (SERS) signal of the diluted water sample. First, design a dedicated high-sensitivity SERS substrate material, select silica modified with gold-copper alloy nanoparticles, and use a spherical carrier with a diameter of 30 to 50 nanometers and a distribution density of 10 9 Up to 10 11 per square centimeter. The spectrometer supports an excitation wavelength range of 633 to 785 nanometers. It uses local fluorescence enhancement technology to add a nanoparticle coating on the sample surface to enhance the SERS signal intensity. It introduces an aspheric lens group to reduce aberrations and improve imaging quality. The built-in temperature control system has a stable working temperature within the range of ±0.1°C to prevent temperature fluctuations from interfering with measurement accuracy. It collects spectral information in batches and reduces the impact of noise interference through signal accumulation enhancement. Step 3: Collect DNA information from water samples. The nanopore sequencer needs to have high-throughput single-molecule sequencing capabilities, read lengths of hundreds of thousands of base pairs, and a sequencing accuracy rate of ≥90%. It needs a built-in temperature control system to keep the sequencing environment constant at 37°C. It uses a fast mode to complete a large number of DNA fragment sequencing tasks within a few hours. It is also equipped with real-time data analysis software to monitor the sequencing progress and quality in real time. Step 4: Prepare the library of the microbial DNA extracted from groundwater using the Nextera XT DNA LibraryPrep Kit to ensure that the library size is uniform and suitable for subsequent sequencing analysis. Before sequencing, the library is quantified and quality checked using the Qubit Fluorometer and Agilent Bioanalyzer instruments to ensure that the library concentration is between 0.2 and 2 nanograms per microliter and the fragment length is mainly distributed between 300 and 700 base pairs. Step 5: Use the TwinNet model to predict the population correlation between the SERS spectrum and the DNA library. First, the original SERS spectrum and the background signal are processed separately through dual-channel input to improve the anti-interference ability. Each channel contains a convolution layer, a pooling layer, and a fully connected layer structure. The convolution kernel sizes are 3×3, 5×5, and 7×7 to extract multi-scale spatial features. The residual connection structure is introduced to avoid the gradient vanishing problem and speed up the training. Batch normalization (BN) technology is applied to enhance the stability of the model. Finally, the final result is output through the Softmax classifier to ensure the prediction accuracy. Step 6, using the improved LSTM network to reversely predict the temporal correlation between the DNA library and the SERS spectrum, first construct an improved LSTM network architecture, each layer contains 128 to 256 hidden units, the activation function uses ReLU to accelerate the convergence process, the introduction of the Dropout mechanism is set to a ratio of 0.2 to 0.5 to prevent overfitting problems, the Adam optimization algorithm is used in the model training process with a learning rate of 0.001 to 0.01, and the batch size is set to 32 to 64 to balance the computational efficiency and model performance; Step 7: Add a reverse transfer path to the standard LSTM to form a bidirectional LSTM structure, so that the model can consider both past and future contextual information, thereby better capturing complex patterns in time series and being suitable for the trend of microbial communities changing over time; Step 8: Combine the pollutant characteristics obtained by SERS detection with the microbial DNA sequences obtained by nanopore sequencing, and use TwinNet and bidirectional LSTM models to comprehensively analyze the correlation between the two sets of data, thereby achieving effective monitoring of the effects of micropollutants in groundwater on microbial communities and timely early warning of groundwater micropollutants due to changes in the DNA library.

2. A method for analyzing groundwater micro-pollution by combining SERS-nanopore sequencing and TwinNet-LSTM according to claim 1, characterized in that: The improved surface enhanced Raman scattering spectrometer acquisition method described in step 2 also involves the introduction of an automatic focusing function, that is, using a laser rangefinder to monitor the liquid level in the sample pool in real time to ensure that the focus position is consistent during each measurement, thereby improving data consistency.

3. A method for analyzing groundwater micro-pollution by combining SERS-nanopore sequencing and TwinNet-LSTM according to claim 1, characterized in that: The nanopore sequencing device described in step 4 also includes error correction algorithms for improving sequencing accuracy, such as Racon and Medaka, which can further correct sequence errors after preliminary assembly to reduce the misjudgment rate.

4. A method for analyzing groundwater micro-pollution by combining SERS-nanopore sequencing and TwinNet-LSTM according to claim 1, characterized in that: The acquisition and preprocessing of the microbial DNA sequence described in step 5 also includes deduplication of the library to remove the repeated sequences generated during the PCR amplification process to ensure the authenticity and reliability of the sequencing results.

5. A method for analyzing groundwater micro-pollution by combining SERS-nanopore sequencing and TwinNet-LSTM according to claim 1, characterized in that: The introduction of bidirectional LSTM described in step 7 improves the ability to model long-term dependencies, and also includes the introduction of an attention mechanism to dynamically adjust the weights of each layer, highlight important feature areas, and suppress noise interference.

Images