Bioaccumulation pollutant chronic toxicity prediction system and method

Abstract

The invention relates to the technical field of environmental toxicology and ecological risk assessment, in particular to a system and a method for predicting chronic toxicity of bioaccumulation pollutants, and the system comprises a data acquisition and processing module, a model training and optimization module, a toxicity prediction module and an ecological risk assessment module. According to the method, full-process prediction is realized through data standardization, feature engineering, model training and model prediction; tissue residual toxicity data is obtained and standardized, a multi-dimensional characteristic matrix containing pollutant physicochemical properties, species biological characteristics and experimental conditions is constructed, a toxicity prediction model is optimized by using a machine learning algorithm, and tissue residual toxicity prediction of fishes in different life stages is realized. And calculating a risk quotient based on the predicted value, and generating a risk assessment report. The problems that an existing system depends on water body concentration toxicity data, is lack of biological accumulation pollutant chronic toxicity prediction capacity and does not aim at an endangered characteristic optimization model are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of environmental toxicology and ecological risk assessment, and specifically relates to a system and method for predicting the chronic toxicity of bioaccumulative pollutants. Background Art

[0002] Global biodiversity is facing unprecedented challenges. According to the International Union for Conservation of Nature (IUCN) Red List of Threatened Species, as of December 2024, 12% of bird species, over 26% of mammal species, and 41% of amphibian species are endangered. The Living Planet Report 2024 further indicates that the decline in biodiversity in freshwater ecosystems is most severe, with freshwater fish populations declining by 85% compared to the baseline period, and approximately 25% of freshwater fish species facing extinction.

[0003] Bioaccumulative pollutants (such as flame retardants and polychlorinated biphenyls) accumulate and amplify through the food chain, posing a significant threat to the chronic toxicity of endangered fish. For example, polybrominated diphenyl ethers (PBDEs) have a bioaccumulation factor (BAF) of up to 10 in Great Lakes trout. 6.7 to 10 7.5 As substitutes for PBDEs, the global production of new brominated flame retardants (NBFRs) and organophosphorus flame retardants (OPFRs) has been increasing year by year, and the BAF of some OPFRs (such as CDPP) can reach 4.2×10 6 , which has been shown to cause teratogenic effects on fish embryos and inhibit growth and development. However, existing ecological risk assessments of bioaccumulative pollutants mostly rely on toxicity data based on water concentrations, lacking toxicity data based on residual concentrations in biological tissues, making it impossible to accurately characterize the effective dose of pollutants in organisms. However, toxicity data based on residual concentrations in biological tissues is extremely scarce, and existing toxicity prediction models have the following limitations: (1) Most systems focus on the prediction of acute toxicity and lack the ability to predict chronic toxicity; (2) The model has not been optimized based on the biological characteristics of endangered fish, making it difficult to use for toxicity prediction of endangered fish, and the prediction accuracy is insufficient.

[0004] Although some studies have attempted to use machine learning models (such as RF and ANN) for toxicity prediction, a prediction system that integrates tissue residue data, characteristics of rare and endangered species, and chronic toxicity endpoints has not yet been established, making it difficult to meet the actual needs of ecological protection.

[0005] In view of this, the present invention is proposed. Summary of the Invention

[0006] The purpose of the present invention is to solve the problems existing in the prior art, such as reliance on water concentration toxicity data, lack of residual toxicity data in biological tissues, inability to accurately characterize the cumulative effects and effective concentrations of pollutants in the body; lack of predictive ability for chronic toxicity and failure to optimize the model for the life cycle sensitivity and data scarcity of rare and endangered species, resulting in insufficient prediction accuracy.

[0007] To achieve the above object, the technical solution of the present invention is as follows: In a first aspect, a system for predicting chronic toxicity of bioaccumulative pollutants comprises: Data acquisition and processing module: used to obtain biological tissue residual toxicity data and related pollutant characteristic data, biological characteristic data and exposure condition data, and construct a data set after standardization; Model training and optimization module: used to construct a multi-dimensional feature matrix based on the data set, and use the multi-dimensional feature matrix to obtain a toxicity prediction model through machine learning algorithm optimization processing; Toxicity prediction module: used to predict the residual toxicity of bioaccumulative pollutants in the tissues of different life stages of organisms using the toxicity prediction model and output toxicity prediction values; Ecological risk assessment module: used to calculate the risk quotient based on the toxicity prediction value and generate a risk assessment report.

