Rapid detection system and method based on pesticide toxicity
By integrating multi-dimensional features and using a gradient boosting tree model, a rapid pesticide toxicity detection system is constructed, which solves the problems of high cost and low accuracy in traditional pesticide toxicity assessment. It achieves efficient and accurate toxicity prediction and experimental design, and is suitable for pesticide registration and research and development.
Patent Information
- Application Number
- CN202511660844.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-13
- Publication Date
- 2026-02-27
AI Technical Summary
Traditional pesticide toxicity assessment methods are costly, time-consuming, have limited predictive models, and lack auxiliary tools, resulting in insufficient prediction accuracy and blind experimental design, making it impossible to accurately assess the potential risks of pesticides to soil ecology.
A rapid pesticide toxicity detection system employing multi-dimensional feature integration includes modules for data acquisition, preprocessing, multi-dimensional feature extraction, prediction model construction, and visualization-assisted design. It constructs a toxicity prediction model through gradient boosting trees and combines dose-toxicity response curves and experimental dose recommendation heatmaps to assist in dose design.
It improves the accuracy of LC50 value prediction, precisely captures the impact of soil environmental parameters on toxicity, reduces experimental costs, solves the blindness of traditional methods, and is applicable to toxicity prediction of other organisms and toxicity screening in the pesticide development stage.
Smart Images
Figure CN121583393A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of pesticide detection and analysis technology, and in particular to a rapid pesticide toxicity detection system and method. Background Technology
[0002] As a key material for controlling pests, diseases and weeds in agricultural production, pesticides must undergo strict toxicity assessments before being registered and marketed, especially toxicity tests on representative organisms of the soil ecosystem. Earthworms, as the core organisms for the decomposition of soil organic matter, directly reflect the potential risks of pesticides to the soil ecosystem. Relevant toxicity data is one of the core indicators for pesticide ecological risk assessment.
[0003] Currently, pesticide toxicity assessment mainly relies on traditional animal testing methods: by setting up multiple dose gradient pesticide exposure tests, observing earthworm mortality rates and calculating LC50 values, and then classifying toxicity levels. However, this method has significant drawbacks: First, the testing costs are high, requiring a large number of pesticide samples, test containers, and manpower. The earthworm toxicity test cycle for a single pesticide can take as long as 20-30 days. Secondly, there is a serious waste of resources. If the actual toxicity of the pesticide is extremely low, but the experiment is designed according to a high dose gradient, it will lead to the ineffective consumption of reagents and time. Third, the prediction models have significant limitations. Some existing toxicity prediction models are based on only single-dimensional data and do not integrate key influencing factors such as the physicochemical properties of pesticides and experimental environmental parameters, resulting in insufficient prediction accuracy, with an accuracy rate usually below 75%. Fourth, there is a lack of auxiliary tools. Existing technologies cannot provide visual support for the dose design of registration trials. Trial personnel need to set the dose range based on experience, which can easily lead to problems such as the dose being too high, causing all earthworms to die, or the dose being too low, making it impossible to accurately calculate LC50. The trial plan needs to be adjusted repeatedly, further extending the cycle. To address the aforementioned technical deficiencies, a solution is proposed. Summary of the Invention
[0004] The purpose of this invention is to improve the accuracy of LC50 value prediction by integrating multi-dimensional features, precisely capturing the impact of soil environmental parameters on toxicity, and making the prediction results more aligned with the actual needs of pesticide ecological risk assessment. Furthermore, by using visualization tools to directly output the dose-toxicity response relationship and recommended dose range, it overcomes the blindness of traditional empirical dose design and can be transferred to the toxicity prediction of other organisms. New models can be built simply by replacing the corresponding historical experimental data. It can also be applied to toxicity screening in the pesticide development stage.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: a rapid pesticide toxicity detection system, comprising a data acquisition module, a data preprocessing module, a multi-dimensional feature extraction module, a prediction model construction module, a toxicity level prediction module, and a visualization-aided design module, wherein: The data acquisition module is used to acquire historical basic data and test data and send them to the data preprocessing module. The historical basic data includes the chemical structure information, physicochemical property parameters, test condition parameters and corresponding earthworm toxicity test results of registered pesticides. The test data includes the chemical structure, physicochemical properties and preset test conditions of new pesticides to be evaluated. The data preprocessing module is used to clean and standardize the collected data and then integrate it into an optimized dataset before sending it to the multi-dimensional feature extraction module. The multi-dimensional feature extraction module is used to obtain and optimize the dataset, extract key features for different types of data to achieve multi-dimensional information fusion, integrate them into a multi-dimensional feature set, and send it to the prediction model building module.
