Method for determining comprehensive toxicity effect of complex pollutants
By constructing a deep learning model to integrate multidimensional behavioral toxicity indicators, the accuracy problem of toxicity assessment of compound pollutants was solved, and efficient quantitative assessment and risk warning of the toxic effects of compound pollutants were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV
- Filing Date
- 2026-02-27
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies are insufficient to effectively reflect the comprehensive impact of complex pollutants on aquatic organisms. Traditional single-indicator or chemical analysis methods lack sensitivity and are difficult to conduct accurate toxicity assessments in complex wastewater samples.
A comprehensive toxicity prediction model based on deep learning is constructed. By integrating multidimensional behavioral toxicity indicators and utilizing autoencoders, spatiotemporal feature fusion, or graph neural network models, an efficient quantitative assessment of the toxic effects of compound pollutants can be achieved.
It achieves high-precision, adaptive prediction of the toxic effects of complex pollutants, improving the accuracy and efficiency of toxicity assessment, and is applicable to practical environmental supervision and risk early warning.
Smart Images
Figure CN122157852A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of comprehensive toxicity effect determination of compound pollutants, specifically to a method for determining the comprehensive toxicity effect of compound pollutants. Background Technology
[0002] Wastewater and its receiving water bodies typically contain multiple categories of pollutants, including organic matter, inorganic salts, metal ions, pharmaceuticals, personal care products, and disinfection byproducts. Due to the complexity of pollution sources, the significant fluctuations in composition over time and under process conditions, and the potential for synergistic, antagonistic, or cumulative effects between different pollutants, relying solely on chemical analysis of target compounds is insufficient to reflect the overall impact of the mixed system on aquatic organisms. Therefore, comprehensive toxicity assessment methods based on biological responses are crucial for identifying the toxic effects of wastewater. Daphnia macrocarpa, as a typical freshwater model organism, is widely used for wastewater and chemical toxicity assessment. However, existing standard methods often rely on acute toxicity endpoints such as LC50 as criteria. These indicators primarily reflect high doses or acute effects, and their sensitivity is limited under low-toxicity or sublethal conditions, making it difficult to reveal the comprehensive disturbance of complex wastewater samples on the behavior and function of organisms.
[0003] With the development of behavioral testing technologies, multidimensional behavioral toxicity indicators based on the movement trajectories of large daphnia are increasingly being used to characterize sublethal effects. However, these behavioral indicators are typically high-dimensional and highly volatile, with correlations and redundancies among different indicators, and are significantly affected by experimental conditions and sample types, making it difficult to comprehensively and objectively represent multidimensional behavioral information. Existing technologies often employ single-indicator comparisons, empirical threshold judgments, or manual weighting and simple summarization of multiple behavioral indicators to obtain comprehensive results, which easily introduces subjective factors, fails to fully reflect the overall toxicity effect, and is not conducive to horizontal comparisons between different pollutants or actual wastewater samples. Furthermore, existing behavioral evaluation results mostly remain at the level of indicator changes, lacking a technical solution to stably map multidimensional behavioral responses into a single comprehensive toxicity indicator for further toxicity effect determination, thus restricting the practical application of behavioral toxicity assessment in complex and compound pollution systems. Summary of the Invention
[0004] To address the above problems, this invention provides a method for determining the comprehensive toxicity effects of complex pollutants.
[0005] A method for determining the comprehensive toxicity effect of a complex pollutant includes the following steps: S101. Construct a comprehensive toxicity prediction model; Identify the test organisms used to characterize the toxicity of pollutants in water; To obtain multidimensional behavioral toxicity indicators to characterize changes in test organisms exposed to pollutants; A comprehensive toxicity prediction model based on deep learning is constructed. The input of the comprehensive toxicity prediction model is the multidimensional behavioral toxicity index, and the output is the predicted value of the comprehensive toxicity index. S102. Comprehensive toxicity assessment of wastewater; Using the multidimensional behavioral toxicity indicators of the wastewater to be tested as input, the comprehensive toxicity model is used to predict the comprehensive toxicity indicators of the wastewater to be tested. The predicted value of the comprehensive toxicity index of the wastewater to be tested is compared with the value of the preset toxicity threshold to determine whether the wastewater to be tested has a toxic effect.
[0006] Note: The above method integrates multidimensional behavioral toxicity indicators and utilizes deep learning models to achieve a comprehensive, sensitive, and efficient quantitative assessment of the toxic effects of compound pollutants. It not only breaks through the limitations of traditional single-indicator or chemical analysis methods and can more realistically reflect the comprehensive impact of pollutants on organisms, but also significantly improves the accuracy and efficiency of toxicity determination through automated prediction models, providing more scientific and dynamic technical support for water environment risk early warning and control.
