Method for predicting acute toxicity of organic compounds to gobiocypris rarus and storage medium
By constructing a machine learning model and utilizing toxicity data from rare gudgeon and other fish species, the acute toxicity of organic compounds to rare gudgeon is predicted. This solves the accuracy problem of existing QSAR software in predicting the toxicity of rare gudgeon, achieving efficient and accurate acquisition of toxicity data, and is suitable for ecotoxicology and environmental risk assessment.
Patent Information
- Application Number
- CN202410952752.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-12
AI Technical Summary
Existing QSAR software is not accurate enough in predicting the acute toxicity of rare gudgeon tested in my country, which affects the environmental risk assessment results of chemical substances, especially the prediction ability of organic compounds.
We constructed a binary classification and regression model based on machine learning. Using acute toxicity data of rare gudgeon, blackhead softmouth minnow, and zebrafish, we calculated molecular descriptors and fingerprints using PaDEL-Descriptor. Combined with algorithms such as support vector machine and gradient boosting, we predicted the acute toxicity category and LC50 value of chemical substances to rare gudgeon.
This study provides an efficient and accurate method for predicting the acute toxicity of organic compounds to rare gudgeon, improving the quality and efficiency of toxicity data acquisition, and is applicable to the fields of ecotoxicology and environmental risk assessment.
Smart Images

Figure CN118942573A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a toxicity prediction method, in particular to a prediction method and a storage medium for the acute toxicity of an organic compound to rare goby. Background Art
[0002] In the environmental risk assessment of new chemicals, chemical pollutants and other chemical substances in my country, as well as the calculation of water quality benchmarks, according to the requirements of my country's Ministry of Ecology and Environment and other management agencies, it is necessary to use my country's test organisms, such as the rare minnow, to complete fish toxicity tests.
[0003] Obtaining acute toxicity data of rare goby through standard test methods is costly and time-consuming and has certain limitations. Therefore, as an alternative method, quantitative structure-activity relationship (QSAR) is increasingly being used in obtaining toxicity data.
[0004] However, the existing fish acute toxicity prediction software, such as the ECOSAR software developed by the U.S. EPA, which is commonly used for aquatic toxicity prediction, is developed based on the toxicity data of non-Chinese test organisms such as zebrafish and rainbow trout, and has poor acute toxicity prediction capabilities for rare goby, a native Chinese organism. For example, researchers have found that the 96h half-lethal concentration (96h-LC50) of 4-acetylaminoantipyrine for rare goby is ≥1531mg / L, which is much higher than the estimated value of 6.05mg / L by the ECOSAR software (Liu Min et al., 2020). In addition, studies have found that rare goby is more sensitive to the ecotoxicological effects of pollutants than zebrafish.
[0005] In summary, it can be seen that directly using existing QSAR software to predict the acute toxicity of rare goby crucian carp in my country will result in inaccurate data, which will affect the environmental risk assessment results of chemical substances and is not conducive to protecting my country's ecological environment. There are fewer types of inorganic compounds, and their toxicity data have been basically complete, while organic compounds are complex, and most new chemical substances and chemical pollutants belong to organic compounds. At present, there is a lack of a QSAR method for predicting the acute toxicity of organic compounds to rare goby crucian carp in my country. Summary of the invention
[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a prediction method and storage medium for the acute toxicity of an organic compound to rare goby, so as to achieve efficient prediction of the acute toxicity of the test organism rare goby, which is conducive to accelerating the environmental risk assessment process of new chemical substances and other chemical substances; the present invention is simple and easy to use, easy to get started, and has a wide application range. It can be used to obtain acute toxicity data of rare goby in disciplines such as ecotoxicology and environmental risk assessment.
[0007] The purpose of the present invention can be achieved by the following technical solutions:
[0008] The first aspect of the present invention provides a method for predicting the acute toxicity of an organic compound to rare goby, comprising the following steps:
[0009] S1: Obtain acute toxicity data for rare goby, fathead minnow, and zebrafish;
[0010] S2: Construct a binary classification prediction model. First, divide the data in S1 into two groups: those with acute toxicity and those without acute toxicity. Then use PaDEL-Descriptor software to calculate the molecular descriptors and fingerprints of the chemical structure. After preprocessing, use machine learning methods to establish a binary classification prediction model. Use the binary classification prediction model to predict the acute toxicity category of chemical substances.