[0006] The prediction model building module constructs a toxicity prediction model based on a gradient boosting tree. The toxicity prediction model takes multi-dimensional features as input and LC50 value and corresponding toxicity level as output. The toxicity level prediction module acquires the basic characteristic data of the new pesticide to be tested, standardizes its physicochemical properties and test conditions through the data preprocessing module, and generates multi-dimensional features to be verified through the feature extraction module. The multi-dimensional features to be verified are input into the trained toxicity prediction model, and the predicted LC50 value and the corresponding toxicity level matched according to the predicted LC50 value are output. The results are integrated into a toxicology report and sent to the visualization auxiliary design module. The visualization-assisted design module is used to obtain toxicology reports, generate visualization tools, and assist in the dosage design of pesticide registration trials. The visualization tools include dose-toxicity response curves and heatmaps of recommended test doses.
[0007] Furthermore, the specific process for cleaning and standardizing the collected data is as follows: S101. Missing value handling: The mean imputation method is used to supplement the numerical missing values in the physicochemical property parameters, and the mode imputation method is used to supplement the categorical missing values in the experimental condition parameters. S102. Outlier handling: Identify and remove outliers that exceed the preset range using the interquartile range method; S103. Data standardization: Z-score standardization is used for physicochemical property parameters to eliminate dimensional differences; S104. Toxicity Classification: Toxicity levels are determined based on the preset LC50 value range of toxicological parameters, including highly toxic, extremely toxic, moderately toxic, and low toxic.
[0008] Specifically: highly toxic (LC50 ≤ 1 mg / kg), highly toxic (1 < LC50 ≤ 10 mg / kg), moderately toxic (10 < LC50 ≤ 100 mg / kg), and low toxic (LC50 > 100 mg / kg), corresponding to label values of 0 - 3 for model training.
[0009] Furthermore, the specific process of extracting key features is as follows: Chemical structure features: Use the RDKit cheminformatics tool to extract 2D molecular fingerprints, topological indices, and functional group features of pesticide molecules; Physical and chemical property features: Select standardized solubility, vapor pressure, and octanol - water partition coefficient Kow as features; Test condition features: Standardize temperature, humidity, and exposure duration as features, and simultaneously construct a temperature - humidity interaction feature, i.e., the product of temperature and humidity, to capture the synergistic effects of environmental parameters; Feature screening: Use recursive feature elimination to remove redundant features and retain the top 500 features that have a significant impact on the LC50 value to reduce the model complexity.
[0010] Furthermore, the specific process of constructing a toxicity prediction model is as follows: S201. Obtain historical baseline data in the optimized dataset as training samples and divide them into a training set and a test set at a ratio of 7:3; S202. Use grid search and 5 - fold cross - validation to optimize key parameters and determine the optimal parameters including learning rate, tree depth, number of estimators, and regularization coefficient; S203. Fit the model using the training set, optimize the prediction accuracy of the LC50 value with the goal of minimizing the mean squared error, and optimize the classification accuracy of the toxicity level with cross - entropy loss; S204. Verify the model performance on the test set until the coefficient of determination of the LC50 value prediction is greater than or equal to the preset coefficient threshold, and at the same time the classification accuracy of the toxicity level is greater than the preset judgment threshold, then obtain the final toxicity prediction model.
