CN116030905A - Integrated learning method for predicting short-term exposure lethal effect of neurotoxic - Google Patents
Integrated learning method for predicting short-term exposure lethal effect of neurotoxic Download PDFInfo
- Publication number
- CN116030905A CN116030905A CN202310122078.XA CN202310122078A CN116030905A CN 116030905 A CN116030905 A CN 116030905A CN 202310122078 A CN202310122078 A CN 202310122078A CN 116030905 A CN116030905 A CN 116030905A
- Authority
- CN
- China
- Prior art keywords
- model
- exposure
- neurotoxic
- features
- short
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an integrated learning method for predicting a short-term exposure lethal effect of a nerve poison. The present invention establishes a heterogeneous data set that encompasses different test animals and multiple exposure pathways, involving multiple toxicity mechanisms. The invention breaks through the traditional machine learning modeling by considering only molecular structural features, ignores the research paradigm of biological exposure features, comprehensively considers different test animals and various exposure ways, performs independent thermal coding on the test animals and various exposure ways, couples the molecular structural features, and develops an integrated learning prediction model based on hard voting combination. The nerve poison short-term exposure lethal effect prediction model established by the invention has higher internal robustness and external prediction capability, has definite application domain, is simple and convenient to operate, and can save the time, cost and animal number of experimental tests; can be used as an advantageous tool for efficiently predicting the short-term exposure lethal effect of the neurotoxic and provides basic toxicity data for evaluating and managing the human health risk of chemicals.
Description
Technical Field
The invention belongs to the technical field of computational toxicology oriented to chemical health risk evaluation and management, and relates to a method for predicting a short-term exposure lethal effect of a nerve poison based on quantitative structure-activity relationship (QSAR) and integrated learning.
Background
A neurotoxic is any chemical, biological or physical substance that may cause neurotoxicity. Hundreds of experimental animal poisoning events caused by more than one neuropoison have been reported worldwide. The toxic mechanisms of these neurotoxics involve inhibition of acetylcholinesterase and blockade of adrenergic receptors, etc., leading to serious complications associated with the central and peripheral nervous systems of the test animals, such as anorexia, tics, skeletal paralysis, ataxia, tremors, disappearance of eversion and death. Short-term exposure mortality is one of the core matters for chemical health risk assessment and management, and can be achieved by testing animals for half-lethal dose (LD 50 ) An evaluation is performed. The number of neurotoxics is great, the lethal effect of short-term exposure on them remains largely unknown, and the experimental test results are irregular. Given the detrimental effects of neurotoxic on health, it is necessary to predict short-term exposure lethal effects under the same approach, before it enters the environment.
The relevant guidelines issued by the economic Cooperation and development Organization (OECD) are rodent neurotoxicity research test guidelines (OECD guideline 424) and can be used for acute neurotoxicity research. The guidelines require assessment of a range of behaviors of animals that may be affected by neurotoxic during each experimental observation, death being one of the key observation indicators. However, this experimental test method is expensive and time-consuming, and has ethical implications for offending animals, and it is difficult to evaluate the short-term exposure lethal effects of numerous neurotoxins one by one, and it is necessary to develop high-throughput quantitative prediction techniques. The correlation between the molecular structural characteristics of the nerve poison and the short-term exposure lethal effect induced by the molecular structural characteristics of the nerve poison is established based on QSAR, so that the efficient prediction of the short-term exposure lethal effect of the nerve poison can be realized, the time is saved, the cost is reduced, and the number of animals required by experimental tests is reduced.
Traditional linear QSARs have difficulty meeting predictive requirements due to the inability to deal with the nonlinear relationship between molecular structural features and short-term exposure lethal effects. In recent years, rapidly developed machine learning algorithms, due to their data-adaptive nature, can provide superior performance for analyzing high-order, high-dimensional and nonlinear relationship data, and are applied to mining the inherent correlation between molecular structural features and short-term exposure lethal effects. The integrated learning can obtain the generalization performance remarkably superior to that of a single machine learning algorithm by combining a plurality of machine learning algorithms, is hopeful to play a positive role in quantitative prediction of the short-term exposure lethal effect of the nerve poison, and realizes efficient filling of the data vacancy of the short-term exposure lethal effect of the nerve poison.