[0007] Furthermore, the test organism includes one of aquatic invertebrates, fish embryos, and / or microorganisms; The test organisms were determined through preliminary experimental assessments based on their sensitivity to pollutants and the richness of extractable behavioral indicators.
[0008] Note: The above screening criteria, based on the sensitivity of pollutants and the richness of extractable behavioral indicators, ensure that organisms have basic sensitivity to pollutants and guarantee that toxic responses have multidimensional and quantifiable characterization potential.
[0009] Furthermore, the test organism is Daphnia macrocarpa, and the multidimensional behavioral toxicity indicators include: Daphnia macrocarpa activity under dark conditions within a specific time period, Daphnia macrocarpa activity under light conditions within a specific time period, total distance Daphnia macrocarpa movement under dark conditions within a specific time period, total distance Daphnia macrocarpa movement under light conditions within a specific time period, Daphnia macrocarpa burst distance within a specific time period, Daphnia macrocarpa cruising distance within a specific time period, and Daphnia macrocarpa freezing distance within a specific time period.
[0010] Note: The selection of the above-mentioned organisms and the determination of indicators can efficiently capture the complex and subtle effects of different pollutants on the behavior of organisms, realize early warning, and ensure that all indicators are highly objective and quantifiable, providing high-quality and information-rich standardized input data for subsequent deep learning models, thereby significantly improving the accuracy, reliability and practicality of comprehensive toxicity prediction.
[0011] Furthermore, the acquisition of multidimensional behavioral toxicity indicators for characterizing changes in test organisms exposed to pollutants includes: Raw multidimensional behavioral toxicity data of test organisms exposed to pollutant samples and wastewater samples from the entire wastewater treatment process were obtained using the ZebraLab system. The original multidimensional behavioral toxicity index data were cleaned and standardized to obtain multidimensional behavioral toxicity indexes used to characterize changes in the test organisms exposed to pollutants. The data cleaning includes removing outliers; the data standardization process uses the Z-Score method.
[0012] Note: The above method achieves objective and efficient data collection by integrating an automated behavior tracking system (ZebraLab), and combines rigorous data cleaning and standardized preprocessing to ensure high quality and high consistency of the original multidimensional behavioral indicators; it eliminates interference from equipment errors and individual differences, and constructs a reliable and comparable standardized dataset.
[0013] Furthermore, the comprehensive toxicity prediction model adopts one of the following: an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model.
[0014] Furthermore, the construction of a comprehensive toxicity prediction model includes: Construct a feature extraction module and a regression output module; The feature extraction module is used to extract deep features of the original data based on the input of the comprehensive toxicity prediction model, through at least one layer of a neural network, such as an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model, and represent them using feature vectors. The regression output module is used to receive the feature vectors and map them to form a comprehensive toxicity index prediction value through a fully connected layer with a linear activation function, such as an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model. Training the model: Using the training dataset, the prediction loss function is minimized by an optimizer to jointly optimize the parameters of the feature extraction module and the regression output module until the model converges and is completed.
[0015] Explanation: The above method, by introducing cutting-edge deep learning architectures such as autoencoders, spatiotemporal feature fusion, or graph neural networks, constructs a predictive model capable of intelligently and automatically extracting and learning deep nonlinear features and intrinsic correlations from complex, high-dimensional behavioral data. Feature extraction and regression output training optimization enable the model not only to efficiently reduce dimensionality and capture key toxicity response patterns, but also to directly and accurately map to comprehensive toxicity values. This achieves high-precision, adaptive prediction of the comprehensive toxicity effects of compound pollutants, significantly surpassing the analytical capabilities and predictive performance of traditional linear or shallow models.
[0016] Furthermore, the step of comparing the predicted value of the comprehensive toxicity index of the wastewater to be tested with the preset toxicity threshold to determine whether the wastewater to be tested has a toxic effect includes: When the predicted value of the comprehensive toxicity index of the wastewater to be tested is less than the preset toxicity threshold, it is determined that the water sample to be tested has an ecotoxicity effect; otherwise, it is determined that the water sample to be tested does not have an ecotoxicity effect.
[0017] Note: By objectively comparing with scientifically preset toxicity thresholds, rapid and standardized classification and determination of wastewater ecological risks are achieved, enabling complex technical analysis results to be directly applied to actual environmental supervision and risk early warning, thus improving the practicality and operability of comprehensive toxicity assessment.