[0011] [[ID= 25]]Furthermore, the dose - toxicity response curve specifically uses the pesticide exposure dose as the horizontal axis and the predicted mortality of earthworms as the vertical axis. By simulating the mortality changes at different doses through the model, automatically mark the dose points corresponding to the LC50 value to visually display the relationship between toxicity and dose; its specific generation process is as follows: The generation of the dose - toxicity response curve specifically includes the following steps: S401: Obtain the center point: Use the predicted value in the toxicological report as the reference point, and the reference point is defined as the predicted concentration of 50% mortality of the test organisms; S402: Determine the curve shape: Use the standard toxicology fitting function to determine the dose-response relationship in toxicology; S403: Fitting Curve: Based on the gradient boosting tree in the toxicity level prediction module, a slope calculation model is constructed. The output is trained as the slope parameters of the response curve. The toxicology report and the predicted slope parameters are received to generate a more specific dose-response curve for the specific compound and specific test conditions. S404: Rendering output: Generates a two-dimensional chart, allowing users to view the predicted mortality rate corresponding to any concentration point on the curve through interactive operation.
[0012] Furthermore, the recommended test dose heatmap uses temperature and humidity as the coordinate axes, with color intensity indicating the rationality of the recommended test dose range under those environmental conditions, assisting test personnel in quickly determining the optimal test environment and dose combination; its specific generation process is as follows: S501: Receive user input: Allows the user to select two preset variable parameters for sensitivity analysis from preset test conditions; S502: Define parameter range: Set the exploration range based on variable parameters; S503: Lock Other Features: The module keeps the chemical structure and physicochemical properties of the new pesticide under test, as well as other test condition parameters not selected as variables in S501, constant. S504: Batch Iterative Prediction: In the two-dimensional parameter space defined in S502, a parameter combination matrix is constructed according to a preset step size. For each cell in the matrix, the cell is recombined with the constant feature locked in S503 to form a complete multi-dimensional feature to be verified, and the toxicity prediction model is called. The feature is input to obtain the predicted value under the specific condition combination.
[0013] S505: Rendering a heatmap: Traverse all matrix cells to obtain a two-dimensional data matrix containing all predicted values, and render the data matrix as a heatmap.
[0014] This invention also provides a rapid detection method for pesticide toxicity, comprising the following steps: Step 1: Obtain historical basic data and test data. The historical basic data includes the chemical structure information, physicochemical property parameters, test condition parameters and corresponding earthworm toxicity test results of registered pesticides. The test data includes the chemical structure, physicochemical properties and preset test conditions of the new pesticide to be evaluated. Step 2: After cleaning and standardizing the collected data, integrate it into an optimized dataset; Step 3: Obtain the optimized dataset, extract key features for different types of data to achieve multi-dimensional information fusion, and integrate them into a multi-dimensional feature set; Step 4: Construct a toxicity prediction model based on gradient boosting tree. The toxicity prediction model takes multi-dimensional features as input and LC50 value and corresponding toxicity level as output. Step 5: Obtain the basic characteristic data of the new pesticide to be tested, standardize its physicochemical properties and test conditions through the data preprocessing module, generate multi-dimensional features to be verified through the feature extraction module, input the multi-dimensional features to be verified into the trained toxicity prediction model, output the predicted LC50 value and the corresponding toxicity level matched according to the predicted LC50 value, and integrate them into a toxicology report. Step 6: Obtain toxicology reports and generate visualization tools to assist in the dosage design of pesticide registration trials. The visualization tools include dose-toxicity response curves and heatmaps of recommended test doses.
[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. Based on the rapid detection system and method for pesticide toxicity, this invention can prioritize the exclusion of invalid tests for pesticides with extremely low or extremely high toxicity through pre-toxicity assessment before registration, thereby reducing test costs. Furthermore, by integrating multi-dimensional features, the accuracy of LC50 value prediction results is improved, which can accurately capture the impact of soil environmental parameters on toxicity, and the prediction results are more in line with the actual needs of pesticide ecological risk assessment.