[0011] S3: Use the model in S2 to predict and obtain information on acutely toxic chemical substances, further establish a regression model, collect data on specific LC50 values, calculate molecular descriptors, and after preprocessing, use machine learning regression methods to predict the acute toxicity LC50 values of chemical substances to rare crucian carp.
[0012] Furthermore, S1 specifically includes:
[0013] Acute toxicity data of rare goby, fathead minnow and zebrafish were obtained. For rare goby, all available acute toxicity data were collected. For fathead minnow and zebrafish, only data of non-acute toxic substances were collected, i.e., data with half lethal concentration LC50>120mg / L or LC50>saturated solution concentration and saturated solution concentration less than 100mg / L.
[0014] Furthermore, S2 specifically includes:
[0015] S2-1: Data classification and model preparation:
[0016] The data obtained in S1 were divided into two groups according to the presence or absence of acute toxicity. Then, the chemical structure formulas of the chemicals to be predicted were drawn, and their molecular descriptors and molecular fingerprints were calculated using PaDEL-Descriptor software;
[0017] S2-2: Model building and prediction:
[0018] The calculated molecular descriptors and molecular fingerprints were preprocessed, and then a machine learning algorithm was selected to establish a binary classification model based on the preprocessed data. The binary classification model was used to predict the acute toxicity category of chemical substances to rare crucian carp.
[0019] Furthermore, S2-1 specifically includes:
[0020] The collected acute toxicity data are divided into two categories according to whether they have acute toxicity or not, namely the first category with acute toxicity and the second category without acute toxicity, providing classification standards for subsequent model training;
[0021] For the chemical substance to be predicted, its chemical structure formula is drawn, and the PaDEL-Descriptor quantum chemistry software is used to calculate the 1D&2D descriptors and fingerprints of the molecule to generate numerical data describing the molecular characteristics, providing the necessary input features for establishing the prediction model.
[0022] Furthermore, S2-2 specifically includes:
[0023] Clean and filter the molecular descriptors and molecular fingerprint data calculated by PaDEL-Descriptor software to achieve preprocessing;
[0024] A machine learning algorithm was selected, and the preprocessed molecular descriptors and molecular fingerprint datasets were used as input features. The presence or absence of acute toxicity was used as the output result. A binary classification prediction model was trained and established. Subsequently, the binary classification prediction model was used to predict the acute toxicity category of new chemicals to determine whether they were acutely toxic to the rare minnow.
[0025] Further, in S2-2, the cleaning and screening include: removing zero variance and near-zero variance variables to eliminate uninformative features, deleting variables containing null values to ensure data integrity, and removing strongly correlated variables to reduce model complexity and avoid multicollinearity problems;
[0026] The machine learning algorithm is selected from one of a support vector machine, a gradient boosting method, and a neural network.
[0027] Furthermore, S3 specifically includes:
[0028] S3-1: Data collection and preprocessing:
[0029] Obtain information on chemical substances with acute toxicity predicted by the model in S2, and obtain acute toxicity data of rare minnow with specific LC50 values, while removing data of fathead minnow and zebrafish and LC50 data that do not meet the conditions. Then, for the chemical substances to be predicted, calculate molecular descriptors and molecular fingerprints, and preprocess the data;
[0030] S3-2: Regression model establishment and application:
[0031] The specific LC50 value was used as the dependent variable and the preprocessed molecular descriptor as the independent variable. The machine learning regression algorithm was used. The molecular descriptor data of the chemical substance to be predicted was input into the regression model to predict its acute toxicity LC50 value to the rare crucian carp.
[0032] Furthermore, S3-1 specifically includes:
[0033] Obtain information on chemical substances with acute toxicity predicted by the model in S2, and obtain the acute toxicity data of rare goby from literature data and experimental information to ensure that the data has a specific LC50 value. At the same time, exclude the data of fathead minnow and zebrafish, as well as the data with LC50 greater than the saturated solution concentration but less than 100 mg / L;
[0034] For the chemical substances to be predicted, the chemical structure formula is drawn, and the molecular descriptors and molecular fingerprints are calculated using PaDEL-Descriptor software. Then, the calculated data is preprocessed, including removing zero variance and near-zero variance variables, null value variables, strongly correlated variables, and performing necessary standardization to prepare for the training of the regression model.