[0018] Furthermore, the method for determining the preset toxicity threshold is as follows: Set up a control sample and calculate the value of the comprehensive toxicity index of the wastewater control sample based on the comprehensive toxicity prediction model; Based on the statistical distribution of comprehensive toxicity index values, a preset toxicity threshold is set.
[0019] Note: The above method establishes an objective and scientific baseline based on uncontaminated control samples. By calculating the toxicity indicators of the control samples using a model and analyzing their statistical distribution, the set thresholds accurately reflect the background fluctuation range under specific environmental or process conditions, thereby effectively distinguishing between normal background fluctuations and actual pollution toxicity effects.
[0020] Furthermore, the predicted value of the comprehensive toxicity index is in the range of (0, 10], and the lower the value, the stronger the comprehensive toxic effect on the test organism.
[0021] The beneficial effects of this invention are: This invention, while maintaining a full characterization of multidimensional behavioral toxicity information, achieves a normalized expression of comprehensive toxicity indicators, avoiding the subjectivity and instability caused by manual weighting or simple summarization. Simultaneously, it enables rapid and sensitive toxicity identification of complex wastewater and mixed pollutant systems without relying on single pollutant concentration analysis, particularly suitable for comprehensive evaluation of low-dose and sublethal effects, improving the method's practical engineering applicability and promotional value. Therefore, this invention constructs an ATI-based comprehensive toxicity effect determination method, allowing comprehensive toxicity evaluation results to be directly used for identifying the toxicity effects of wastewater samples, applicable to comprehensive toxicity assessment scenarios involving both chemical pollutants and actual wastewater samples. Attached Figure Description
[0022] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation
[0023] To further illustrate the methods and effects of this invention, the technical solution of this invention will be clearly and completely described below in conjunction with experiments.
[0024] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.
[0025] Example 1: A method for determining the comprehensive toxicity effect of a complex pollutant, comprising the following steps: S101. Construct a comprehensive toxicity prediction model; (1) Identify the test organisms used to characterize the toxicity of pollutants in water; The test organisms include one of aquatic invertebrates, fish embryos, and / or microorganisms; The test organisms were determined through preliminary experimental assessments based on their sensitivity to pollutants and the richness of extractable behavioral indicators.
[0026] For example, the determination of test organisms for characterizing the toxicity of pollutants in water includes: determining the median lethal concentration (LD50) or median effective concentration (LD50) of each organism in the initial set of test organisms for typical pollutants through preliminary experiments, screening them according to their toxicity sensitivity, and preferentially selecting species with low LD50 or LD50 concentrations and high responsiveness to pollutants. For example, screening species with LD50 or LD50 concentrations less than or equal to a preset threshold (e.g., 100 mg / L).
[0027] For example, preliminary experiments can be conducted to determine the median lethal concentration (LD50) or median effect concentration (LD50) of each organism in the initial set of test organisms for typical pollutants. Based on the comparison of the relative toxicity response levels among different species, species that show high toxicity sensitivity to pollutants can be screened, for example, whose LD50 or LD50 concentration is located in the top 30% or top 50% of the candidate species distribution.
[0028] It should be understood that, in addition to the two methods mentioned above, there are other methods that can screen out aquatic organisms that respond positively to pollution, and the embodiments of the present invention do not limit this.
[0029] For example, aquatic invertebrates such as Daphnia macrocarpa, Daphnia davidii, and Artemia; fish embryos such as Zebrafish and Killifish; and microorganisms such as luminescent bacteria can be selected using either of the two methods described above. One to three organisms with high response sensitivity, rich behavioral indicators, and strong environmental adaptability are selected as test organisms, such as Daphnia macrocarpa in this embodiment.
[0030] (2) Obtain multidimensional behavioral toxicity indicators to characterize changes in test organisms exposed to pollutants; The test organism is Daphnia macrocarpa, and the multidimensional behavioral toxicity indicators include: Daphnia macrocarpa activity under dark conditions within a specific time period, Daphnia macrocarpa activity under light conditions within a specific time period, total distance Daphnia macrocarpa movement under dark conditions within a specific time period, total distance Daphnia macrocarpa movement under light conditions within a specific time period, Daphnia macrocarpa burst distance within a specific time period, Daphnia macrocarpa cruising distance within a specific time period, and Daphnia macrocarpa freezing distance within a specific time period.