[0008] Furthermore, the data acquisition and processing module includes: Data collection unit: used to obtain tissue residual toxicity data of chronic toxicity endpoints including no observed effect concentration and lowest observed effect concentration from multi-source databases, as well as related pollutant characteristic data, biological characteristic data and exposure condition data; Data preprocessing unit: used to encode categorical variables, normalize numerical data, and use automatic class balancing algorithm to deal with data imbalance problems.

[0009] Furthermore, the machine learning algorithm is an extreme gradient boosting algorithm.

[0010] Furthermore, it also includes: The additive explanatory analysis determines the key feature weights, where the key features refer to characteristic variables in the multidimensional feature matrix that have a significant impact on the prediction results of the chronic toxicity of bioaccumulative pollutants.

[0011] Furthermore, the model training and optimization module includes: A feature engineering unit, used to construct a multi-dimensional feature matrix of the physical and chemical properties of pollution, biological characteristics of species, and experimental conditions based on the dataset; Model training unit, used to Perform 5-fold cross validation to optimize the parameters of extreme gradient boosting, and achieve fitting accuracy processing for the toxicity prediction model through the test set; A model validation unit is used to validate the toxicity prediction model through 10-fold cross validation and external validation after the fitting accuracy processing.

[0012] Furthermore, the Perform 5-fold cross validation to optimize the parameters of extreme gradient boosting, including: Define the parameter search space as: ,in and Satisfy the empirical correlation:

[0013] The mean square error and the coefficient of determination were calculated using 5-fold cross validation. The formula for calculating the mean square error is:

[0014] The formula for calculating the coefficient of determination is:

[0015] in, is the true toxicity value, is the predicted value, is the mean; the optimal parameter is determined to be .

[0016] Furthermore, the toxicity prediction module includes: a toxicity prediction unit, configured to input pollutant structure, species characteristics, and exposure conditions into the toxicity prediction model and output a predicted value of tissue residual toxicity; The sensitive feature identification unit is used to identify sensitive features in the tissue residual toxicity prediction model, where the sensitive features refer to features related to the highly sensitive response of an organism to the chronic toxicity of bioaccumulative pollutants.

[0017] Furthermore, the toxicity prediction unit verifies the prediction result by 100 times of bootstrap resampling technology.

[0018] Furthermore, the ecological risk assessment module includes: The risk quotient calculation unit is used to calculate the risk quotient value based on the predicted no-effect concentration and the measured concentration in the organism, and identify high-risk pollutants with RQ>1.

[0019] In a second aspect, a method for predicting the chronic toxicity of bioaccumulative pollutants comprises: S1. Obtain residual toxicity data in biological tissues and related pollutant characteristic data, biological characteristic data, and exposure condition data, and construct a data set after standardization; S2. constructing a multidimensional feature matrix based on the data set, and using the multidimensional feature matrix to obtain a toxicity prediction model through optimization processing using a machine learning algorithm; S3. Using the toxicity prediction model, predict the residual toxicity of bioaccumulative pollutants in tissues at different life stages of organisms and output a toxicity prediction value; S4. Calculate a risk quotient based on the toxicity prediction value and generate a risk assessment report.

[0020] Compared with the existing technology, the present invention provides a system and method for predicting the chronic toxicity of bioaccumulative pollutants. The system includes a data acquisition and processing module, a model training and optimization module, a toxicity prediction module, and an ecological risk assessment module. The method realizes full-process prediction through data standardization, feature engineering, model training, and model prediction; by acquiring tissue residual toxicity data and standardizing it, a multi-dimensional feature matrix including the physical and chemical properties of pollution, biological characteristics of species, and experimental conditions is constructed, and the toxicity prediction model is optimized using a machine learning algorithm to achieve tissue residual toxicity prediction for different life stages of fish, and the risk quotient is calculated based on the predicted value to generate a risk assessment report. This technology solves the problems of existing systems that rely on water concentration toxicity data, lack the ability to predict the chronic toxicity of bioaccumulative pollutants, and fail to optimize models for endangered characteristics. It provides a systematic solution for accurately evaluating the chronic toxicity of bioaccumulative pollutants to rare and endangered species, and has important application value for ecological risk assessment and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is an architectural diagram of a system for predicting the chronic toxicity of bioaccumulative pollutants provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be clearly described below in conjunction with the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0023] It should be noted that, unless otherwise specifically stated, the relative arrangements of components and steps, and numerical expressions set forth in these embodiments should not be construed as limiting the scope of the present invention.