[0035] Furthermore, S3-2 specifically includes:
[0036] Select a machine learning regression algorithm, including elastic network regression or support vector machine regression, with the collected specific LC50 values as the dependent variable and the preprocessed molecular descriptor data as the independent variable, to train the regression model and adjust and optimize the regression model parameters;
[0037] The trained regression model is applied to the molecular descriptor data of the chemical substance to be predicted, and these data are input into the regression model. The model is used to calculate the predicted LC50 value of the acute toxicity of the chemical substance to rare crucian carp, thereby providing quantitative results for the toxicity assessment of the chemical substance.
[0038] A second aspect of the present invention provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, is used to execute the above-mentioned method for predicting the acute toxicity of organic compounds to rare minnows.
[0039] Compared with the prior art, the present invention has the following technical advantages:
[0040] 1) The present invention applies machine learning methods to the prediction of acute toxicity of chemical substances to the rare goby crucian carp, a test organism in my country, and can provide an efficient and accurate prediction method and tool to improve the quality and efficiency of obtaining acute toxicity data of the rare goby crucian carp;
[0041] 2) The present invention is simple and easy to use, and can be used to obtain acute toxicity data of rare crucian carp in the fields of ecotoxicology, environmental risk assessment, etc.;
[0042] 3) When the present invention is used for a corresponding computer program, the overall program is concise, has low requirements on computer hardware, and is highly practical. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 It is a simplified flow chart of the present invention;
[0044] Figure 2 It is a specific flow chart of the present invention;
[0045] Figure 3 It is a scatter plot of the model predicted values and measured values in the elastic net regression model;
[0046] Figure 4 This is the application domain diagram of the Euclidean distance of the binary classification SVM model;
[0047] Figure 5 Apply domain Williams plot for EN regression model. DETAILED DESCRIPTION
[0048] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms and other features not clearly described in this technical solution are all considered to be common technical features disclosed in the prior art.
[0049] Example 1
[0050] In this embodiment, a method for predicting the acute toxicity of an organic compound to rare crucian carp based on machine learning comprises the following steps:
[0051] 1) Step 1: Collect the acute toxicity data of rare goby obtained from literature and experiments, as well as fathead minnow and zebrafish with similar acclimation temperature, acute toxicity test temperature, test fish body length and sexual maturity period as rare goby. For rare goby, collect all available acute toxicity data, and for fathead minnow and zebrafish, only collect data of non-acute toxic substances, i.e., median lethal concentration (LC50). 50 )>120mg / L or LC 50> Saturated solution concentration and the saturated solution concentration is less than 100 mg / L;
[0052] 2) Step 2: Establish a binary classification model for the prediction of acute toxicity of chemical substances to rare crucian carp: a) Based on LC 50 >120mg / L or LC 50 > saturated solution concentration and saturated solution concentration less than 100 mg / L as the boundary, the data obtained in step 1 are divided into two groups: those with acute toxicity and those without acute toxicity, which are recorded as category 1 (with acute toxicity) and category 2 (without acute toxicity) respectively; b) search or draw the chemical structure formula of the chemical substance that needs to be predicted for acute toxicity to rare goby, recorded as A; at the same time, draw the chemical structure formula of the chemical substance with acute toxicity data for rare goby, fathead minnow and zebrafish in step 1, recorded as B; and save both A and B as mol files; c) use PaDEL-Descriptor quantum chemistry software, take all mol files as input, select 1D&2D and Fingerprints, calculate the molecular descriptor and molecular fingerprint, where A is calculated The obtained is recorded as A1, and the calculated one is recorded as B1; d) then preprocessing the obtained molecular descriptors