[0031] The acquisition of multidimensional behavioral toxicity indicators used to characterize changes in test organisms exposed to pollutants includes: Raw multidimensional behavioral toxicity data of Daphnia magna were obtained from pollutant samples and wastewater samples from the entire process of the wastewater treatment plant under exposure using the ZebraLab system. The original multidimensional behavioral toxicity index data were cleaned and standardized to obtain multidimensional behavioral toxicity indexes used to characterize changes in the test organisms exposed to pollutants. The data cleaning includes removing outliers; the data standardization process uses the Z-Score method.
[0032] (3) Construct a comprehensive toxicity prediction model based on deep learning. The input of the comprehensive toxicity prediction model is the multidimensional behavioral toxicity index, and the output is the predicted value of the comprehensive toxicity index. The comprehensive toxicity prediction model adopts one of the following: an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model.
[0033] It should be understood that the Integrated Toxicity Index (ATI) is a comprehensive index proposed by Jamshidi-Zanjani and Saeedi to assess the degree of pollution of potentially toxic elements (PTEs) in aquatic sediments by integrating the toxic effects of multiple heavy metals. In this invention, the Integrated Toxicity Index (ATI) is specifically used to characterize the overall toxicity effect of the test organism (Daphnia macrocarpa) under exposure to multiple pollutants, output by nonlinear fusion of its multidimensional behavioral characteristics through a deep learning model. This index transforms the core logic of traditional ATI, which assesses sediment toxicity through weighted summation of multi-element chemical concentrations, into assessing wastewater toxicity through deep learning of multidimensional biological behavioral responses. Its physical meaning is the intensity of biological effects; it is used to quantify the deviation between the actual degree of toxic stress exhibited by the organism and its normal physiological state, reflecting the real functional damage suffered by the living system.
[0034] The construction of a comprehensive toxicity prediction model includes: Construct a feature extraction module and a regression output module; The feature extraction module is used to extract deep features of the original data based on the input of the comprehensive toxicity prediction model, through at least one layer of a neural network, such as an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model, and represent them using feature vectors. The regression output module is used to receive the feature vectors and map them to form a comprehensive toxicity index prediction value through a fully connected layer with a linear activation function, such as an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model. Training the model: Using the training dataset, the prediction loss function is minimized by an optimizer to jointly optimize the parameters of the feature extraction module and the regression output module until the model converges and is completed.
[0035] It should be understood that the autoencoder dimensionality reduction model includes an encoder and a decoder. The encoder maps multidimensional behavioral toxicity indicators to low-dimensional latent variables, which are directly used as the Advanced Toxicity Index (ATI) or output as the ATI through a fully connected layer. The spatiotemporal feature fusion model includes a convolutional neural network (CNN) module and a recurrent neural network (RNN) module. The CNN extracts spatial correlation features of the behavioral indicators, and the RNN extracts temporal evolution features. The fusion layer concatenates the spatiotemporal features and outputs the ATI through a fully connected layer. The graph neural network model treats individual subjects as nodes and constructs behavioral interaction relationships as edges. It aggregates group behavioral features through graph convolutional layers and outputs a group-level ATI.
[0036] Specifically, this embodiment uses an autoencoder dimensionality reduction model; For example, in this embodiment, the data acquisition and model construction in the autoencoder dimensionality reduction model include the following steps S1-1 to S1-5: S1-1, Behavioral Data Collection In this embodiment, 25 pollutants were used to establish the model, corresponding to 25 pollutant samples; wastewater samples came from 31 wastewater treatment plants across their entire process flow, totaling 186 samples. The test organism was a Daphnia macrocarpa, which had been stably propagated for more than 10 generations in our laboratory.
[0037] In this embodiment, the collected behavioral toxicity indicators consisted of 7-dimensional behavioral characteristics, including: Daphnia macrocarpa activity under 5 minutes of darkness (activity-dark), Daphnia macrocarpa activity under 5 minutes of light (activity-light), total distance traveled under 5 minutes of darkness (total distance-dark), total distance traveled under 5 minutes of light (total distance-light), burst distance under both darkness and light conditions within 10 minutes, cruising distance under both darkness and light conditions within 10 minutes, and freezing distance under both darkness and light conditions within 10 minutes. Daphnia macrocarpa behavioral data were collected using ZebraLab video acquisition and behavioral analysis equipment.
[0038] S1-2, Data Standard Processing The collected data undergoes data cleaning to remove sample data that fails to identify trajectories, has severe missing data, or is obviously abnormal; subsequently, the retained data is standardized. In this embodiment, the Z-score standardization method is used for data standardization.