[0024] The following description of exemplary embodiments is merely illustrative and is not intended to limit the present invention, its application, or use in any sense. Technologies, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but to the extent applicable, such technologies, methods, and apparatuses should be considered part of this specification.

[0025] Example 1 First, see Figure 1 , which is an architectural diagram of a chronic toxicity prediction system for bioaccumulative pollutants proposed in the present invention. The technical methods provided by the present invention can be applied to the prediction of any one or more of different chronic toxic effects, such as developmental toxicity, reproductive toxicity, and behavioral toxicity. Taking chronic developmental toxicity prediction as an example, the system implementation process specifically includes: A1. Data acquisition and processing module: used to obtain biological tissue residual toxicity data and related pollutant characteristic data, biological characteristic data and exposure condition data, and construct a data set after standardization. Specifically, it may include: A11, data acquisition unit The tissue toxicity data of organic pollutants on fish were automatically captured from the Ecotoxicology Knowledge Base (ECOTOX) and the Environmental Residue-Effect Database (ERED) databases. The screening criteria included laboratory research data of single chemicals (not mixtures), freshwater environments, developmental toxicity endpoints, and whole-body / muscle tissue concentration measurements.

[0026] A12, data preprocessing unit By encoding categorical variables, normalizing numerical data, and balancing data, we can solve the problems of inconsistent data formats and unbalanced samples, and provide standardized input for model training: A121, adopt r encoded "life stages" (egg, embryo, larvae, and adult) as continuous values ​​(egg = 1, embryo = 2, larvae = 3, and adult = 4) to prevent the model from misjudging the order of categories. Toxic effects (such as developmental malformations and growth inhibition) were one-hot encoded, and non-numeric variables were converted to binary vectors (such as [1,0,0] for developmental malformations) to facilitate model recognition.

[0027] A122. For all obtained numerical variables, first log 10 The transformation is used to compress the data range and reduce the impact of outliers, and then all numerical features are scaled to the [0,1] interval through minimum-maximum normalization. The formula is:

[0028] in, is the original value, and are the minimum and maximum values ​​of the logarithmic transformation values, respectively, to ensure that features of different dimensions are comparable.

[0029] A123. To address the problem of excessively high proportion of data on dominant species (for example, rainbow trout data account for 26.4%), an automatic class balancing algorithm is used to dynamically adjust sample weights, giving higher weights to samples of disadvantaged species (such as endangered fish).

[0030] A2, model training and optimization module: used to construct a multi-dimensional feature matrix based on the data set, and use the multi-dimensional feature matrix to optimize the machine learning algorithm to obtain a toxicity prediction model; the machine learning algorithm is extreme gradient boosting ( ) algorithm; through multi-dimensional feature engineering and algorithm optimization, a high-precision chronic toxicity prediction model is constructed to address the problem that the existing water-based acute toxicity prediction model cannot meet the long-term protection needs of rare and endangered species. Specifically, it may include: A21, Feature Engineering Unit: Used to construct a multi-dimensional feature matrix of the physical and chemical properties of pollution, biological characteristics of species and experimental conditions based on the data preprocessed by A12.

[0031] The physical and chemical characteristics of the pollution were characterized by ECFP4 molecular fingerprint to characterize the pollutant structure specificity, and combined with the logarithm of the n-octanol-water partition coefficient ( ) and bioaccumulation factor (BAF) to quantify the bioaccumulation potential, where the BAF data were derived from the environmental assessment tool EPISuite (v4.1.34) of the United States Environmental Protection Agency (USEPA).

[0032] Species biological characteristics integrate trophic level, body length, weight and life stage (egg = 1 to adult = 4) to capture the differences in toxicity sensitivity among different species and life cycles; experimental condition characteristics combine exposure time and toxic effect type to reflect the impact of exposure duration and effect type on toxicity.

[0033] A22, model training unit: used to A 5-fold cross validation was performed to optimize the parameters of the extreme gradient boosting algorithm, and the fitting accuracy processing of the toxicity prediction model was achieved through the test set.

[0034] Improve prediction accuracy through a systematic process: first, reserve 10% of the data points as an external validation set, and use 90% of the remaining data as a training set and 10% as a test set; then Perform 5-fold cross validation optimization Parameters, define the parameter search space as: ; in, and Satisfy the empirical correlation:

[0035] The mean square error and the coefficient of determination were calculated using 5-fold cross validation. The formula for calculating the mean square error is:

[0036] The formula for calculating the coefficient of determination is:

[0037] in, is the true toxicity value, is the predicted value, is the mean; the optimal parameter is determined to be . The model's prediction accuracy for chronic developmental toxicity based on tissue residual concentrations was significantly improved.