[0016] 2. This rapid pesticide toxicity detection system and method uses visualization tools to directly output the dose-toxicity response relationship and recommended dose range, which solves the blindness of traditional empirical dose design and can be transferred to the toxicity prediction of other organisms. Only the corresponding historical test data needs to be replaced to build a new model. It can also be applied to the toxicity screening in the pesticide research and development stage. Attached Figure Description
[0017] Figure 1 A schematic diagram of the overall external structure of the present invention is shown; Figure 2 This diagram shows the overall external structure of the present invention from another angle. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Example 1: like Figure 1As shown in the figure, a rapid detection system for pesticide toxicity includes a data acquisition module, a data preprocessing module, a multi-dimensional feature extraction module, a prediction model construction module, a toxicity level prediction module, and a visualization auxiliary design module, where: The data acquisition module is used to obtain historical basic data and待测 data and send them to the data preprocessing module. The historical basic data includes the chemical structure information, physical and chemical property parameters, test condition parameters of registered pesticides, and the corresponding earthworm toxicity test results. The待测 data includes the chemical structure, physical and chemical properties, and preset test conditions of the new pesticide to be evaluated. The data preprocessing module is used to clean and standardize the collected data and then integrate it into an optimized data set and send it to the multi-dimensional feature extraction module. The specific process of cleaning and standardizing the collected data is as follows: S101. Missing value processing: The mean filling method is used to supplement the numerical missing values in the physical and chemical property parameters, and the mode filling method is used to supplement the categorical missing values in the test condition parameters. S102. Outlier processing: The interquartile range method is used to identify and remove outliers outside the preset range. S103. Data standardization: The Z-score standardization is used for the physical and chemical property parameters to eliminate the dimensional difference. S104. Toxicity level coding: The toxicity level is determined according to the preset toxicological parameter LC50 value range, including highly toxic, highly toxic, moderately toxic, and low toxic.
[0020] Specifically: Highly toxic (LC50 ≤ 1mg / kg), highly toxic (1 < LC50 ≤ 10mg / kg), moderately toxic (10 < LC50 ≤ 100mg / kg), low toxic (LC50 > 100mg / kg), and the corresponding encoded label values are 0-3 for model training.
[0021] The multi-dimensional feature extraction module is used to obtain the optimized data set, extract key features for different types of data to achieve multi-dimensional information fusion, and integrate them into a multi-dimensional feature set and send it to the prediction model construction module.
[0022] The specific process of extracting key features is as follows: Chemical structure features: The RDKit chemical informatics tool is used to extract the 2D molecular fingerprint, topological index, and functional group features of the pesticide molecule. Physical and chemical property features: The standardized solubility, vapor pressure, and octanol-water partition coefficient Kow are selected as features. Test condition features: The temperature, humidity, and exposure duration are standardized as features, and at the same time, a temperature-humidity interaction feature, that is, the product of temperature and humidity, is constructed to capture the synergistic effect of environmental parameters. Feature selection: Recursive feature elimination method is used to remove redundant features and retain the top 500 features that have a significant impact on the LC50 value, thereby reducing model complexity.
[0023] The prediction model building module constructs a toxicity prediction model based on gradient boosting trees. The toxicity prediction model takes multi-dimensional features as input and outputs LC50 values and corresponding toxicity levels. The specific process of constructing a toxicity prediction model is as follows: S201. Obtain historical basic data from the optimization dataset as training samples, and divide it into training set and test set in a 7:3 ratio; S202. The key parameters were optimized using grid search and 5-fold cross-validation. The optimal parameters were determined, including the learning rate, tree depth, number of estimators, and regularization coefficient. S203. Fit the model using the training set, optimize the LC50 value prediction accuracy with the goal of minimizing the mean squared error, and optimize the toxicity level classification accuracy with cross-entropy loss. S204. Validate the model performance on the test set until the determination coefficient of the LC50 value prediction is greater than or equal to the preset coefficient threshold, and the accuracy of the toxicity level classification is greater than the preset judgment threshold. Then the final toxicity prediction model is obtained.
[0024] The toxicity level prediction module acquires the basic characteristic data of the new pesticide to be tested, standardizes its physicochemical properties and test conditions through the data preprocessing module, and generates multi-dimensional features to be verified through the feature extraction module. The multi-dimensional features to be verified are input into the trained toxicity prediction model, and the predicted LC50 value and the corresponding toxicity level matched according to the predicted LC50 value are output. The results are integrated into a toxicology report and sent to the visualization auxiliary design module. The visualization-assisted design module is used to obtain toxicology reports, generate visualization tools, and assist in the dosage design of pesticide registration trials. The visualization tools include dose-toxicity response curves and heatmaps of recommended test doses.
