Method and system for identifying agricultural runoff pollution in surface water using microbial fingerprinting

Abstract

The application discloses a method and system for identifying farmland runoff pollution in surface water by using microbial fingerprints. Microbial samples of multiple types of water bodies are collected, and microbial composition data are obtained through 16S rDNA sequencing; based on a sensitivity-specificity screening index, combined with specific anaerobic characteristics, a farmland runoff pollution microbial fingerprint is determined, a double-model integrated judgment mechanism is established: an artificial neural network is used to process microbial category data, an XGBoost model is used to analyze microbial relative abundance data, and a control variable method is used to extract the classification decision basis of the double models; and the double models are used to make result judgment based on logical rules. The application innovatively integrates microbial fingerprints and machine learning classification models, can realize accurate and simple identification of farmland runoff pollution in surface water, and improves the identification accuracy and efficiency of farmland runoff pollution.

Description

Technical Field

[0001] This invention belongs to the field of environmental pollution source identification technology, specifically involving a method and system for identifying agricultural runoff pollution in surface water using microbial fingerprints. Background Technology

[0002] With the expansion of agricultural production, agricultural non-point source pollution has become a significant source of water pollution. Compared to orchards, grasslands, and woodlands, intensively managed farmland has a greater impact on surface water quality. The fertilizers and pesticides used in farmland carry pollutants into water bodies through rainfall and other processes, threatening human health and ecological security. Current pollution source identification methods rely on fingerprints such as stable isotopes, water chemical indicators, and characteristic compounds. While accurate in some cases, these methods are limited to specific pollutants and cannot distinguish between farmland pollution and pollution from other agricultural sources. Microbial fingerprinting, however, is a promising alternative due to its sensitivity to environmental changes and ability to reflect complex pollution patterns.

[0003] In recent years, significant progress has been made in water pollution identification based on microbial fingerprints. Existing technologies primarily construct microbial fingerprint databases for specific pollution sources. For example, patent CN117512147A uses 16S rRNA genes to establish species fingerprint profiles for aquaculture and agricultural sources, and CN117535435A proposes a rice paddy pollution-specific gene fingerprint profile. However, three major technical bottlenecks remain in this field: First, traditional fingerprint matching methods rely on the detection of microbial fingerprints, which have biological limitations in terms of host specificity and sensitivity, leading to a high false positive rate. Second, machine learning-based pollution identification focuses on the abundance data of microbial fingerprints while neglecting the importance of microbial category information. This can result in the inability to construct a robust and accurate pollution identification framework due to the neglect of the indicative role of certain fingerprints in identifying pollution sources. Third, although machine learning models excel at discovering correlations and associations to improve prediction accuracy, they often require cumbersome model construction, optimization, and screening, and lack the ability for causal reasoning or statistical interpretation. Especially neural networks and ensemble models, their black-box nature makes decision-making paths untraceable, making it difficult to meet the evidence chain construction requirements in environmental law enforcement. Therefore, there is an urgent need to develop a precise and convenient method for identifying pollution sources that integrates multi-dimensional microbial fingerprint features with interpretable and transparent machine learning. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method and system for identifying agricultural runoff pollution in surface water using microbial fingerprints, so as to improve the identification efficiency and accuracy of agricultural runoff pollution.

[0005] According to one aspect of this application, a method for identifying agricultural runoff pollution in surface water using microbial fingerprinting is disclosed, the method comprising:

[0006] Multiple water samples of different types were obtained, and the composition data of polluting microorganisms for each water sample were obtained based on the 16S rDNA sequencing method. Each water sample corresponds to a water environment.

[0007] Based on the sensitivity-specificity analysis method and combined with the characteristics of obligate anaerobic organisms, multiple microorganisms associated with farmland runoff pollution were screened from the microbial composition data of multiple pollution sources. These microorganisms were used as the microbial fingerprints of farmland runoff pollution, including f_Desulfuromonadaceae, g_Geobacter, f_AKAU3564_sediment_group, o_Dehalococcoidales, and g_Citrifermentans.

[0008] Based on the data of the microbial categories of farmland runoff pollutants corresponding to each of the multiple microbial fingerprints of farmland runoff pollutants and the relative abundance data of the microbial categories of farmland runoff pollutants corresponding to each of the multiple microbial fingerprints of farmland runoff pollutants, an artificial neural network model and an XGBoost model were constructed respectively.

[0009] The target data of the target microbial fingerprint in the target surface water sample to be identified is obtained, and the target data is input into the artificial neural network model and the XGBoost model. The target data includes the target microbial fingerprint category data corresponding to the target microbial fingerprint and the target microbial fingerprint relative abundance data corresponding to the target microorganism.

[0010] Based on the judgment results output by the artificial neural network model and the XGBoost model, logical rules are used to determine whether there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint target data.

[0011] In some embodiments, the method further includes:

[0012] By changing the microbial fingerprint category data of farmland runoff pollution input to the artificial neural network model using the control variable method, multiple first output results are obtained;

[0013] Based on multiple first output results, determine the binarization judgment condition that satisfies the positive judgment result output by the artificial neural network model;

[0014] By changing the relative abundance data of the microbial fingerprint of farmland runoff pollution input to the XGBoost model using the control variable method, multiple second output results are obtained.

[0015] Based on the multiple second output results, a relative abundance threshold range that satisfies the positive judgment result output by the XGBoost model is determined.

[0016] In some embodiments, based on multiple first output results, determining the binarization criteria that satisfy the positive judgment result output by the artificial neural network model includes:

[0017] The microbial fingerprint category data of farmland runoff pollution is binarized to obtain a binarized 0 / 1 matrix, where 0 represents non-existence and 1 represents existence.