[0038] A23, model validation unit: used to validate the toxicity prediction model through 10-fold cross validation and external validation after the fitting accuracy processing.

[0039] Internal validation uses 10-fold cross validation to evaluate the robustness of the model on the training data and ensure the stability of the model fitting; External validation utilizes an independent dataset not used in training, focusing on verifying the model's ability to generalize to unknown data. This dual validation mechanism effectively addresses prediction biases caused by model overfitting, enabling highly accurate predictions of chronic developmental toxicity, such as embryonic malformations and growth inhibition, providing a scientifically reliable basis for subsequent toxicity assessments and risk decision-making.

[0040] A3. Toxicity prediction module: used to predict the residual toxicity of bioaccumulative pollutants in tissues at different life stages of organisms using the toxicity prediction model and output toxicity prediction values.

[0041] A31. Data acquisition and prediction unit: used to input pollutant structure, species characteristics and exposure conditions, and output tissue residual toxicity prediction values.

[0042] This unit supports the input of multiple parameters to initiate the prediction process, including physicochemical parameters characterizing the pollutant, biological parameters characterizing the species, parameters characterizing the exposure conditions, and parameters reflecting the toxic effects of the pollutant. Physicochemical parameters of the substance include BAF and ECFP4 molecular fingerprints obtained through SMILES structures; biological parameters include body length, weight, trophic level, and life stage coding; life stage is coded from 1 to 4 according to the developmental progression, corresponding to the egg, embryo, juvenile, and adult stages respectively; and the exposure condition parameter is the duration of exposure.

[0043] Based on these parameters, the model integrates a cross-dimensional feature matrix of the contaminant's physical and chemical characteristics, the species' biological properties, and experimental conditions to output predicted tissue residual toxicity values, including the No Observed Effect Concentration (NOEC) and the Lowest Observed Effect Concentration (LOEC). To ensure the reliability of the predictions, the system validates the results through 100 bootstrap resampling cycles, strictly controlling the prediction error to less than 10%.

[0044] A32. Sensitive feature identification unit: used to identify sensitive features in the tissue residual toxicity prediction model, wherein the sensitive features refer to features related to the highly sensitive response of organisms to the chronic toxicity of bioaccumulative pollutants.

[0045] A33. Toxicity prediction unit: Toxicity values ​​for sensitive life stages and species are derived based on comparative predictions. At the life stage level, the embryonic and juvenile stages are clearly defined as highly sensitive stages, and their predicted toxicity values ​​are generally 2 to 10 times lower than those of the adult stage. Taking the organophosphorus flame retardant IDPP as an example, its NOEC for juvenile fish is 0.41 mg / kg, while the NOEC for adults is 1.8 mg / kg, indicating that the juvenile stage is significantly more sensitive to this pollutant. At the species level, the system can accurately locate sensitive rare and endangered fish (such as the dry river loach and Shennong goby distributed in the Yangtze River basin), ensuring accurate capture of the toxicity responses of sensitive, rare and endangered species.

[0046] A4. Ecological Risk Assessment Module: This module calculates the risk quotient based on the toxicity prediction values ​​and generates a risk assessment report. This module quantifies the bioaccumulative effects by measuring residual concentrations in tissues, enabling accurate risk assessment of bioaccumulative pollutants.

[0047] A41, Risk Quotient Calculation Unit: used to calculate the risk quotient according to 、 and Calculate the RQ value and identify high-risk pollutants with RQ>1; First, the predicted no-effect concentration was determined based on the toxicity prediction value , the specific formula is:

[0048] Among them, the evaluation factor It is a safety factor commonly used in ecological risk assessment and is used to protect rare and endangered species. The risk quotient (RQ) is then calculated. The specific expression is:

[0049] when It is judged as high risk at the time, realizing automatic identification of high-risk pollutants.