[0039] S1-3, Construction of Comprehensive Toxicity Model Based on a preprocessed 7-dimensional behavioral toxicity index, a comprehensive toxicity model is constructed using an autoencoder dimensionality reduction model (including an encoder and a decoder). The dataset is divided into an 80% training set and a 20% test set. The encoder is trained to reconstruct the behavioral features, mapping the 7-dimensional behavioral features to a one-dimensional comprehensive representation. This one-dimensional output is defined as the Comprehensive Toxicity Index (ATI), used to characterize the overall behavioral damage to *Daphnia magna*. In this embodiment, a smaller ATI value indicates a stronger behavioral toxicity effect.
[0040] S1-5, Model Performance Evaluation The model performance is evaluated using a test set. In this embodiment, the performance is based on the mean squared error (MSE) and the coefficient of determination (R²). 2 The model's performance was evaluated, and the validation result was MSE=0.173; R0 2 =0.827. This indicates that the model has a good comprehensive characterization ability for multidimensional behavioral toxicity information and can be used for subsequent ATI output and judgment of water samples to be tested.
[0041] S102. Comprehensive toxicity assessment of wastewater; (4) Using the multidimensional behavioral toxicity index of the wastewater to be tested as input, the comprehensive toxicity model is used to make predictions and obtain the predicted value of the comprehensive toxicity index of the wastewater to be tested. For example, the prediction steps using the integrated toxicity model are as follows: The actual influent of a municipal wastewater treatment plant was selected as sample 1 for testing. The pH value of this water sample was 7.7, the COD value was 212.15 mg / L, and the total organic carbon (TOC) value was 43.36 mg / L.
[0042] A large-scale Daphnia exposure experiment was conducted on sample 1. Seven-dimensional behavioral toxicity indicators were collected using the same ZebraLab equipment as in step S1, with five parallel groups for each water sample. The collected behavioral toxicity indicators were still seven-dimensional behavioral characteristics (including activity-brightness, activity-darkness, total distance-brightness, total distance-darkness, burst distance, cruising distance, and freezing distance; units consistent with the equipment output). After cleaning and Z-score standardization according to the same rules as in step S1, the raw data are shown in Table 1. Table 1. Behavioral toxicity data of Daphnia macrocarpa after exposure to sample 1 (n=5)
[0043] Substituting the pretreated behavioral toxicity indexes of the water sample into the comprehensive toxicity model, the one-dimensional comprehensive toxicity characterization ATI=0.35 was output.
[0044] (5) Compare the predicted value of the comprehensive toxicity index of the wastewater to be tested with the preset toxicity threshold to determine whether the wastewater to be tested has a toxic effect.
[0045] Specifically, it includes: When the predicted value of the comprehensive toxicity index of the wastewater to be tested is less than the preset toxicity threshold, it is determined that the water sample to be tested has an ecotoxicity effect; otherwise, it is determined that the water sample to be tested does not have an ecotoxicity effect.
[0046] The method for determining the preset toxicity threshold is as follows: set a negative control sample (standard dilution water), calculate the value of the comprehensive toxicity index of the wastewater control sample based on the comprehensive toxicity prediction model, and set the preset toxicity threshold based on the statistical distribution of the comprehensive toxicity index value. The behavioral indicators of the negative control samples were collected, and the data cleaning and standardization methods were the same as those of the test samples. The same feature set and standardized parameters as those used in the model training phase were also employed.
[0047] It should be understood that the preset toxicity threshold can be taken as the mean ATI of the negative control sample minus k times the standard deviation, where k is a real number from 1 to 3, preferably 2; or it can be taken as a specific percentile of the ATI of the negative control sample, preferably the 1st to 10th percentile (e.g., the 5th percentile). This threshold defines the lower limit of the normal fluctuation range of the ATI value. When the ATI value of the test sample is lower than this threshold, its toxic effect has exceeded the normal fluctuation range, and therefore it can be determined that there is a significant toxic effect.
[0048] The predicted value of the comprehensive toxicity index is in the range of (0~10), and the lower the value, the stronger the comprehensive toxic effect on the test organism.
[0049] For example, in this embodiment, a threshold method is used to determine the risk, and the toxicity threshold is set to ATI=1.73. If ATI≥1.73, it indicates that the sample to be tested has no significant effect on the behavior of Daphnia macrocarpa, and it is determined that there is no toxic effect; if ATI<1.73, it indicates that the sample to be tested has a significant effect on the behavior of Daphnia macrocarpa, and it is determined that there is a toxic effect.