[0018] Based on the binarized 0 / 1 matrix, the analysis of single farmland runoff pollution microbial fingerprints and / or combinations of multiple farmland runoff pollution microbial fingerprints is performed. The binarization criteria are determined as follows:

[0019] When the presence of the g_Citrifermentans microbial fingerprint is detected, the extraction result is 1, indicating that the binarization criteria are met; or,

[0020] When any two or more microbial fingerprints other than the combination of f_Desulfuromonadaceae and g_Geobacter are detected, the extraction result is 1, indicating that the binarization judgment condition is met.

[0021] In some embodiments, determining the relative abundance threshold range that satisfies the positive judgment result output by the XGBoost model based on the plurality of second output results includes:

[0022] The relative abundance data of microbial fingerprints of farmland runoff pollution were adjusted with a precision gradient of 0.0000001% to determine the relative abundance threshold range, which includes:

[0023] If only a single microbial fingerprint of farmland runoff is detected, the relative abundance threshold range must meet the following conditions: 0.0067339% ≥ f_AKAU3564_sediment_group ≥ 0.0050275%, f_Desulfuromonadaceae ≥ 0.0177225%, 0.0397715% ≥ g_Geobacter ≥ 0.0077625%, 0.0078340% ≥ o_Dehalococcoidales ≥ 0.0031500% or 0.0105874% ≥ o_Dehalococcoidales ≥ 0.0091936%;

[0024] If both f_Desulfuromonadaceae and g_Geobacter are detected, the relative abundance threshold range must satisfy: f_Desulfuromonadaceae ≥ 0.0177225% or f_Desulfuromonadaceae < 0.0177225% and 0.0397719% ≥ g_Geobacter ≥ 0.0077645%.

[0025] In some embodiments, determining whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data using logical rules based on the judgment results output by the artificial neural network model and the XGBoost model includes:

[0026] Match whether the target microbial fingerprint category data meets the binarization determination criteria;

[0027] When the target microbial fingerprint category data meets the binarization judgment condition, and the judgment result output by the artificial neural network model is determined to be a positive judgment result, it is determined that there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint category data.

[0028] In some embodiments, determining whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data using logical rules based on the judgment results output by the artificial neural network model and the XGBoost model includes:

[0029] When the judgment result output by the artificial neural network model is a negative judgment result;

[0030] Obtain the relative abundance data of the target microbial fingerprint in the target surface water sample;

[0031] Whether the relative abundance data of the target microbial fingerprint meets the relative abundance threshold range;

[0032] When the relative abundance data of the target microbial fingerprint meets the relative abundance threshold range, the judgment result output by the XGBoost model is determined to be a positive judgment result, and it is determined that there is farmland runoff pollution in the target surface water sample corresponding to the relative abundance data of the target microbial fingerprint.

[0033] In some embodiments, determining whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data using logical rules based on the judgment results output by the artificial neural network model and the XGBoost model includes:

[0034] Match whether the target microbial fingerprint category data meets the binarization determination criteria;

[0035] Check whether the relative abundance data of the target microbial fingerprint in the target surface water sample meets the relative abundance threshold range;

[0036] When the target microbial fingerprint category data meets the binarization determination condition and / or the relative abundance data of the target microbial fingerprint meets the relative abundance threshold range, it is determined whether there is farmland pollution in the target surface water sample corresponding to the target microbial fingerprint target data.

[0037] According to another aspect of this application, a system for identifying agricultural runoff pollution in surface water using microbial fingerprinting is also disclosed, the system comprising:

[0038] The water sample acquisition module is used to acquire multiple water samples of different types and obtain the microbial composition data of each water sample based on the 16S rDNA sequencing method. Each water sample corresponds to a water environment.

[0039] The microbial fingerprinting module for farmland runoff pollution is used to screen multiple farmland runoff pollution-associated microorganisms from multiple pollution source microbial composition data based on sensitivity-specificity analysis methods and specific anaerobic characteristics. The farmland runoff pollution-associated microorganisms are used as farmland runoff pollution microbial fingerprints, which include f_Desulfuromonadaceae, g_Geobacter, f_AKAU3564_sediment_group, o_Dehalococcoidales, and g_Citrifermentans.

[0040] The model building module is used to construct artificial neural network models and XGBoost models based on the data of the categories of microbial fingerprints of multiple microbial fingerprints of agricultural runoff and the relative abundance data of microbial fingerprints of multiple microbial fingerprints of agricultural runoff.

[0041] The data acquisition module is used to acquire target data of target microbial fingerprints in the target surface water sample to be identified, and input the target data into the artificial neural network model and the XGBoost model. The target data includes target microbial fingerprint category data corresponding to the target microbial fingerprint and target microbial fingerprint relative abundance data corresponding to the target microbial fingerprint.

[0042] The farmland runoff determination module is used to determine, based on the judgment results output by the artificial neural network model and the XGBoost model, whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data using logical rules.