At present, some studies have constructed short-term exposure lethal effects of neurotoxic (end point value of toxicity is LD) 50 ) QSAR predictive model of (c). Document 1"Chemical Research in Toxicology,2006,19 (2): 209-216" considers the lethal effects of oral short-term exposure of male rats of 38 organophosphorus compounds, the toxic mechanisms of which mainly include the inhibition of acetylcholinesterase. Document 1 constructs a QSAR model for predicting short-term exposure lethal effects of an organophosphorus compound based on descriptors of absorption, distribution, metabolism, and excretion processes, and the like, with a leave-one-out cross validation coefficient of 0.82. Reference 2"Toxicology Research,2020,9 (3): 164-172" based on 422 organic compound autonomic nervous system toxicity data collected in ChemIDplus database, an extreme tree regression model of short term exposure lethal effect of mice by intraperitoneal injection was constructed using the pybio med descriptor, with an external validation coefficient of 0.784. Furthermore, document 3"Environmental Science&Technology,2022,56 (1): 335-348 "data were collected from the pesticide ecotoxicity database of the pesticide program of the U.S. environmental protection agency and the toxicity of 128 compounds in the literature on various birds, the mechanism of toxicity of these compounds being related to the inhibition of acetylcholinesterase. Document 3 uses 11 two-dimensional descriptors to construct a QSAR model for predicting acute oral toxicity of pesticides to birds, wherein the toxicity prediction model for quails in america has an external validation coefficient of 0.648. In summary, the above models are predictive models of short term exposure lethal effects of neurotoxic but all employ conventional systemsModeling by a learning method or a single machine learning algorithm, wherein the robustness and the prediction capability of the model are to be improved; and the biological exposure path is single, which leads to limitation of the application range of the model to toxicity prediction under specific exposure conditions; furthermore, the data sets of these models contain less structural diversity of neurotoxics, resulting in a small model application domain. Specifically, the modeling data set of document 1 is small, the included compounds are not diversified, only the lethal effect of 38 organophosphorus compounds male rats is involved in oral short-term exposure, the application range of the prediction model is narrow, and no clear application domain characterization is performed. Document 2, although having a large data set, only covers the lethal effect of intraperitoneal injection exposure of mice; although document 3 considers experimental data of a plurality of test animals, it only relates to oral exposure toxicity of birds. Neither document 2 nor 3 can be used for the prediction of neurotoxic effects of various biological and exposure pathways. Therefore, it is necessary to break through the research paradigm of single algorithm modeling and develop integrated learning models of short term exposure lethal effects of neurotoxic based on heterogeneous data sets of different test animals and multiple exposure pathways.
For the reasons described above, 574 pieces of high-dimensional, diverse short-term exposure lethal effect data of neurotoxic that cover different test animals and multiple exposure pathways were collected and organized from the PubChem database. The experimental information of different test animals and various exposure ways is subjected to independent thermal coding, chemical information of molecular structural features obtained through coupling calculation is taken as modeling features, three different machine learning algorithms are taken as base regressors, a hard voting strategy is adopted, an integrated learning model for predicting the short-term exposure lethal effect of the nerve poison is built, and the application domain of the model is characterized.