and molecular fingerprints: i) removing variables with zero variance and near-zero variance, ii) removing variables with null values, iii) removing strongly correlated variables, wherein after preprocessing, A1 is recorded as A2, and B1 is recorded as B2; e) using machine learning methods such as support vector machine, gradient boosting method, neural network, random forest, K nearest neighbor method, etc., with the classification in step 2a as the dependent variable and B2 obtained after preprocessing in step 2d as the independent variable, a binary classification model is established; f) using this model to calculate the classification of A2, if A2 is classified as category 2, the prediction is completed, and it is concluded that the chemical substance to be predicted has no acute toxicity to rare crucian carp, if A2 is classified as category 1, then continue to step three;
[0053] 3) Step 3: Establish a regression model for predicting the acute toxicity of the chemical substance to rare goby: a) Collect the acute toxicity data of rare goby with specific values obtained from literature and experiments (i.e., remove the acute toxicity data of fathead minnow and zebrafish from the data in step 1, and remove the LC 50 > saturated solution concentration and the saturated solution concentration is less than 100 mg / L); b) Use B in step 2b to remove the acute toxicity data and LC of fathead minnow and zebrafish 50 >saturated solution concentration and the saturated solution concentration is less than 100 mg / L, recorded as C, and the molecular descriptor and molecular fingerprint are calculated according to step 2c, recorded as C1; c) using the pretreatment method of step 2d, pretreatment A1 in step 2c and C1 in step 3b, recorded as A3 and C2; d) the acute toxicity data lgLC in step 3a 50As the dependent variable, C2 in step 3c is used as the independent variable, and a regression model is established using machine learning methods such as elastic network regression, support vector machine, random forest, gradient boosting method, neural network, etc.; e) Using this model, A3 in step 3c is used as the model input data to obtain the acute toxicity LC value of the test substance to rare crucian carp. 50 .
[0054] The data processing process of step 1, step 2 and step 3 can be implemented using programming languages such as R, Julia, Python, etc.
[0055] The present embodiment also proposes a computer-readable storage medium, wherein the storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the above-mentioned prediction method of the acute toxicity of organic compounds to rare crucian carp. The storage medium can be an electronic medium, a magnetic medium, an optical medium, an electromagnetic medium, an infrared medium or a semiconductor system or a propagation medium. The storage medium can also include a semiconductor or solid-state memory, a magnetic tape, a removable computer disk, a random access memory (RAM), a read-only memory (ROM), a hard disk and an optical disk. The optical disk can include a compact disk-read-only memory (CD-ROM), a compact disk-read-write (CD-RW) and a DVD.
[0056] Application Example 1
[0057] This embodiment provides a method for predicting acute toxicity of rare crucian carp for 22 organic compounds of different properties. These 22 substances have test results, including 4-acetylaminoantipyrine, which is inaccurately calculated using the foreign ecotoxicity prediction software ECOSAR. These substances are not the training set substances used in the model construction process in step 1, and are used as test sets to verify the accuracy of model prediction. A simple schematic diagram of the process is shown in FIG. Figure 1 The specific implementation process is as shown in Figure 2 shown.
[0058] Step 1. Collect the acute toxicity data of rare goby, fathead minnow and zebrafish obtained from literature and experiments. Collect the acute toxicity data of rare goby published in Chinese and foreign languages from 2000 to 2022. After removing mixtures and inorganic compounds, there are 100 acute toxicity data, of which 89 are acutely toxic and 11 are non-acutely toxic; collect the acute toxicity data of fathead minnow and zebrafish published in Chinese and foreign languages from 1958 to 2022, select the data without acute toxicity, a total of 39, remove 3 substances that are repeated with rare goby, and 36 remain; in addition, there are 80 acute toxicity data of rare goby obtained in the experiment, of which 19 are acutely toxic and 61 are non-acutely toxic.