[0043] The present invention includes, but is not limited to, the following beneficial effects: (1) The present invention utilizes 16S (1) rDNA sequencing was used to determine the microbial composition of samples from various types of pollution sources. Combined with sensitivity-specificity analysis and specific anaerobic characteristics, multiple microorganisms associated with agricultural runoff pollution were screened out. Based on the microorganisms associated with agricultural runoff pollution, a microbial fingerprint for identifying agricultural runoff pollution was constructed, filling the gap in the field of agricultural runoff pollution identification in surface water where there is a lack of specific microbial indicators. (2) This invention provides a simple, model-free method for identifying agricultural runoff pollution in surface water using microbial fingerprints. It breaks through the dependence of pollution source identification on machine learning algorithm modeling. By replacing complex data analysis with explicit judgment rules, it simplifies the pollution source identification process and improves the operability of the pollution identification method. (3) This invention proposes a dual-dimensional fusion criterion for microbial fingerprints to improve the robustness of pollution identification. By jointly analyzing the category and relative abundance threshold of microbial fingerprints, it effectively overcomes the limitation that a single data model cannot fully capture the complex information of the microbiome. (4) The technology of this invention has strong scalability. The pollution source identification framework (microbial fingerprint screening + dual-dimensional criterion) constructed can be transferred to other pollution source identification scenarios such as domestic sewage and industrial wastewater. It has high application value in environmental pollution source analysis. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0045] Figure 1 This is a flowchart of a method for identifying farmland runoff pollution in surface water using microbial fingerprints, according to an embodiment of this application.

[0046] Figure 2 This is another flowchart of a method for identifying farmland runoff pollution in surface water using microbial fingerprints, as described in this application.

[0047] Figure 3 This is another flowchart of a method for identifying farmland runoff pollution in surface water using microbial fingerprints, as described in this application.

[0048] Figure 4 This is another flowchart of a method for identifying farmland runoff pollution in surface water using microbial fingerprints, as described in this application.

[0049] Figure 5 This is another flowchart of a method for identifying farmland runoff pollution in surface water using microbial fingerprints, as described in this application.

[0050] Figure 6 This is a structural block diagram of a system for identifying farmland runoff pollution in surface water using microbial fingerprints, according to an embodiment of this application.

[0051] Figure 7 This is an NMDS dimensionality reduction grouping diagram of various pollution sources based on Bray-Curtis distance in an embodiment of the present invention. The left side represents pollution source categories that can be significantly divided into agricultural runoff pollution and domestic sewage, the middle side represents pollution source categories that can be significantly divided into agricultural runoff pollution and aquaculture wastewater, and the right side represents pollution source categories that can be significantly divided into agricultural runoff pollution and orchard runoff pollution.

[0052] Figure 8 These are the balanced accuracy and F1 score of the category data model based on microbial fingerprints in this embodiment of the invention, where the left side is the balanced accuracy and the right side is the F1 score.

[0053] Figure 9 These are the balanced accuracy and F1 score of the relative abundance data model based on microbial fingerprints in this embodiment of the invention, where the left side is the balanced accuracy and the right side is the F1 score.

[0054] Figure 10 This is a performance comparison of model integration, single model, and traditional methods on external test set 2 for embodiments of the present invention;

[0055] Figure 11 This is the detection of microbial fingerprints of farmland runoff pollution in surface water samples according to an embodiment of the present invention. The left side is a seawater sample and the right side is a river water sample. Des, Geo, AKA, Deh and Cit are abbreviations for f_Desulfuromonadaceae, g_Geobacter, f_AKAU3564_sediment_group, o_Dehalococcoidales and g_Citrifermentans, respectively. Detailed Implementation

[0056] This invention discloses a method for identifying agricultural runoff pollution in surface water using microbial fingerprinting. The method includes: acquiring multiple water samples of different types, and obtaining microbial composition data for each water sample based on 16S rDNA sequencing, with each water sample corresponding to a specific aquatic environment; screening multiple agricultural runoff pollution-associated microorganisms from the microbial composition data of multiple pollution sources based on sensitivity-specificity analysis and specific anaerobic characteristics, and using these agricultural runoff pollution-associated microorganisms as agricultural runoff pollution microbial fingerprints, including f_Desulfuromonadaceae, g_Geobacter, f_AKAU3564_sediment_group, o_Dehalococcoidales, and g_Citrifermentans; and analyzing the agricultural runoff corresponding to each of the multiple agricultural runoff pollution microbial fingerprints. This application constructs artificial neural network (ANN) and XGBoost models for the identification of agricultural runoff pollution, using data on the fingerprint categories of polluting microorganisms and the relative abundance data of each fingerprint. Target data of the target microorganism fingerprints in the target surface water sample to be identified are obtained and input into the ANN and XGBoost models. The target data includes the target microorganism fingerprint category data and the relative abundance data of the target microorganism fingerprints. Based on the judgment results output by the ANN and XGBoost models, logical rules are used to determine whether agricultural runoff pollution exists in the target surface water sample corresponding to the target microorganism fingerprint target data. This application improves the identification efficiency and accuracy of agricultural runoff pollution.

[0057] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0058] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Figure 1 A flowchart illustrating a method for identifying agricultural runoff pollution in surface water using microbial fingerprinting, as shown below. Figure 1 As shown, it includes the following steps:

[0059] S100: Obtain multiple water samples of different types and acquire microbial composition data for each water sample based on 16S rDNA sequencing.

[0060] Each water sample corresponds to a specific aquatic environment.

[0061] Specifically, in one instance, this step can be implemented based on the following steps:

[0062] (1) Water sample collection: 158 farmland runoff samples, 37 garden runoff samples, 17 domestic sewage samples, 174 aquaculture tailwater samples and 186 surface water samples were collected in 1L, stored at 4℃ and transported back to the laboratory, and processed within 24 hours.

[0063] (2) Water sample pretreatment: The water samples were filtered using a 0.45 μm glass fiber membrane. After filtration, the membrane was removed and placed in a 2 mL centrifuge tube and stored at -80°C. The 16S rDNA high-throughput sequencing experiment was performed on the Illumina platform of Hangzhou Lianchuan Biotechnology Co., Ltd.