[0050] Therefore, the ATI of sample 1 was 0.35 < 1.73, indicating that it had a significant toxic effect.
[0051] Example 2: This example uses the same method as Example 1, except that different test samples are used. Specific implementation examples are as follows: This embodiment selects actual effluent from a municipal wastewater treatment plant as sample 2 and uses the method of this invention to determine its comprehensive toxicity. For details regarding the establishment of the comprehensive toxicity model, the selection of the behavioral toxicity index system (7-dimensional behavioral features), data cleaning and standardization principles, and model training and performance evaluation methods (MSE and R2), please refer to the corresponding sections of Embodiment 1; these will not be repeated here. This embodiment focuses on illustrating the process and results of substituting the behavioral data of sample 2 into the model to obtain the ATI under the same model and preprocessing strategy, and determining the toxicity effect based on the ATI.
[0052] S2. Substitute actual sample data: Determine the behavioral toxicity index of Daphnia macrocarpa after exposure to sample 2 and calculate ATI. S2-1, Measurement of Behavioral Indicators of Large Daphnia Sample 2 was taken from the actual effluent of a municipal wastewater treatment plant. Its pH value was 6.9, COD value was 42.15 mg / L, and TOC value was 9.63 mg / L.
[0053] Using the same test organisms, exposure conditions, and ZebraLab video acquisition and behavioral analysis equipment as in Example 1, behavioral toxicity indicators of *Daphnia macrocarpa* exposed to test sample 2 were determined. Five parallel groups were set up for each water sample. The collected behavioral toxicity indicators were still 7-dimensional behavioral characteristics (including activity-brightness, activity-darkness, total distance-brightness, total distance-darkness, bursting distance, cruising distance, and freezing distance; units consistent with equipment output). The obtained raw data are shown in Table 2.
[0054] Table 2. Behavioral toxicity data of Daphnia macrocarpa under exposure of sample 2 (n=5)
[0055] S2-2, Model Substitution Calculation The behavioral toxicity index data shown in Table 2 were preprocessed according to the same data cleaning rules and standardization methods as in Example 1 (including using the same feature set, standardization parameters, and processing procedures) to ensure the consistency of the model input distribution. Subsequently, the preprocessed 7-dimensional behavioral feature data were input into the comprehensive toxicity model established in Example 1. The model calculated that the comprehensive toxicity index (ATI) of sample 2 was 2.68.
[0056] S3. Comprehensive Toxicity Effect Assessment: Toxicity effect assessment based on ATI. This embodiment uses the same threshold method as in Embodiment 1 for risk assessment, with the threshold set at ATI=1.73: when ATI>1.73, it is determined that the water sample to be tested has no significant impact on the behavior of Daphnia macrocarpa and the water sample to be tested has no significant toxic effect; when ATI<1.73, it is determined that the water sample to be tested has a significant impact on the behavior of Daphnia macrocarpa and the water sample to be tested has a significant toxic effect.
[0057] Since the ATI of sample 2 is 2.68 > 1.73, it is determined that sample 2 has no significant toxic effect.
[0058] Example 3: This example uses the same method as Example 1, except that different test samples are used. Specific implementation examples are as follows: In this embodiment, a mixture of multiple nonsteroidal anti-inflammatory drugs was selected as the test sample 3 to simulate the comprehensive toxicity effect under the condition of coexistence of similar drugs in actual sewage.
[0059] The concentration of each chemical in the mixture is configured according to its own toxicity proportion, meaning that each chemical contributes an equal share of toxicity to the mixture. Specifically, the concentration of each chemical is determined proportionally to its LC10 value, ensuring that the toxicity units of each component are equal.
[0060] In this embodiment, the total exposure concentration of the nonsteroidal anti-inflammatory drug mixture was set at 37.42 mg / L, and the concentrations of each component were prepared according to the above-mentioned principle of equal toxicity ratio. The LC10 values of each component in this embodiment are as follows: diclofenac 25.22 mg / L, ibuprofen 15.68 mg / L, acetylsalicylic acid 21.19 mg / L, gemfibrozil 15.63 mg / L, and acetaminophen 62.24 mg / L. Therefore, the concentration ratio of each chemical in sample 3 to be tested was set as follows: diclofenac: ibuprofen: acetylsalicylic acid: gemfibrozil: acetaminophen = 25.22:15.68:21.19:15.63:62.24. The remaining culture conditions, exposure methods, video acquisition and behavior analysis procedures are the same as in Example 1, and will not be repeated here.