Disclosure of Invention
The invention constructs an integrated learning method for predicting the short-term exposure lethal effect of the nerve poison with high efficiency and low cost, and the method can quantitatively predict the short-term exposure lethal effect directly according to the molecular characteristics calculated by the molecular structure of the nerve poison and the designated biological exposure path. The method is based on various nerve poisons, breaks through the fact that only molecular structural features are considered in the traditional machine learning modeling, ignores the research paradigm of biological exposure features, comprehensively considers different test animals and various exposure ways, performs independent thermal coding on the test animals and various exposure ways, couples the molecular structural features, and develops an integrated learning prediction model based on a hard voting strategy. The model built by the invention has higher internal robustness and external prediction capability, is simple and convenient to operate, and can save the time, cost and animal number of experimental tests; can be used as an advantageous tool for efficiently predicting the short-term exposure lethal effect of the neurotoxic and provides basic toxicity data for evaluating and managing the human health risk of chemicals.
The technical scheme of the invention is as follows:
an integrated learning method for predicting short-term exposure lethal effect of neurotoxic comprises the following steps:
(1) Construction of a data set for short-term exposure lethal effects of neurotoxic
Short-term exposure lethal effect data of neurotoxic was collected and collated from the PubChem database, which covers both rats and mice, and four exposure routes of intraperitoneal injection, intravenous injection, oral injection and subcutaneous injection, totaling 574 toxicity test data; short-term exposure lethal effect endpoint value (LD 50 Mg/kg) to molar concentration (mol/kg) and then to negative logarithm (pLD) 50 ) The method comprises the steps of carrying out a first treatment on the surface of the The toxicity mechanism involves alpha adrenergic receptor blocking, beta adrenergic receptor blocking, ganglion blocking, acetylcholinesterase inhibition, etc.; the collected neurotoxic belongs to pesticides, antipsychotic drugs, dyes, rubber auxiliary intermediates and the like, and comprises organic acids, ethers, esters, ketones, alcohols, amides, anilines, polycyclic aromatic hydrocarbons and substitutes thereof, halogenated alkanes, halogenated olefins, heterocyclic compounds and derivatives thereof and the like, and does not comprise inorganic compounds, organic metal compounds and mixtures (mainly macromolecular salt compounds);
(2) Representation of molecular structural features and biological exposure features of neurotoxic compounds
PubCHem CIDs of 574 nerve poisons and corresponding 2D structures are obtained in batches based on PubCHem Py, and are stored in SDF format files; inputting the SDF file into PaDEL-Descriptor version 2.21 software, and calculating 1D and 2D molecular structure characteristics; single heat encoding was performed on two test animals and four exposure pathways; firstly removing the features with variance of 0 and then removing the features with the pearson correlation coefficient larger than 0.9, and adopting a recursive feature elimination method to select the features;
(3) Model training
Coupling the 1D and 2D molecular structural features of the neurotoxic with experimental features of different test animals and multiple exposure paths, and taking the coupled features as the feature input of a model, pLD 50 The value is used as a prediction end point of the model, and a machine learning regression model is constructed; randomly splitting the data set into a training set and a verification set according to the ratio of 4:1, and performing internal verification by adopting ten-fold cross verification to reduce random errors, wherein the verification set is used for external verification of the model; four machine learning algorithms, namely K nearest neighbor (K-nearest neighbors, KNN), support vector machine (support vector machine, SVM), random Forest (RF) and gradient lifting decision tree (gradient boosting decision tree, GBDT) are adopted to respectively construct models, three algorithms with optimal model performance are used as base regressors, a hard voting strategy is adopted to construct an integrated learning model, and the weights of the base regressors in the hard voting strategy are the same; determining the optimal super-parameters of a machine learning algorithm through grid search, wherein the optimal super-parameters of the integrated learning model are derived from the optimal super-parameters of each base regressor;