[0050] The second aspect is a flow chart of a method for predicting the chronic toxicity of bioaccumulative pollutants proposed by the present invention. Specifically, it may include: S1. Obtain residual toxicity data in biological tissues and related pollutant characteristic data, biological characteristic data, and exposure condition data, and construct a data set after standardization; S2. constructing a multidimensional feature matrix based on the data set, and using the multidimensional feature matrix to obtain a toxicity prediction model through optimization processing using a machine learning algorithm; S3. Using the toxicity prediction model, predict the residual toxicity of bioaccumulative pollutants in tissues at different life stages of organisms and output a toxicity prediction value; S4. Calculate a risk quotient based on the toxicity prediction value and generate a risk assessment report.

[0051] Example 2 The embodiment of the present invention can be applied to the prediction of chronic toxicity of bioaccumulative pollutants to various organisms. In this second embodiment, only the chronic developmental toxicity of bioaccumulative pollutants to fish is used as an example for a specific description.

[0052] A risk assessment based on chronic developmental toxicity was conducted in June 2023 for 22 bioaccumulative flame retardants in Yibin (YB), Yichang (YC), Poyang Lake (PY), and Shanghai (SH), China, targeting 98 rare and endangered fish species in the Yangtze River Basin, including typical species such as the Chinese sturgeon, Gobius crucianus, and Trichuris longissima. Specific steps may include: Table 1 Point table

[0053] Developmental toxicity data on chemical pollutants to freshwater fish based on tissue residue concentrations (NOEC) and LOEC were preliminarily screened from the Ecotoxicology Knowledge Base (ECOTOX) and Environmental Residue-Effect Database (ERED) databases. The NOEC and LOEC data of the same effect were integrated using the geometric mean method, resulting in a dataset of 606 data points for 87 substances on 36 species (endangered fish accounted for 15.3%).

[0054] The constructed multidimensional matrix includes the physical and chemical parameters of pollution (molecular fingerprints, BAF), species biological parameters (body length, weight, life stage and trophic level), experimental conditions (duration of exposure) and toxic effects.

[0055] Data preprocessing included encoding categorical variables, performing log10 transformation on numerical variables, and performing minimum-maximum normalization on the dataset. Categorical variables included life stage and toxicity effects, which were digitally encoded using label encoding and one-hot encoding, respectively.

[0056] Build with Python Predictive model, and through Traversing parameter combinations , find the parameter configuration that can optimize the performance of the XGBoost model (i.e., the coefficient of determination R²) in the above dataset; that is, through R 2 Characterize the predictive ability of the model, including model prediction accuracy, robustness and applicability, such as the test set , 10-fold cross validation , external validation ; Also, identify key features through SHAP analysis, such as the weight of 0.21 for body weight, to clarify the key features that affect the model output.

[0057] Predicted the chronic developmental toxicity of 22 flame retardants to 98 rare and endangered fish species in the Yangtze River. The results showed that embryos / juveniles were 2-10 times more sensitive than adults ( ), small fish such as the blunt-headed catfish family and the thrips family are highly sensitive to pollutants.

[0058] Based on this assessment of the ecological risk of important habitats to rare and endangered fish, it was found that the detection rate of CDPP in the water bodies of the Yangtze River and Shanghai was as high as 100%, and the highest concentration was detected in SH-5, which was 0.17ng / L. , confirming that the risk of SH-5 points exceeds the safety threshold by 10 times. Shanghai Average , 4 times the safety threshold. Generate risk ranking for the Yangtze River Basin , it is recommended to implement stricter water quality monitoring for water bodies during the breeding and juvenile rearing periods, establish protected areas for rare and endangered fish, especially sensitive fish, and give priority to the control of aromatic OPFRs.

[0059] This example reveals the high risk of aromatic OPFRs such as CDPP to endangered fish in the Yangtze River, and provides accurate assessment results based on tissue residues for the protection of rare and endangered species worldwide.

[0060] In summary, the present invention has the following advantages: 1. By constructing a data set based on tissue residue concentrations and combining the geometric mean method and automatic class balance algorithm, the assessment bias caused by water exposure data was resolved; 2. Based on The model constructs a multi-dimensional feature matrix and passes Optimization and dual validation achieved high-precision prediction of chronic toxicity and successfully identified the embryonic / juvenile stage as a highly sensitive stage; 3. Through SHAP feature analysis and data balancing algorithm, the prediction accuracy of rare and endangered species is improved, sensitive features are accurately identified, high-risk pollutants are located, and a quantitative decision-making basis is provided for the protection of rare and endangered species.

[0061] The above specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the scope of the technical solutions of the present invention, and all of these should be included in the scope of the claims of the present invention.