[0061] S2. Substitute actual sample data: Determine the behavioral toxicity index of Daphnia macrocarpa after exposure to sample 3 and calculate ATI. S2-1, Measurement of Behavioral Indicators of Large Daphnia Using the same experimental system and ZebraLab video acquisition and behavior analysis equipment as in Example 1, the 7-dimensional behavioral toxicity index of Daphnia magna under exposure conditions was obtained for sample 3. Five replicates were set up for each group, and the data obtained are shown in Table 3.
[0062] Table 3. Behavioral toxicity data of Daphnia macrocarpa under exposure of sample 3 (n=5)
[0063] S2-2, Model Substitution Calculation The behavioral toxicity indices shown in Table 3 were preprocessed using the same data cleaning rules and standardized parameters as in Example 1 to ensure consistency with the input distribution during the model training phase. Subsequently, the preprocessed 7-dimensional behavioral feature data were input into the comprehensive toxicity model established in Example 1. According to the model calculation, the comprehensive toxicity index ATI of the test sample 3 at 37.42 mg / L was 0.78.
[0064] S3. Comprehensive Toxicity Effect Assessment: Toxicity effect assessment based on ATI. Using the same threshold determination criteria as in Example 1, the threshold was set to ATI = 1.73: when the ATI was below the threshold, a toxic effect was determined to exist; when the ATI was above or equal to the threshold, no significant toxic effect was determined to exist. Since ATI = 0.78 < 1.73 in this example, it was determined that sample 3 had a significant toxic effect under the exposure condition of 37.42 mg / L.
[0065] Example 4: This example uses the same method as Example 1, except that different test samples are used. Specific implementation examples are as follows: In this embodiment, a mixture of multiple typical endocrine disruptors was selected as the test sample 4 to simulate the comprehensive toxicity effect of similar drugs coexisting in actual wastewater.
[0066] The concentration of each chemical in the mixture is configured according to its own toxicity proportion, meaning that each chemical contributes an equal share of toxicity to the mixture. Specifically, the concentration of each chemical is determined proportionally to its LC10 value, ensuring that the toxicity units of each component are equal.
[0067] In this embodiment, the total exposure concentration of the nonsteroidal anti-inflammatory drug mixture was set at 33.59 mg / L, and the concentrations of each component were prepared according to the above-mentioned equitoxicity ratio principle. The LC10 values of each component in this embodiment are as follows: ethinyl estradiol 13.93 mg / L, 17β-estradiol 1.95 mg / L, estrone 18.24 mg / L, nonylphenol 0.11 mg / L, bisphenol S 10.94 mg / L, and estriol 78.63 mg / L. Therefore, the concentration ratio of each chemical in sample 4 to be tested was set as follows: ethinyl estradiol: 17β-estradiol: estrone: nonylphenol: bisphenol S: estriol = 13.93:1.95:18.24:0.11:10.94:78.63. The remaining culture conditions, exposure methods, video acquisition and behavior analysis procedures are the same as in Example 1, and will not be repeated here.
[0068] S2. Substitute actual sample data: Determine the behavioral toxicity index of Daphnia macrocarpa after exposure to sample 4 and calculate ATI. S2-1, Measurement of Behavioral Indicators of Large Daphnia Using the same experimental system and ZebraLab video acquisition and behavior analysis equipment as in Example 1, the 7-dimensional behavioral toxicity index of Daphnia magna under exposure conditions was obtained for sample 4. Five replicates were set up for each group, and the data obtained are shown in Table 4.
[0069] Table 4. Behavioral toxicity data of Daphnia macrocarpa under exposure to sample 4 (n=5)
[0070] S2-2, Model Substitution Calculation The behavioral toxicity indicators shown in Table 4 were preprocessed using the same data cleaning rules and standardized parameters as in Example 1 to ensure consistency with the input distribution during the model training phase. Subsequently, the preprocessed 7-dimensional behavioral feature data were input into the comprehensive toxicity model established in Example 1. According to the model calculation, the comprehensive toxicity index ATI of the test sample 4 at 33.59 mg / L was 0.84.
[0071] S3. Comprehensive Toxicity Effect Assessment: Toxicity effect assessment based on ATI. Using the same threshold determination criteria as in Example 1, the threshold was set to ATI = 1.73: when the ATI was below the threshold, a toxic effect was determined to exist; when the ATI was above or equal to the threshold, no significant toxic effect was determined to exist. Since ATI = 0.84 < 1.73 in this example, it was determined that sample 4 had a significant toxic effect under the exposure condition of 33.59 mg / L.