the following are model-optimized hyper-parameters, the best hyper-parameters of KNN: the number of neighboring points (n_neighbors) is 4, and the weight function (weights) is distance; optimal super parameters of SVM: the radial basis is used as a kernel function, the regularization parameter (C) is 200, the kernel coefficient (gamma) is 0.005, and epsilon is 0.1; optimal super parameters for RF: the number of decision trees (n_identifiers) is 290, the maximum feature number (max_features) of each decision tree is 11, and the random seed (random_state) is 86; optimal superparameter for GBDT: n_evastiators is 200, max_features is 20, and random_state is set to 0; the rest parameters of the algorithm are all default values;
(4) Model evaluation and verification
Using the square (R 2 adj ) And Root Mean Square Error (RMSE) evaluation model, training set ten-fold cross validation coefficient (Q 2 CV ) Evaluation of modelRobustness, validation set external validation coefficient (Q 2 ext ) Evaluating the predictive ability of the model;
the predicted performance of the model is:
model 1 performance constructed by KNN algorithm: training set R 2 adj =0.772,RMSE CV =0.394,Q 2 CV =0.783; verification set RMSE ext =0.414,Q 2 ext =0.799;
The 3 algorithms with the optimal model effect are SVM, RF, GBDT, which are used as a base regressor, and the performance of the integrated learning model 5 constructed by adopting a strategy of hard voting is: training set R 2 adj =0.863,RMSE CV =0.307,Q 2 CV =0.870; verification set RMSE ext =0.344,Q 2 ext =0.861; the integrated learning model has optimal performance, and is used as a final model for predicting the short-term exposure lethal effect of the nerve poison;
(5) Application domain characterization
Based on the normalized residual and the leverage distance (h i ) Drawing a Williams diagram, and representing the application domain of a regression model 5, wherein the training set features comprise biological exposure features besides molecular structure features; h is a i The formula of (2) is as follows:
H=X(X T X) -1 X T (1)
h i =[H] ii (2)
wherein X is a matrix of a plurality of nerve poisons and a plurality of features in the training set, H is a hat matrix, H i Lever distance for ith nerve poison, [ H ]] ii The matrix H diagonal value is H, the guard value of the defined lever distance is H, p is the number of training set features, and n is the number of training set nerve poisons;
if the nerve is poisoned by h i Greater than h, the neurotoxic is considered to be outside the application domain. Thus, the ensemble learning model 5 is applied to the pair h i Less than 0.137 of the neurotoxic pLD 50 And (5) predicting a value.
The beneficial effects of the invention are as follows:
(1) The built ensemble learning model 5 combines the advantages of three different machine learning algorithms (SVM, RF, GBDT), has superior robustness and predictive ability and has well-characterized application domains compared to the models of documents 1, 2, 3 and models 1, 2, 3, 4 built by a single machine learning algorithm;
(2) The modeling data set covers toxicity data of two test animals of rats and mice and four exposure approaches of abdominal cavity, vein, oral injection and subcutaneous injection, breaks through the research paradigm that only molecular structural features are considered in the traditional machine learning modeling, ignores biological exposure features, and has wider application field;
(3) The modeling has the advantages that the quantity of the nerve poison used by the modeling is large, the structure is various, the nerve poison comprises organic acid, ether, ester, ketone, alcohol, amide, aniline, polycyclic aromatic hydrocarbon and substitutes thereof, halogenated alkane, halogenated olefin, heterocyclic compound and derivatives thereof, and the like, and the modeling has good generalization capability;
(4) The method is simple, convenient and efficient, can save the time, cost and animal number of experimental tests, is expected to play an important role in quantitative prediction of the short-term exposure lethal effect of the nerve poison, fills up the data gap of the short-term exposure lethal effect of the nerve poison efficiently, provides a basic tool for evaluating and managing the health risks of chemicals, and serves the national important requirements of chemical risk management and new pollutant treatment.
Drawings
FIG. 1 shows training set pLD 50 Fitting the experimental and predicted values, the training set was 459 pieces of toxicity data.
FIG. 2 shows a verification set pLD 50 Fitting of the experimental values to the predicted values, the validation set was 115 pieces of toxicity data.
Fig. 3 is a Williams diagram of the model, the circles represent training set neurotoxicity, the triangles represent validation set neurotoxicity, and the alert value h is 0.137.
Detailed Description
The following describes the embodiments of the present invention further with reference to the drawings and technical schemes.