[0025] The dose-toxicity response curve uses pesticide exposure dose as the horizontal axis and predicted earthworm mortality rate as the vertical axis. It simulates mortality rate changes under different doses using a model, automatically labeling dose points corresponding to LC50 values to visually demonstrate the relationship between toxicity and dose. The specific generation process is as follows: The generation of dose-toxicity response curves includes the following steps: S401: Obtain the center point: Use the predicted value in the toxicology report as the baseline point, which is defined as the predicted concentration at which 50% mortality of the test organism occurs. S402: Determine the curve shape: Use the standard toxicology fitting function to determine the dose-response relationship in toxicology; S403: Fitting Curve: Based on the gradient boosting tree in the toxicity level prediction module, a slope calculation model is constructed. The output is trained as the slope parameters of the response curve. The toxicology report and the predicted slope parameters are received to generate a more specific dose-response curve for the specific compound and specific test conditions. S404: Rendering output: Generates a two-dimensional chart, allowing users to view the predicted mortality rate corresponding to any concentration point on the curve through interactive operation.
[0026] The recommended test dose heatmap uses temperature and humidity as the coordinate axes, with color intensity indicating the rationality of the recommended test dose range under those environmental conditions. This helps test personnel quickly determine the optimal test environment and dose combination. Its specific generation process is as follows: S501: Receive user input: Allows the user to select two preset variable parameters for sensitivity analysis from preset test conditions; S502: Define parameter range: Set the exploration range based on variable parameters; S503: Lock Other Features: The module keeps the chemical structure and physicochemical properties of the new pesticide under test, as well as other test condition parameters not selected as variables in S501, constant. S504: Batch Iterative Prediction: In the two-dimensional parameter space defined in S502, a parameter combination matrix is constructed according to a preset step size. For each cell in the matrix, the cell is recombined with the constant feature locked in S503 to form a complete multi-dimensional feature to be verified, and the toxicity prediction model is called. The feature is input to obtain the predicted value under the specific condition combination.
[0027] S505: Rendering a heatmap: Traverse all matrix cells to obtain a two-dimensional data matrix containing all predicted values, and render the data matrix as a heatmap.
[0028] This invention, through pre-registration toxicity assessment, can prioritize the exclusion of ineffective tests for pesticides with extremely low or high toxicity. Furthermore, by integrating multi-dimensional features, the LC50 value prediction results are more accurate, precisely capturing the impact of soil environmental parameters on toxicity. The prediction results are more in line with the actual needs of pesticide ecological risk assessment. At the same time, the invention uses visualization tools to directly output the dose-toxicity response relationship and recommended dose range, solving the blindness of traditional empirical dose design. It can also be transferred to the toxicity prediction of other organisms, and new models can be built simply by replacing the corresponding historical test data. It can also be applied to toxicity screening in the pesticide research and development stage.
[0029] Example 2: like Figure 2 As shown, the present invention also provides a rapid detection method based on pesticide toxicity, comprising the following steps: Step 1: Obtain historical basic data and data to be tested. Historical basic data includes the chemical structure information, physicochemical property parameters, test condition parameters and corresponding earthworm toxicity test results of registered pesticides. Data to be tested includes the chemical structure, physicochemical properties and preset test conditions of the new pesticide to be evaluated. Step 2: After cleaning and standardizing the collected data, integrate it into an optimized dataset; Step 3: Obtain the optimized dataset, extract key features for different types of data to achieve multi-dimensional information fusion, and integrate them into a multi-dimensional feature set; Step 4: Construct a toxicity prediction model based on gradient boosting tree. The toxicity prediction model takes multi-dimensional features as input and LC50 value and corresponding toxicity level as output. Step 5: Obtain the basic characteristic data of the new pesticide to be tested, standardize its physicochemical properties and test conditions through the data preprocessing module, generate multi-dimensional features to be verified through the feature extraction module, input the multi-dimensional features to be verified into the trained toxicity prediction model, output the predicted LC50 value and the corresponding toxicity level matched according to the predicted LC50 value, and integrate them into a toxicology report. Step 6: Obtain toxicology reports and generate visualization tools to assist in the dosage design of pesticide registration trials. The visualization tools include dose-toxicity response curves and heatmaps of recommended test doses.