[0059] Step 2: Establish a binary classification model for the prediction of acute toxicity of chemical substances to rare crucian carp: a) Based on LC 50>120mg / L or LC 50 > saturated solution concentration and saturated solution concentration less than 100 mg / L as the boundary, the data obtained in step 1 are divided into two groups: acute toxicity and non-acute toxicity, respectively recorded as category 1 (acute toxicity) and category 2 (non-acute toxicity), where there are 108 data in category 1 and 108 data in category 2, and the substance name is used as the first column and the category is used as the second column to save as a csv file; b) the collected data are randomly divided into a training set and a test set according to 8:2, where the 22 organic compounds with different properties to be predicted are the substances in the test set; search the Internet to download the chemical structure mol files of these 22 chemical substances, recorded as A; search the Internet at the same time, download the chemical structure mol files of the 194 chemical substances in the training set, recorded as B; c) use PaDEL-Descriptor (v2.2 1) Quantum chemistry software, take all mol files as input, select 1D&2D and Fingerprints, calculate the molecular descriptors and molecular fingerprints, where A is recorded as A1 file and saved as csv file, the first column is the substance name, the second column is the category measured by the experiment, the subsequent columns are the calculated molecular descriptors and molecular fingerprints, the path is recorded as path1, B is the molecular descriptors and molecular fingerprints calculated, enter the corresponding rows in the csv file in step 2a, recorded as B1 file, the path is path2; d) Then preprocess the obtained B1 file: i) remove variables with zero variance and near-zero variance, ii) remove variables with null values, iii) remove strongly correlated variables (remove variables with correlation coefficient ≥ 0.9, and only retain one variable), this process is implemented by the following R program:
[0060] library(caret)
[0061] train.data<-read.csv(path2,header=TRUE)
[0062] trainy<-train.data[,2]
[0063] x.dataframe<-data.frame(train.data[,3:length(train.data[2,])])
[0064] zerovar<-nearZeroVar(x.dataframe)
[0065] x.newdata<-x.dataframe[,-zerovar]
[0066] col.NA<-unique(which(is.na(x.newdata),arr.ind=TRUE)[,2])
[0067] x.newdata1<-x.newdata[,-col.NA]
[0068] descrCorr<-cor(x.newdata1)
[0069] highCorr<-findCorrelation(descrCorr,0.9)
[0070] B2<-x.newdata1[,-highCorr]
[0071] Use the same method to preprocess the A1 file:
[0072] test.data<-read.csv(path1,header=TRUE)
[0073] testy<-test.data[,2]
[0074] x.dataframe2<-data.frame(test.data[,3:length(test.data[2,])])
[0075] x.testdata<-x.dataframe2[,-zerovar]
[0076] x.testdata2<-x.testdata[,-col.NA]
[0077] A2<-x.testdata2[,-highCorr]
[0078] After preprocessing, A1 is recorded as A2, and B1 is recorded as B2;
[0079] e) Use the support vector machine method, take the classification in step 2a as the dependent variable, and B2 obtained after preprocessing in step 2d as the independent variable to establish a binary classification model; this process is implemented by the following R language program:
[0080] library(e1071)
[0081] train_data<-data.frame(y=as.factor(trainy),B2)
[0082] svm.fit<-svm(y~.,data=train_data,kernel="linear",cost=0.005)
[0083] Check the accuracy of the model's prediction on the training set and find that the model's classification prediction accuracy for the data in the training set (i.e., the 194 substances in step 2a) is 92.4%, the sensitivity is 97.7%, the specificity is 87.2%, the F1-score is 92.8%, the Matthews correlation coefficient MCC is 85.4%, and the leave-one-out cross-validation accuracy is 75.0%;
[0084] f) Use this model to calculate the classification of A2. If A2 is classified as category 2, the prediction ends and it is concluded that the chemical substance to be predicted has no acute toxicity to rare crucian carp. If A2 is classified as category 1, continue to step 3. This process is implemented by the following R language program:
[0085] pred2<-predict(svm.fit,A2)
[0086] Checking the accuracy of the model's prediction on the validation set, it was found that the model's classification prediction accuracy for the validation set (i.e., the 22 substances in step 2b) reached 88.6%, with a sensitivity of 90.9%, a specificity of 86.4%, an F1-score of 88.9%, and a Matthews correlation coefficient (MCC) of 77.4%.
[0087] Especially for 4-acetylaminoantipyrine, the estimated value of the foreign ecotoxicity prediction software ECOSAR is 6.05 mg / L, while the model constructed here predicts its classification as category 2, that is, no acute toxicity to rare crucian carp, which is consistent with the test value (LC 50 For Category 2 substances, the prediction process ends here, i.e., there is no acute toxicity to rare goby; for Category 1 substances, proceed to Step 3.