[0072] Example 5: The method in this example is roughly the same as that in Example 1. The difference is that the comprehensive toxicity prediction model adopts a spatiotemporal feature fusion model. The architecture of this model is as described in Example 1, and the specific training method is a conventional training method.
[0073] Example 6: The method in this example is roughly the same as that in Example 1. The difference is that the comprehensive toxicity prediction model uses a graph neural network model. The architecture of this model is as described in Example 1, and the specific training method is a conventional training method.
Claims
1. A method for determining the comprehensive toxicity effect of a complex pollutant, characterized in that, Includes the following steps: S101. Construct a comprehensive toxicity prediction model; Identify the test organisms used to characterize the toxicity of pollutants in water; To obtain multidimensional behavioral toxicity indicators to characterize changes in test organisms exposed to pollutants; A comprehensive toxicity prediction model based on deep learning is constructed. The input of the comprehensive toxicity prediction model is the multidimensional behavioral toxicity index, and the output is the predicted value of the comprehensive toxicity index. S102. Comprehensive toxicity assessment of wastewater; Using the multidimensional behavioral toxicity indicators of the wastewater to be tested as input, the comprehensive toxicity model is used to predict the comprehensive toxicity indicators of the wastewater to be tested. The predicted value of the comprehensive toxicity index of the wastewater to be tested is compared with the value of the preset toxicity threshold to determine whether the wastewater to be tested has a toxic effect.
2. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 1, characterized in that, The test organism is one of the following: aquatic invertebrates, fish embryos, and / or microorganisms; The test organisms were determined through preliminary experimental assessments based on their sensitivity to pollutants and the richness of extractable behavioral indicators.
3. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 2, characterized in that, The test organism is Daphnia macrocarpa, and the multidimensional behavioral toxicity indicators include: Daphnia macrocarpa activity under dark conditions within a specific time period, Daphnia macrocarpa activity under light conditions within a specific time period, total distance Daphnia macrocarpa movement under dark conditions within a specific time period, total distance Daphnia macrocarpa movement under light conditions within a specific time period, Daphnia macrocarpa burst distance within a specific time period, Daphnia macrocarpa cruising distance within a specific time period, and Daphnia macrocarpa freezing distance within a specific time period.
4. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 1, characterized in that, The acquisition of multidimensional behavioral toxicity indicators used to characterize changes in test organisms exposed to pollutants; include: Raw multidimensional behavioral toxicity data of test organisms exposed to pollutant samples and wastewater samples from the entire wastewater treatment process were obtained using the ZebraLab system. The original multidimensional behavioral toxicity index data were cleaned and standardized to obtain multidimensional behavioral toxicity indexes used to characterize changes in the test organisms exposed to pollutants. The data cleaning includes removing outliers; the data standardization process uses the Z-Score method.
5. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 1, characterized in that, The comprehensive toxicity prediction model adopts one of the following: an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model.
6. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 5, characterized in that, The construction of a comprehensive toxicity prediction model includes: Construct a feature extraction module and a regression output module; The feature extraction module is used to extract deep features of the original data based on the input of the comprehensive toxicity prediction model, through at least one layer of a neural network, such as an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model, and represent them using feature vectors. The regression output module is used to receive the feature vectors and map them to form a comprehensive toxicity index prediction value through a fully connected layer with a linear activation function, such as an autoencoder dimensionality reduction model, a spatiotemporal feature fusion model, or a graph neural network model. Training the model: Using the training dataset, the prediction loss function is minimized by an optimizer to jointly optimize the parameters of the feature extraction module and the regression output module until the model converges and is completed.
7. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 1, characterized in that, The step of comparing the predicted value of the comprehensive toxicity index of the wastewater to be tested with the value of the preset toxicity threshold to determine whether the wastewater to be tested has a toxic effect includes: When the predicted value of the comprehensive toxicity index of the wastewater to be tested is less than the preset toxicity threshold, it is determined that the water sample to be tested has an ecotoxicity effect; otherwise, it is determined that the water sample to be tested does not have an ecotoxicity effect.
8. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 7, characterized in that, The method for determining the preset toxicity threshold is as follows: Set up a control sample and calculate the value of the comprehensive toxicity index of the wastewater control sample based on the comprehensive toxicity prediction model; Based on the statistical distribution of comprehensive toxicity index values, a preset toxicity threshold is set.
9. The method for determining the comprehensive toxicity effect of a complex pollutant as described in claim 1, characterized in that, The predicted value of the comprehensive toxicity index ranges from 0 to 10. The lower the value, the stronger the comprehensive toxic effect on the test organism.