[0030] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.
[0031] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation. In the two embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways; for example, the device embodiments described above are merely illustrative, and the division of modules is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed; another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or modules can be electrical, mechanical or other forms. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A rapid pesticide toxicity detection system, characterized in that, It includes a data acquisition module, a data preprocessing module, a multi-dimensional feature extraction module, a prediction model construction module, a toxicity level prediction module, and a visualization-aided design module, among which: The data acquisition module is used to acquire historical basic data and test data and send them to the data preprocessing module. The historical basic data includes the chemical structure information, physicochemical property parameters, test condition parameters and corresponding earthworm toxicity test results of registered pesticides. The test data includes the chemical structure, physicochemical properties and preset test conditions of new pesticides to be evaluated. The data preprocessing module is used to clean and standardize the collected data and then integrate it into an optimized dataset before sending it to the multi-dimensional feature extraction module. The multi-dimensional feature extraction module is used to obtain the optimized dataset, extract key features for different types of data to achieve multi-dimensional information fusion, integrate them into a multi-dimensional feature set and send it to the prediction model building module; The prediction model building module constructs a toxicity prediction model based on a gradient boosting tree. The toxicity prediction model takes multi-dimensional features as input and LC50 value and corresponding toxicity level as output. The toxicity level prediction module acquires the basic characteristic data of the new pesticide to be tested, standardizes its physicochemical properties and test conditions through the data preprocessing module, and generates multi-dimensional features to be verified through the feature extraction module. The multi-dimensional features to be verified are input into the trained toxicity prediction model, and the predicted LC50 value and the corresponding toxicity level matched according to the predicted LC50 value are output. The results are integrated into a toxicology report and sent to the visualization auxiliary design module. The visualization-assisted design module is used to obtain toxicology reports, generate visualization tools, and assist in the dosage design of pesticide registration trials. The visualization tools include dose-toxicity response curves and heatmaps of recommended test doses.
2. The rapid pesticide toxicity detection system according to claim 1, characterized in that, The specific process of cleaning and standardizing the collected data is as follows: S101. Missing value handling: The mean imputation method is used to supplement the numerical missing values in the physicochemical property parameters, and the mode imputation method is used to supplement the categorical missing values in the experimental condition parameters. S102. Outlier handling: Identify and remove outliers that exceed the preset range using the interquartile range method; S103. Data standardization: Z-score standardization is used for physicochemical property parameters to eliminate dimensional differences; S104. Toxicity Classification: Toxicity levels are determined based on the preset LC50 value range of toxicological parameters, including highly toxic, extremely toxic, moderately toxic, and low toxic.
3. The rapid pesticide toxicity detection system according to claim 1, characterized in that, The specific process for extracting key features is as follows: Chemical structure characteristics: The RDKit cheminformatics tool was used to extract the 2D molecular fingerprint, topological index and functional group characteristics of pesticide molecules; Physicochemical properties: Standardized solubility, vapor pressure, and octanol-water partition coefficient Kow were selected as characteristics; Experimental condition characteristics: Temperature, humidity, and exposure duration were standardized and used as features. At the same time, temperature-humidity interaction features were constructed, that is, the product of temperature and humidity, to capture the synergistic influence of environmental parameters. Feature selection: A recursive feature elimination method is used to remove redundant features and retain the top 500 features that have a significant impact on the LC50 value, thereby reducing model complexity.
4. The rapid pesticide toxicity detection system according to claim 1, characterized in that, The specific process of constructing a toxicity prediction model is as follows: S201. Obtain historical basic data from the optimization dataset as training samples, and divide it into training set and test set in a 7:3 ratio; S202. The key parameters were optimized using grid search and 5-fold cross-validation. The optimal parameters were determined, including the learning rate, tree depth, number of estimators, and regularization coefficient. S203. Fit the model using the training set, optimize the LC50 value prediction accuracy with the goal of minimizing the mean squared error, and optimize the toxicity level classification accuracy with cross-entropy loss. S204. Validate the model performance on the test set until the determination coefficient of the LC50 value prediction is greater than or equal to the preset coefficient threshold, and the accuracy of the toxicity level classification is greater than the preset judgment threshold. Then the final toxicity prediction model is obtained.