[0064] (3) Total microbial DNA extraction: The CTAB method was selected to extract total microbial DNA from samples from various sources. The quality of DNA extraction was detected by agarose gel electrophoresis, and the DNA was quantified by ultraviolet spectrophotometer.

[0065] (4) PCR amplification and sequencing: Using the DNA extracted from each sample as a template, the V3-V4 region of the 16S rDNA gene was amplified. The sequencing primer sequences were as follows:

[0066] Pre-primer: 341F (5'-CCTACGGGNGGCWGCAG-3');

[0067] Back primer: 805R (5'-GACTACHVGGGTATCTAATCC-3').

[0068] The specific reaction system and conditions are shown in Tables 1 and 2:

[0069] Table 1 PCR reaction system

[0070] PCR reaction components PCR reaction volume PhusionHotstartflex2XMasterMix 12.5μL ForwardPrimer 2.5μL ReversePrimer 2.5μL TemplateDNA 20ng <![CDATA[ddH2O]]> 25μL

[0071] Table 2 PCR reaction conditions

[0072] PCR reaction temperature PCR reaction time Cycle number 98℃ 30s 98℃ 10s 54℃ 30s 35 cycles 72℃ 45s 72℃ 10min

[0073]

[0074] The purified and quality-checked PCR products were mixed in the same proportion and sequenced on the Illumina NovaSeq platform.

[0075] (5) ASV Clustering and Species Annotation: The obtained microbial gene sequences were clustered into ASVs and their abundance was determined based on 100% similarity. The abundance results were then normalized. To obtain the species classification information corresponding to each ASV, the SILVA and NT-16s databases were used for species classification. To ensure the accuracy of the classification annotation results, the annotation confidence level was set to be greater than 0.7. Based on the ASV annotation results and the ASV abundance table of each sample, the relative abundance of species at the kingdom, phylum, class, order, family, genus, and species levels was obtained.

[0076] S102. Based on the sensitivity-specificity analysis method and combined with the characteristics of obligate anaerobic organisms, multiple microorganisms associated with farmland runoff pollution were screened from the microbial composition data of multiple pollution sources, and the microorganisms associated with farmland runoff pollution were used as the microbial fingerprints of farmland runoff pollution.

[0077] In one example, the microbial fingerprint of farmland runoff pollution includes f_Desulfuromonadaceae, g_Geobacter, f_AKAU3564_sediment_group, o_Dehalococcoidales, and g_Citrifermentans.

[0078] Specifically, in conjunction with the above steps (1) to (5), step (6) is continued to determine the microbial fingerprint.

[0079] (6) Based on the count values ​​of ASVs, the Bray-Curtis distance was calculated using the `vegdist` function in R. Non-metric multidimensional scaling (NMDS) was then used to perform dimensionality reduction and grouping of pollution source samples based on the Bray-Curtis distance. The results are as follows: Figure 7 As shown, the results indicate that the pollution source samples can be significantly divided into agricultural runoff, aquaculture wastewater, garden runoff, and domestic sewage. Ten specific microorganisms for agricultural runoff were screened using source sensitivity (≥0.5) and source specificity (≥0.8), as shown in Table 3. Based on obligate anaerobic characteristics, five specific anaerobic microorganisms were identified as microbial fingerprints for agricultural runoff pollution, including f_Desulfuromonadaceae, g_Geobacter, f_AKAU3564_sediment_group, o_Dehalococcoidales, and g_Citrifermentans.

[0080] Table 3. Sensitivity and specificity of specific microorganisms in farmland pollution.

[0081]

[0082] S104. Based on the data of the microbial fingerprint categories of multiple microbial fingerprints of agricultural runoff and the relative abundance data of microbial fingerprints of multiple microbial fingerprints of agricultural runoff, an artificial neural network model and an XGBoost model are constructed respectively.

[0083] Specifically, the model construction can refer to the following steps:

[0084] (a) Configuration of simulated pollution sinks: Based on the hydrological parameters of the target area (annual average rainfall, farmland irrigation volume, effective utilization coefficient of farmland irrigation water, surface water resources, and farmland area), the proportion of farmland runoff pollution was calculated to be 0%–20% of the surface water resources. Farmland runoff pollution was mixed with other pollution sources (orchard runoff, domestic sewage, and aquaculture wastewater) at proportional gradients (0%, 5%, 10%, 20%) to prepare 104 simulated pollution sinks. When constructing the composite simulation system, surface water samples without detectable microbial fingerprints of farmland runoff pollution were selected as background water samples and mixed with water samples from each pollution source at preset proportions (total volume 1L) to construct a composite simulation system including single, binary, ternary, and quaternary pollution sources.

[0085] (b) Data Preparation: Microbial fingerprint data from all source samples (n=386) were used as the basic dataset. The categorical and relative abundance data of the microbial fingerprints were used as input features to the model, with pollution source type as the output variable. In the categorical data, detection of a microbial fingerprint was denoted as 1, and non-detection as 0. The output variable was also binary: 0 indicated the absence of agricultural runoff pollution, and 1 indicated the presence of agricultural runoff pollution. The basic dataset was divided into a training set (n=272, containing 108 positive samples and 164 negative samples) and a test set 1 (n=114, containing 45 positive samples and 69 negative samples) in a 7:3 ratio. Microbial fingerprint data from 104 simulated pollution sinks in agricultural runoff pollution were used as an external test set 2 (n=104, including 79 positive samples and 25 negative samples) to further evaluate the model's generalization ability, universality, and stability across different datasets.