Example 1
Given a neuropoison diisopropanolamine (CAS: 110-97-4/PubCHem CID: 8086), its short-term exposure lethal effect (endpoint value pLD) 50 ). Firstly, according to PubCHem CID of diisopropanolamine, SDF structure files are obtained, and PaDEL-Descriptor software is utilized to calculate 1D and 2D molecular structures. The h value calculated according to formulas (1) and (2) is 0.032, which is less than the lever guard value h of the neurotoxic in fig. 3 (0.137), so diisopropanolamine is in the model application domain. The test animal is designed to be a mouse, the exposure route is intraperitoneal injection, and the calculated values of the molecular structural characteristics are input into a model to obtain pLD 50 Is 3.003, and pLD determined by experimental tests 50 The value was 3.142 and the predicted and experimental data were very consistent.
Example 2
Given a neurotoxic guaifenesin (CAS: 93-14-1/PubCHem CID: 3516), the short-term exposure lethal effect (endpoint value pLD) 50 ). Firstly, according to the PubCHem CID of guaifenesin, an SDF structure file is obtained, and the 1D and 2D molecular structures of the SDF structure file are calculated by using PaDEL-Descriptor software. H calculated according to formulas (1) and (2) is 0.042, which is less than the lever guard value h of the neurotoxic in figure 3 (0.137), so guaifenesin is in the modelType application domain. Assigning the test animal to be a mouse, injecting the exposure route to be subcutaneous injection, and inputting the calculated values of the molecular structural characteristics into a model to obtain the pLD 50 Is 2.545, and pLD determined by experimental tests 50 The value was 2.394 and the predicted and experimental data were very consistent.
Example 3
Given a neurotoxic 3- (m-tolyl) octanone (CAS: 3483-17-8/PubCHem CID: 198874), its short-term exposure lethal effect was predicted (endpoint value pLD) 50 ). Firstly, according to the PubCHem CID of 3- (m-tolyl) octanone, an SDF structure file is obtained, and the 1D and 2D molecular structures of the SDF structure file are calculated by using PaDEL-Descriptor software. The h value calculated according to formulas (1) and (2) is 0.033, which is less than the lever guard value h of the neurotoxic in fig. 3 (0.137), so 3- (m-tolyl) octanone is in the model application domain. The test animal is designed to be a mouse, the exposure route is oral injection, and the calculated values of the molecular structural characteristics are input into a model to obtain pLD 50 Is 2.551, and pLD determined by experimental tests 50 The values were 2.550, and the predicted and experimental values were very consistent.
Claims (5)
1. An integrated learning method for predicting short-term exposure lethal effect of neurotoxic is characterized by comprising the following steps:
(1) Construction of a data set for short-term exposure lethal effects of neurotoxic
Short-term exposure lethal effect data of neurotoxicity is collected and organized from PubCHem database, and toxicity endpoint value is half lethal dose LD 50 It is converted to molar concentration and then to negative log pLD 50 The method comprises the steps of carrying out a first treatment on the surface of the The data set covers two test animals, namely rats and mice, and four exposure routes of intraperitoneal injection, intravenous injection, oral injection and subcutaneous injection, and contains 574 toxicity test data in total; the data set includes a variety of toxic mechanisms of action, specifically alpha adrenergic receptor blockade, beta adrenergic receptor blockade, ganglionic block, and acetylcholinesterase inhibition; the neurotoxic agent comprises organic acid, ether, ester, ketone, alcohol, amide, aniline, polycyclic aromatic hydrocarbon and its substitute, halogenated alkane, and halogenSubstituted olefins, heterocyclic compounds and derivatives thereof;
(2) Representation of molecular structural features and biological exposure features of neurotoxic compounds
Obtaining 574 PubCHem CIDs of nerve toxicants in batches based on a PubCHem database and a corresponding 2D structure thereof; calculating the 1D and 2D structural characteristics of the molecules by using PaDEL-Descriptor software; single heat encoding was performed on two test animals and four exposure pathways; the preprocessing process comprises the steps of firstly removing the features with variance of 0 and then removing features with the pearson correlation coefficient larger than 0.9, and adopting a recursive feature elimination method to perform feature screening;
(3) Model training