5. The rapid pesticide toxicity detection system according to claim 1, characterized in that, The dose-toxicity response curve specifically uses pesticide exposure dose as the horizontal axis and predicted earthworm mortality rate as the vertical axis. It simulates mortality rate changes under different doses using a model, automatically labels the dose points corresponding to the LC50 value, and visually demonstrates the relationship between toxicity and dose. The specific generation process is as follows: The generation of the dose-toxicity response curve specifically includes the following steps: S401: Obtain the center point: Use the predicted value in the toxicology report as the benchmark point, which is defined as the predicted concentration at which the mortality rate of the test organism is 50%. S402: Determine the curve shape: Use the standard toxicology fitting function to determine the dose-response relationship in toxicology; S403: Fitting Curve: Based on the gradient boosting tree in the toxicity level prediction module, a slope calculation model is constructed. The output is trained as the slope parameters of the response curve. The toxicology report and the predicted slope parameters are received to generate a more specific dose-response curve for the specific compound and specific test conditions. S404: Rendering output: Generates a two-dimensional chart, allowing users to view the predicted mortality rate corresponding to any concentration point on the curve through interactive operation.
6. The rapid pesticide toxicity detection system according to claim 1, characterized in that, The recommended test dose heatmap uses temperature and humidity as the coordinate axes, with color intensity indicating the rationality of the recommended test dose range under those environmental conditions. This helps test personnel quickly determine the optimal test environment and dose combination. Its specific generation process is as follows: S501: Receive user input: Allows the user to select two preset variable parameters for sensitivity analysis from preset test conditions; S502: Define parameter range: Set the exploration range based on variable parameters; S503: Lock Other Features: The module keeps the chemical structure and physicochemical properties of the new pesticide under test, as well as other test condition parameters not selected as variables in S501, constant. S504: Batch Iterative Prediction: In the two-dimensional parameter space defined in S502, a parameter combination matrix is constructed according to a preset step size. For each cell in the matrix, the cell is recombined with the constant feature locked in S503 to form a complete multi-dimensional feature to be verified, and the toxicity prediction model is called. The feature is input to obtain the predicted value under the specific condition combination. S505: Rendering a heatmap: Traverse all matrix cells to obtain a two-dimensional data matrix containing all predicted values, and render the data matrix as a heatmap.
7. A rapid pesticide toxicity detection method, applied to a rapid pesticide toxicity detection system as described in claims 1-6, characterized in that, Includes the following steps: Step 1: Obtain historical basic data and test data. The historical basic data includes the chemical structure information, physicochemical property parameters, test condition parameters and corresponding earthworm toxicity test results of registered pesticides. The test data includes the chemical structure, physicochemical properties and preset test conditions of the new pesticide to be evaluated. Step 2: After cleaning and standardizing the collected data, integrate it into an optimized dataset; Step 3: Obtain the optimized dataset, extract key features for different types of data to achieve multi-dimensional information fusion, and integrate them into a multi-dimensional feature set; Step 4: Construct a toxicity prediction model based on gradient boosting tree. The toxicity prediction model takes multi-dimensional features as input and LC50 value and corresponding toxicity level as output. Step 5: Obtain the basic characteristic data of the new pesticide to be tested, standardize its physicochemical properties and test conditions through the data preprocessing module, generate multi-dimensional features to be verified through the feature extraction module, input the multi-dimensional features to be verified into the trained toxicity prediction model, output the predicted LC50 value and the corresponding toxicity level matched according to the predicted LC50 value, and integrate them into a toxicology report. Step 6: Obtain toxicology reports and generate visualization tools to assist in the dosage design of pesticide registration trials. The visualization tools include dose-toxicity response curves and heatmaps of recommended test doses.