[0086] (c) Model Construction: Seven classification models were constructed using R, including Logistic Regression (LR), Support Vector Machine (SVM), Naive Bayes (NB), Artificial Neural Network (ANN), Random Forest (RF), XGBoost, and K-Nearest Neighbors (KNN). To evaluate the robustness and randomness of the models, 15 random seeds were created, generating 15 different dataset partitions for independent computation. Before training each model, a grid search method was used for hyperparameter optimization with 10x cross-validation. In this example, the Artificial Neural Network (ANN) model and the XGBoost model are used as examples. The ANN model forms a complex network structure through a large number of interconnected neurons. During training, two classes of microbial fingerprint data are input, and the model continuously adjusts the connection weights between neurons. In this process, the ANN model learns the intrinsic relationship between microbial fingerprint data and farmland runoff pollution, such as the correlation pattern between specific microbial combinations and their abundance levels and farmland runoff pollution. XGBoost Model: As a gradient boosting tree algorithm, XGBoost constructs a series of decision trees iteratively. Each iteration focuses on samples that the previous model predicted incorrectly, continuously optimizing the model. During training, it constructs a decision tree based on microbial fingerprint data to mine the features and patterns in the data in order to accurately predict the pollution of farmland runoff.

[0087] (d) Model Evaluation: The classification models were evaluated using four metrics: Balanced Accuracy, Sensitivity, Specificity, Precision, and F1 score. To determine the optimal contamination identification model and reduce redundancy in the evaluation metrics, Balanced Accuracy and F1 score were used as the final screening metrics. The results are as follows: Figure 8 and Figure 9 As shown in the figure. The results show that the ANN model is the best performing model for classification based on microbial fingerprint category data, with a balanced accuracy of 0.8133±0.0006 and an F1 score of 0.7891±0.0062, respectively; the XGBoost model is the best performing model for classification based on microbial fingerprint relative abundance data, with a balanced accuracy of 0.8261±0.0029 and an F1 score of 0.8105±0.0057, respectively.

[0088] (e) Model Ensemble: To ensure the representativeness of the dataset, we used the basic dataset (n=386) as the training set and the external test set 2 (n=104) as the test set. After hyperparameter optimization, the parameter settings are as follows: for the ANN model, size, decay, maxit, and abstol are set to 1, 0.001, 500, and 0.001, respectively; for the XGBoost model, max_depth, eta, nrounds, gamma, subsample, and colsample_bytree are set to 6, 0.01, 172, 0, 1, and 1, respectively. A logical "OR" rule was used to ensemble the ANN and XGBoost models; an output of 1 from either model was considered as agricultural runoff pollution. We compared the performance of the model ensemble with that of a single model and traditional methods (where the detection of microbial fingerprints was considered as the presence of agricultural runoff pollution) on the external test set 2, such as... Figure 10 As shown in the figure, the performance of the ensembled models was significantly improved, with the balanced accuracy and F1 score reaching 0.8661 and 0.8522, respectively. Compared with the highest values ​​of the single models, these represent improvements of 2.24% and 2.97%, respectively, and improvements of 17.97% and 14.28%, respectively, compared with the traditional methods.

[0089] S106. Obtain target data of the target microbial fingerprint in the target surface water sample to be identified, and input the target data into the artificial neural network model and the XGBoost model.

[0090] The target data includes the target microbial fingerprint category data corresponding to the target microbial fingerprint and the target microbial fingerprint relative abundance data corresponding to the target microbial fingerprint.

[0091] S108. Based on the judgment results output by the artificial neural network model and the XGBoost model, logical rules are used to determine whether there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint target data.

[0092] Understandably, this embodiment utilizes 16S rDNA sequencing to determine the microbial composition of samples from various pollution sources. Combining sensitivity-specificity analysis and specific anaerobic characteristics, it screens out multiple microorganisms associated with agricultural runoff pollution. Based on these microorganisms, it constructs a microbial fingerprint for identifying agricultural runoff pollution, filling the gap in the field of surface water agricultural runoff pollution identification due to the lack of specific microbial indicators. Furthermore, this simple, model-free method for identifying surface water agricultural runoff pollution using microbial fingerprints overcomes the dependence of pollution source identification on machine learning algorithm modeling. By replacing complex data analysis with explicit judgment rules, it simplifies the pollution source identification process and improves the operability of the pollution identification method. It proposes a two-dimensional fusion criterion for microbial fingerprints to enhance the robustness of pollution identification. By jointly analyzing the category and relative abundance threshold of microbial fingerprints, it effectively overcomes the limitation of a single data model in comprehensively capturing the complex information of the microbiome.

[0093] Furthermore, Figure 2 This is another flowchart illustrating the method for identifying farmland runoff pollution in surface water using microbial fingerprinting, as described in this application. Figure 2 As shown, it includes the following steps:

[0094] S200. Based on the control variable method, the microbial fingerprint category data of farmland runoff pollution input into the artificial neural network model is changed to obtain multiple first output results.

[0095] S202. Based on multiple first output results, determine the binarization judgment conditions that satisfy the positive judgment result output by the artificial neural network model.

[0096] S204. Based on the control variable method, the relative abundance data of microbial fingerprints of farmland runoff pollution input to the XGBoost model are changed to obtain multiple second output results.

[0097] S206. Based on multiple second output results, determine the relative abundance threshold range that satisfies the positive judgment result output by the XGBoost model.

[0098] Specifically, for the Artificial Neural Network (ANN) model: using the controlled variable method, each time other variables are fixed, only one microbial fingerprint category data of farmland runoff pollution is changed (set to exist or not exist), and the changes in the model output are observed. Through multiple experiments, it is determined under which combinations of microbial fingerprint categories exist or do not exist that the ANN model will output a positive judgment (i.e., a judgment of farmland runoff pollution), thus determining the binary judgment condition that satisfies the positive judgment result output of the artificial neural network model. Here, "binarization" is reflected in the fact that the microbial fingerprint category only has two states: "exist" (which can be regarded as 1) and "not exist" (which can be regarded as 0).