[0088] Step 3: Establish a regression model for predicting the acute toxicity of the chemical substance rare goby: a) Collect the acute toxicity data of rare goby obtained from literature and experiments, that is, remove the acute toxicity data of fathead minnow and zebrafish from the data in step 1, and remove the LC 50 > saturated solution concentration and the saturated solution concentration is less than 100 mg / L, a total of 91 data; b) Use B in step 2b to remove the acute toxicity data and LC of fathead minnow and zebrafish 50 > saturated solution concentration and the saturated solution concentration is less than 100 mg / L, recorded as C, and the molecular descriptor and molecular fingerprint are calculated according to step 2c and stored as a csv file, with the first column being the substance name and the second column being lgLC 50, denoted as C1; c) using the pretreatment method of step 2d, pretreatment of category 1 substances (12 species) in step 2f and C1 in step 3b, denoted as A3 and C2; d) the acute toxicity data lgLC in step 3a 50 As the dependent variable, C2 in step 3c is used as the independent variable, and the regression model is established using the elastic network method; this process is implemented by the following R language program:
[0089] alp<-0.1
[0090] lamda<-0.7
[0091] glmnet.fit<-glmnet(C2,C2_lgLC50,alpha=alp,lambda=lamda)
[0092] e) Using this model, take A3 in step 3c as the model input data and obtain the acute toxicity LC value of the test substance to rare crucian carp 50 This process is implemented by the following program in R language:
[0093] pre.glmnet.fit_test<-predict(glmnet.fit,newx=A3,s=lamda)
[0094] Draw a scatter plot of the model's predicted values and measured values in the training set and test set, see Figure 3 It can be seen that the predicted value and the measured value are closely around the 45° line, and the model prediction result is good. The adjusted determination coefficient (R 2 adj ), root mean square error (RMSE), leave-one-out cross-validation coefficient (Q 2 LOO ), external validation coefficient (Q 2 EXT ) and consistency correlation coefficient (CCC), see Table 1, meet the requirements of accuracy, predictive ability, and robustness (Q 2 >0.5, R 2 >0.5, CCC>0.75, Ghosh et al., 2020).
[0095] Table 1 Elastic net regression model characterization
[0096]
[0097] f) Model application domain characterization: i) Use Euclidean distance to characterize the application domain of the binary SVM model, see Figure 4; Using the maximum distance as the threshold, all the data in the test set fall within the application domain; Using the 95th percentile of the training set as the threshold, 90.9% of the data in the test set are within the application domain, and 2 points are beyond the application domain, but the model still classifies them correctly. ii) Use Williams diagram to verify the application domain of EN regression model, see Figure 5 ; It is generally believed that the application domain range is the standardized residual |δ|≤3 and the leverage value h≤h * ;from Figure 5 It can be seen that the training sets are all within the application domain, and there is one data in the test set with a leverage value h>h * , but |δ|≤3 does not exceed the threshold, indicating that the model has good extrapolation properties for this substance. Figure 4 and Figure 5 It can be seen that this application has a wider application domain.
[0098] The above description of the embodiments is to facilitate the understanding and use of the invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.
Claims
1. A method for predicting the acute toxicity of an organic compound to rare crucian carp, characterized in that: The following steps are involved: S1: Obtain acute toxicity data for rare goby, fathead minnow, and zebrafish; S2: Construct a binary classification prediction model. First, divide the data in S1 into two groups: those with acute toxicity and those without acute toxicity. Then use PaDEL-Descriptor software to calculate the molecular descriptors and fingerprints of the chemical structure. After preprocessing, use machine learning methods to establish a binary classification prediction model. Use the binary classification prediction model to predict the acute toxicity category of chemical substances. S3: Use the model in S2 to predict and obtain information on acutely toxic chemical substances, further establish a regression model, collect data on specific LC50 values, calculate molecular descriptors, and after preprocessing, use machine learning regression methods to predict the acute toxicity LC50 values of chemical substances to rare crucian carp.
2. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 1, characterized in that: S1 specifically includes: Acute toxicity data of rare goby, fathead minnow and zebrafish were obtained. For rare goby, all available acute toxicity data were collected. For fathead minnow and zebrafish, only data of non-acute toxic substances were collected, i.e., data with half lethal concentration LC50>120mg / L or LC50>saturated solution concentration and saturated solution concentration less than 100mg / L.
3. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 1, characterized in that: S2 specifically includes: S2-1: Data classification and model preparation: The data obtained in S1 were divided into two groups according to the presence or absence of acute toxicity. Then, the chemical structure formulas of the chemicals to be predicted were drawn, and their molecular descriptors and molecular fingerprints were calculated using PaDEL-Descriptor software; S2-2: Model building and prediction: The calculated molecular descriptors and molecular fingerprints were preprocessed, and then a machine learning algorithm was selected to establish a binary classification model based on the preprocessed data. The binary classification model was used to predict the acute toxicity category of chemical substances to rare crucian carp.
4. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 3, characterized in that: S2-1 specifically includes: The collected acute toxicity data are divided into two categories according to whether they have acute toxicity or not, namely the first category with acute toxicity and the second category without acute toxicity, providing classification standards for subsequent model training; For the chemical substance to be predicted, its chemical structure formula is drawn, and the PaDEL-Descriptor quantum chemistry software is used to calculate the 1D&2D descriptors and fingerprints of the molecule to generate numerical data describing the molecular characteristics, providing the necessary input features for establishing the prediction model.
5. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 3, characterized in that: S2-2 specifically includes: Clean and filter the molecular descriptors and molecular fingerprint data calculated by PaDEL-Descriptor software to achieve preprocessing; A machine learning algorithm was selected, and the preprocessed molecular descriptors and molecular fingerprint datasets were used as input features. The presence or absence of acute toxicity was used as the output result. A binary classification prediction model was trained and established. Subsequently, the binary classification prediction model was used to predict the acute toxicity category of new chemicals to determine whether they were acutely toxic to the rare minnow.
6. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 5, characterized in that: In S2-2, the cleaning and screening include: removing zero-variance and near-zero-variance variables to eliminate uninformative features, deleting variables containing null values to ensure data integrity, and removing strongly correlated variables to reduce model complexity and avoid multicollinearity problems; The machine learning algorithm is selected from one of a support vector machine, a gradient boosting method, and a neural network.
7. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 1, characterized in that: S3 specifically includes: S3-1: Data collection and preprocessing: Obtain information on chemical substances with acute toxicity predicted by the model in S2, and obtain acute toxicity data of rare minnow with specific LC50 values, while removing data of fathead minnow and zebrafish and LC50 data that do not meet the conditions. Then, for the chemical substances to be predicted, calculate molecular descriptors and molecular fingerprints, and preprocess the data; S3-2: Regression model establishment and application: The specific LC50 value was used as the dependent variable and the preprocessed molecular descriptor as the independent variable. The machine learning regression algorithm was used. The molecular descriptor data of the chemical substance to be predicted was input into the regression model to predict its acute toxicity LC50 value to the rare crucian carp.
8. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 7, characterized in that: S3-1 specifically includes: Obtain information on chemical substances with acute toxicity predicted by the model in S2, and obtain the acute toxicity data of rare goby from literature data and experimental information to ensure that the data has a specific LC50 value. At the same time, exclude the data of fathead minnow and zebrafish, as well as the data with LC50 greater than the saturated solution concentration but less than 100 mg / L; For the chemical substances to be predicted, the chemical structure formula is drawn, and the molecular descriptors and molecular fingerprints are calculated using PaDEL-Descriptor software. Then, the calculated data is preprocessed, including removing zero variance and near-zero variance variables, null value variables, strongly correlated variables, and performing necessary standardization to prepare for the training of the regression model.
9. The method for predicting the acute toxicity of an organic compound to rare crucian carp according to claim 7, characterized in that: S3-2 specifically includes: Select a machine learning regression algorithm, including elastic network regression or support vector machine regression, with the collected specific LC50 values as the dependent variable and the preprocessed molecular descriptor data as the independent variable, to train the regression model and adjust and optimize the regression model parameters; The trained regression model is applied to the molecular descriptor data of the chemical substance to be predicted, and these data are input into the regression model. The model is used to calculate the predicted LC50 value of the acute toxicity of the chemical substance to rare crucian carp, thereby providing quantitative results for the toxicity assessment of the chemical substance.
10. A storage medium containing computer executable instructions, characterized in that: When the storage medium of the computer executable instructions is executed by a computer processor, it is used to execute the method for predicting the acute toxicity of an organic compound to rare minnows according to any one of claims 1 to 9.
Citation Information
Patent Citations
Method for predicting acute toxicity of organophosphorous flame retardant on zebra fish based on QSAR / QEcoSAR method combination
CN108733970A
Rare gobiocypris rarus embryotoxicity determination method and water body monitoring and early warning method
CN111896727A
Aquatic organism acute toxicity multi-classification prediction method based on machine learning and integration method
CN116403659A