Claims

1. A chronic toxicity prediction system for bioaccumulative pollutants, characterized in that: include: Data acquisition and processing module: used to obtain biological tissue residual toxicity data and related pollutant characteristic data, biological characteristic data and exposure condition data, and construct a data set after standardization; Model training and optimization module: used to construct a multi-dimensional feature matrix based on the data set, and use the multi-dimensional feature matrix to obtain a toxicity prediction model through machine learning algorithm optimization processing; Toxicity prediction module: used to predict the residual toxicity of bioaccumulative pollutants in the tissues of different life stages of organisms using the toxicity prediction model and output toxicity prediction values; Ecological risk assessment module: used to calculate the risk quotient based on the toxicity prediction value and generate a risk assessment report.

2. The bioaccumulative pollutant chronic toxicity prediction system according to claim 1, characterized in that: The data acquisition and processing module includes: Data collection unit: used to obtain tissue residual toxicity data of chronic toxicity endpoints including no observed effect concentration and lowest observed effect concentration from multi-source databases, as well as related pollutant characteristic data, biological characteristic data and exposure condition data; Data preprocessing unit: used to encode categorical variables, normalize numerical data, and use automatic class balancing algorithm to deal with data imbalance problems.

3. The bioaccumulative pollutant chronic toxicity prediction system according to claim 1, characterized in that: The machine learning algorithm is the extreme gradient boosting algorithm.

4. The bioaccumulative pollutant chronic toxicity prediction system according to claim 3, characterized in that: Also includes: pass The additive explanatory analysis determines the key feature weights, where the key features refer to characteristic variables in the multidimensional feature matrix that have a significant impact on the prediction results of the chronic toxicity of bioaccumulative pollutants.

5. The chronic toxicity prediction system for bioaccumulative pollutants according to claim 3, characterized in that: The model training and optimization module includes: A feature engineering unit, used to construct a multi-dimensional feature matrix of the physical and chemical properties of pollution, biological characteristics of species, and experimental conditions based on the dataset; Model training unit, used to Perform 5-fold cross validation to optimize the parameters of the extreme gradient boosting algorithm, and achieve fitting accuracy processing for the toxicity prediction model through the test set; A model validation unit is used to validate the toxicity prediction model through 10-fold cross validation and external validation after the fitting accuracy processing.

6. The chronic toxicity prediction system for bioaccumulative pollutants according to claim 5, characterized in that: Said through Perform 5-fold cross validation to optimize the parameters of the extreme gradient boosting algorithm, including: Define the parameter search space as: ,in and Satisfy the empirical correlation: The mean square error and the coefficient of determination were calculated using 5-fold cross validation. The formula for calculating the mean square error is: The formula for calculating the coefficient of determination is: in, is the true toxicity value, is the predicted value, is the mean; the optimal parameter is determined to be .

7. The bioaccumulative pollutant chronic toxicity prediction system according to claim 1, characterized in that: The toxicity prediction module includes: a toxicity prediction unit, configured to input pollutant structure, species characteristics, and exposure conditions into the toxicity prediction model and output a predicted value of tissue residual toxicity; The sensitive feature identification unit is used to identify sensitive features in the tissue residual toxicity prediction model, where the sensitive features refer to features related to the highly sensitive response of an organism to the chronic toxicity of bioaccumulative pollutants.

8. The chronic toxicity prediction system for bioaccumulative pollutants according to claim 7, characterized in that: The toxicity prediction unit verifies the prediction results by 100 times of bootstrap resampling technology.

9. The bioaccumulative pollutant chronic toxicity prediction system according to claim 1, characterized in that: The ecological risk assessment module includes: The risk quotient calculation unit is used to calculate the risk quotient value based on the predicted no-effect concentration and the measured concentration in the organism, and identify high-risk pollutants with RQ>1.

10. A method for predicting the chronic toxicity of bioaccumulative pollutants, characterized in that: include: S1. Obtain residual toxicity data in biological tissues and related pollutant characteristic data, biological characteristic data, and exposure condition data, and construct a data set after standardization; S2. constructing a multidimensional feature matrix based on the data set, and using the multidimensional feature matrix to obtain a toxicity prediction model through optimization processing using a machine learning algorithm; S3. Using the toxicity prediction model, predict the residual toxicity of bioaccumulative pollutants in tissues at different life stages of organisms and output a toxicity prediction value; S4. Calculate a risk quotient based on the toxicity prediction value and generate a risk assessment report.

Images