[0099] In one example, suppose there are five microbial fingerprints for farmland pollution, labeled A, B, C, D, and E. Experiments using the controlled variable method show that when microbial fingerprints A, B, and C are present simultaneously, and D and E are absent, the ANN model outputs a positive result, identifying the sample as having farmland pollution. Therefore, "A, B, and C are present, and D and E are absent" constitutes a binary judgment condition. In reality, there will be multiple such combinations of conditions, collectively forming the basis for the ANN model to determine farmland runoff pollution. These conditions provide clear rules for subsequent use of the ANN model to identify farmland runoff pollution in surface water samples.

[0100] For the XGBoost model, the controlled variable method is also used, fixing the category data of microbial fingerprints while changing their relative abundance data. By gradually adjusting the relative abundance values ​​and observing the model output, it is determined which intervals the XGBoost model will output a positive result for the relative abundance of microbial fingerprints, thus obtaining the relative abundance threshold intervals.

[0101] Furthermore, in one example, for the ANN model, whether it's a single microbial fingerprint of farmland runoff pollution or a combination of multiple microbial fingerprints, the data for the target microbial fingerprint is set to 1 to indicate its detection, while the data for other fingerprints are uniformly set to 0 to indicate no detection. Then, the impact of each fingerprint and fingerprint combination on the model's prediction is discussed sequentially. In one example, to address the limitations of the ANN model, the XGBoost model is configured as follows: when a specific single fingerprint or fingerprint combination is detected but not identified as farmland pollution by the ANN model, the relative abundance range for which the XGBoost model identifies farmland runoff pollution is determined by changing the relative abundance of the fingerprint. The relative abundance range can be set from 0.002% to 100%. To minimize error, the relative abundance data is accurate to seven decimal places, with the data for two adjacent samples increasing by 0.0000001% sequentially. The results show that for the ANN model, detecting g_Citrifermentans or any two or more microbial fingerprints other than the combination of f_Desulfuromonadaceae and g_Geobacter is considered as evidence of farmland runoff pollution. For the XGBoost model, if only a single microbial fingerprint is detected, farmland runoff pollution is identified if any of the following conditions are met: 0.0067339% ≥ f_AKAU3564_sediment_group ≥ 0.0050275%, f_Desulfuromonadaceae ≥ 0.0177225%, 0.0397715% ≥ g_Geobacter ≥ 0.0077625%, 0.0078340% ≥ o_Dehalococcoidales ≥ 0.0031500% or 0.0105874% ≥ o_Dehalococcoidales ≥ 0.0091936%; If both f_Desulfuromonadaceae and g_Geobacter are detected, then agricultural runoff pollution will be identified if either of the following conditions is met: f_Desulfuromonadaceae ≥ 0.0177225% or f_Desulfuromonadaceae < 0.0177225% and 0.0397719% ≥ g_Geobacter ≥ 0.0077645%.

[0102] Through explicit binarization and specific judgment conditions, the ANN model can quickly analyze and judge samples without complex calculations, improving the efficiency of identifying farmland runoff pollution. In actual environmental monitoring, it can quickly provide preliminary judgment results, saving time for subsequent processing. The judgment conditions are constructed from two dimensions: single fingerprint and multi-fingerprint combination. This approach considers both the indicative role of key individual microorganisms and the overall characteristics of the microbial community, enabling comprehensive judgment from multiple perspectives, reducing false positives and false negatives, and improving the accuracy and reliability of the model in identifying farmland pollution.

[0103] In one example, Figure 3 This is another flowchart of the method for identifying farmland runoff pollution in surface water using microbial fingerprinting, as described in this application. Figure 3 As shown, it includes the following steps:

[0104] S300. Whether the target microbial fingerprint category data meets the binarization judgment condition.

[0105] S302. When the target microbial fingerprint category data meets the binarization judgment condition, and the judgment result output by the artificial neural network model is determined to be a positive judgment result, it is determined that there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint category data.

[0106] In one example, Figure 4 This is another flowchart of the method for identifying farmland runoff pollution in surface water using microbial fingerprinting, as described in this application. Figure 4 As shown, it includes the following steps:

[0107] S400. When the judgment result output by the artificial neural network model is negative, obtain the relative abundance data of the target microbial fingerprint in the target surface water sample.

[0108] S402. Does the relative abundance data of the target microbial fingerprint meet the relative abundance threshold range?

[0109] S404. When the relative abundance data of the target microbial fingerprint meets the relative abundance threshold range, the judgment result output by the XGBoost model is determined to be a positive judgment result, and it is determined that there is farmland runoff pollution in the target surface water sample corresponding to the relative abundance data of the target microbial fingerprint.

[0110] In one example, Figure 5 This is another flowchart of the method for identifying farmland runoff pollution in surface water using microbial fingerprinting, as described in this application. Figure 5 As shown, it includes the following steps:

[0111] S500, whether the target microbial fingerprint category data meets the binarization judgment condition.

[0112] S502. Match whether the relative abundance data of the target microorganism fingerprint in the target surface water sample meets the relative abundance threshold range.

[0113] S504. When the target microbial fingerprint category data meets the binarization judgment condition and / or the target microbial fingerprint relative abundance data meets the relative abundance threshold range, logical rules are used to determine whether there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint target data.

[0114] Based on this example method, in one example, the target data of the target microbial fingerprint in the target surface water sample to be identified is as follows: Figure 11 As shown, the results indicate that 47 out of 186 surface water samples collected tested positive for microbial fingerprints indicating farmland runoff pollution.

[0115] The target data of the target microbial fingerprints in the surface water samples to be identified were compared with the judgment criteria, and the results are shown in Table 4. The results show that 32 out of the 47 surface water samples with detected microbial fingerprints were identified as having agricultural runoff pollution.

[0116] Table 4. Identification results of agricultural runoff pollution in surface water. 1 represents the presence of agricultural runoff pollution, and 0 represents the absence of agricultural runoff pollution.

[0117]

[0118] Furthermore, Figure 6 A structural diagram of a system for identifying agricultural runoff pollution in surface water using microbial fingerprinting, such as... Figure 6 As shown, the system includes:

[0119] The water sample acquisition module is used to acquire multiple water samples of different types and obtain the microbial composition data of each water sample based on the 16S rDNA sequencing method. Each water sample corresponds to a water environment.

[0120] The microbial fingerprinting module for farmland runoff pollution is used to screen multiple farmland runoff pollution-associated microorganisms from multiple pollution source microbial composition data based on sensitivity-specificity analysis methods and specific anaerobic characteristics. These farmland runoff pollution-associated microorganisms are used as farmland runoff pollution microbial fingerprints, which include f_Desulfuromonadaceae, g_Geobacter, f_AKAU3564_sediment_group, o_Dehalococcoidales, and g_Citrifermentans.

[0121] The model building module is used to construct artificial neural network models and XGBoost models based on the data of the categories of microbial fingerprints of multiple microbial fingerprints of agricultural runoff and the relative abundance data of microbial fingerprints of multiple microbial fingerprints of agricultural runoff.

[0122] The data acquisition module is used to acquire target data of target microbial fingerprints in the target surface water samples to be identified, and input the target data into the artificial neural network model and the XGBoost model. The target data includes the target microbial fingerprint category data corresponding to the target microbial fingerprint and the relative abundance data of the target microbial fingerprint corresponding to the target microbial fingerprint.

[0123] The farmland runoff pollution determination module is used to determine whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data based on the judgment results output by the artificial neural network model and the XGBoost model, using logical rules.

[0124] The application of the relevant modules of the system in this example can be found in the above introduction to the principles of the method, and will not be repeated here.

[0125] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying agricultural runoff pollution in surface water using microbial fingerprinting, characterized in that, The method includes: Multiple water samples of different types were obtained, and the composition data of pollutant microorganisms for each water sample were obtained based on the 16S rDNA sequencing method. Each water sample corresponds to a water environment. Based on sensitivity-specificity analysis and combined with obligate anaerobic characteristics, multiple microorganisms associated with farmland runoff pollution were screened from the microbial composition data of multiple pollution sources. These farmland runoff pollution-associated microorganisms were used as the microbial fingerprint of farmland runoff pollution. The microbial fingerprint of farmland runoff pollution includes... f_Desulfuromonadaceae , g_Geobacter , f_ AKAU3564_sediment_group , o_Dehalococcoidales and g_Citrifermentans ; Based on the data of the microbial fingerprint categories of multiple microbial fingerprints of agricultural runoff and the relative abundance data of microbial fingerprints of multiple microbial fingerprints of agricultural runoff, an artificial neural network model and an XGBoost model were constructed respectively. The target data of the target microbial fingerprint in the target surface water sample to be identified is obtained, and the target data is input into the artificial neural network model and the XGBoost model. The target data includes the target microbial fingerprint category data corresponding to the target microbial fingerprint and the target microbial fingerprint relative abundance data corresponding to the target microbial fingerprint. Based on the judgment results output by the artificial neural network model and the XGBoost model respectively, logical rules are used to determine whether there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint target data. By changing the microbial fingerprint category data of farmland runoff pollution input to the artificial neural network model using the control variable method, multiple first output results are obtained; Based on multiple first output results, determine the binarization judgment condition that satisfies the positive judgment result output by the artificial neural network model; By changing the relative abundance data of the microbial fingerprint of farmland runoff pollution input to the XGBoost model using the control variable method, multiple second output results are obtained. Based on the multiple second output results, a relative abundance threshold range that satisfies the positive judgment result output by the XGBoost model is determined. Among them, based on multiple first output results, the binary judgment conditions that satisfy the positive judgment result output by the artificial neural network model include: The microbial fingerprint category data of farmland runoff pollution is binarized to obtain a binarized 0 / 1 matrix, where 0 represents non-existence and 1 represents existence. Based on the binarized 0 / 1 matrix, the analysis of single farmland runoff pollution microbial fingerprints and / or combinations of multiple farmland runoff pollution microbial fingerprints is performed. The binarization criteria are determined as follows: In detected g_Citrifermentans When microbial fingerprints are present, the extraction result is 1, indicating that the binarization criteria are met; or, Simultaneously detected f_Desulfuromonadaceae and g_Geobacter When any two or more microbial fingerprints other than the combination are present, the extraction result is 1, indicating that the binarization judgment condition is met.

2. The method for identifying farmland runoff pollution in surface water using microbial fingerprinting according to claim 1, characterized in that, The determination of the relative abundance threshold range that satisfies the positive judgment result output by the XGBoost model based on multiple second output results includes: The relative abundance data of microbial fingerprints of farmland runoff pollution were adjusted with a precision gradient of 0.0000001% to determine the relative abundance threshold range, which includes: If only a single microbial fingerprint of farmland runoff pollution is detected, the relative abundance threshold range must meet the requirement of 0.0067339% ≥ f_ AKAU3564_sediment_group ≥0.0050275%, f_Desulfuromonadaceae ≥0.0177225%, 0.0397715%≥ g_Geobacter ≥0.0077625%, 0.0078340%≥ o_Dehalococcoidales ≥0.0031500% or 0.0105874% ≥ o_Dehalococcoidales ≥0.0091936%; If both are detected f_Desulfuromonadaceae and g_Geobacter When this is the case, the relative abundance threshold range must satisfy: f_Desulfuromonadaceae ≥0.0177225% or f_Desulfuromonadaceae <0.0177225% and 0.0397719% ≥ g_Geobacter ≥0.0077645%.

3. The method for identifying farmland runoff pollution in surface water using microbial fingerprinting according to claim 1, characterized in that, The determination of whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint category data, based on the judgment results output by the artificial neural network model and the XGBoost model respectively, using logical rules, includes: Match whether the target microbial fingerprint category data meets the binarization determination criteria; When the target microbial fingerprint category data meets the binarization judgment condition, and the judgment result output by the artificial neural network model is determined to be a positive judgment result, it is determined that there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint category data.

4. The method for identifying farmland runoff pollution in surface water using microbial fingerprinting according to claim 1, characterized in that, The determination of whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data, based on the judgment results output by the artificial neural network model and the XGBoost model respectively, using logical rules, includes: When the judgment result output by the artificial neural network model is a negative judgment result, the relative abundance data of the target microbial fingerprint in the target surface water sample is obtained; Whether the relative abundance data of the target microbial fingerprint meets the relative abundance threshold range; When the relative abundance data of the target microbial fingerprint meets the relative abundance threshold range, the judgment result output by the XGBoost model is determined to be a positive judgment result, and it is determined that there is farmland runoff pollution in the target surface water sample corresponding to the relative abundance data of the target microbial fingerprint.

5. The method for identifying farmland runoff pollution in surface water using microbial fingerprinting according to claim 1, characterized in that, The determination of whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data, based on the judgment results output by the artificial neural network model and the XGBoost model respectively, using logical rules, includes: Match whether the target microbial fingerprint category data meets the binarization determination criteria; Check whether the relative abundance data of the target microbial fingerprint in the target surface water sample meets the relative abundance threshold range; When the target microbial fingerprint category data meets the binarization determination condition and / or the relative abundance data of the target microbial fingerprint meets the relative abundance threshold range, it is determined whether there is farmland runoff pollution in the target surface water sample corresponding to the target microbial fingerprint target data.

6. A system for identifying agricultural runoff pollution in surface water using microbial fingerprinting, characterized in that, The system includes: The water sample acquisition module is used to acquire multiple water samples of different types and obtain the microbial composition data of each water sample based on the 16S rDNA sequencing method. Each water sample corresponds to a water environment. A module for determining the microbial fingerprint of farmland runoff pollution is used to screen multiple farmland runoff pollution-associated microorganisms from microbial composition data of multiple pollution sources based on sensitivity-specificity analysis methods and specific anaerobic characteristics. These farmland runoff pollution-associated microorganisms are then used as the farmland runoff pollution microbial fingerprint. The farmland runoff pollution microbial fingerprint includes... f_ Desulfuromonadaceae , g_Geobacter , f_AKAU3564_sediment_group , o_Dehalococcoidales and g_Citrifermentans ; The model building module is used to construct artificial neural network models and XGBoost models based on the data of the categories of microbial fingerprints of multiple microbial fingerprints of agricultural runoff and the relative abundance data of agricultural runoff pollution corresponding to each of the multiple microbial fingerprints of agricultural runoff. The data acquisition module is used to acquire target data of the target microbial fingerprint in the target surface water sample to be identified, and input the target data into the artificial neural network model and the XGBoost model. The target data includes the target microbial fingerprint category data corresponding to the target microorganism and the target microbial relative abundance data corresponding to the target microorganism. The farmland runoff pollution determination module is used to determine whether farmland runoff pollution exists in the target surface water sample corresponding to the target microbial fingerprint target data based on the judgment results output by the artificial neural network model and the XGBoost model, using logical rules. The first output result determination module is used to change the microbial fingerprint category data of farmland runoff pollution input to the artificial neural network model based on the control variable method to obtain multiple first output results; The binarization determination condition module is used to determine, based on multiple first output results, the binarization determination conditions that satisfy the positive determination result output by the artificial neural network model. The second output result determination module is used to change the relative abundance data of the microbial fingerprint of farmland runoff pollution input to the XGBoost model based on the control variable method to obtain multiple second output results; The relative abundance threshold interval determination module is used to determine the relative abundance threshold interval that satisfies the positive judgment result output by the XGBoost model based on the multiple second output results. The binarization determination condition module includes: The binarization processing unit is used to binarize the microbial fingerprint category data of farmland runoff pollution to obtain a binarized 0 / 1 matrix, where 0 represents non-existence and 1 represents existence. The binarization determination unit is used to analyze single farmland runoff pollutant microbial fingerprints and / or combinations of multiple farmland runoff pollutant microbial fingerprints based on a binarized 0 / 1 matrix, and to determine the binarization determination conditions. The binarization determination condition unit includes: The first decision condition subunit is used to detect... g_Citrifermentans When microbial fingerprints are present, the extraction result is 1, indicating that the binarization criteria are met; or, The second decision condition subunit is used to simultaneously detect the exception. f_Desulfuromonadaceae and g_Geobacter When any two or more microbial fingerprints other than the combination are present, the extraction result is 1, indicating that the binarization judgment condition is met.

Images