Coupling the 1D and 2D molecular structural features of the neurotoxic with experimental features of different test animals and multiple exposure paths, and taking the coupled features as the feature input of a model, pLD 50 The value is used as a prediction end point of the model, and a machine learning regression model is constructed; randomly splitting the data set into a training set and a verification set according to the ratio of 4:1, and performing internal verification by adopting ten-fold cross verification, wherein the verification set is used for external verification of the model; four machine learning algorithms, namely K nearest neighbor, a support vector machine, a random forest and a gradient lifting decision tree are adopted to respectively construct a model, three algorithms with optimal model performance are used as base regressors, a hard voting strategy is adopted to construct an integrated learning model, and the weights of the base regressors in the hard voting strategy are the same; determining the optimal super-parameters of a machine learning algorithm through grid search, and constructing an integrated learning model based on the optimal super-parameters;
the following is the model-optimized hyper-parameters, the best hyper-parameters for K-nearest neighbors: the number of adjacent points is 4, and the weight function is distance; optimal super parameters of support vector machine: the radial basis is taken as a kernel function, the regularization parameter is 200, the kernel coefficient is 0.005, and epsilon is 0.1; optimal superparameter for random forest: the number of decision trees is 290, the maximum feature number of each decision tree is 11, and the random seed is 86; optimal superparameter of gradient boosting decision tree: the number of decision trees is 200, the maximum feature number of each decision tree is 20, and the random seed is set to 0;
3 algorithms with optimal model performance, namely a support vector machine, a random forest and a gradient lifting decision tree are adopted as a base regressor, and a strategy of hard voting is adopted to construct an integrated learning model;
(4) Model evaluation, verification and application domain characterization
Using square R of the degree-of-freedom adjusted correlation coefficient 2 adj Fitting ability of RMSE evaluation model and internal ten-fold cross validation coefficient Q 2 CV Evaluating robustness of model, external verification coefficient Q 2 ext Evaluating the predictive ability of the model; an application domain using a Williams characterization model, wherein the training set features include biological exposure features in addition to molecular structural features;
integrated learning model training set R 2 adj =0.863,RMSE CV =0.307,Q 2 CV =0.870; verification set RMSE ext =0.344,Q 2 ext =0.861; this model was used as the final model for predicting the short-term exposure lethal effects of neurotoxic.
2. The method of claim 1, wherein the training set features in the model application domain representation include biological exposure features in addition to molecular structural features, the defined application domain threshold being: lever guard value (h) =0.137.
3. The method of claim 1, wherein the short term exposure lethal effect data of the neurotoxic agent covers both rats and mice tested animals, and four exposure routes of intraperitoneal injection, intravenous injection, oral injection, and subcutaneous injection.
4. The method of claim 1, wherein the neurotoxic agent comprises a pesticide, an antipsychotic agent, a dye, and a rubber additive intermediate.
5. The method of claim 1, wherein the neurotoxic agent comprises an organic acid, ether, ester, ketone, alcohol, amide, aniline, polycyclic aromatic hydrocarbon and substituents thereof, halogenated alkane, halogenated alkene, heterocyclic compound and derivatives thereof.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310122078.XA CN116030905A (en) | 2023-02-16 | 2023-02-16 | Integrated learning method for predicting short-term exposure lethal effect of neurotoxic |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310122078.XA CN116030905A (en) | 2023-02-16 | 2023-02-16 | Integrated learning method for predicting short-term exposure lethal effect of neurotoxic |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116030905A true CN116030905A (en) | 2023-04-28 |
Family
ID=86073943
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310122078.XA Pending CN116030905A (en) | 2023-02-16 | 2023-02-16 | Integrated learning method for predicting short-term exposure lethal effect of neurotoxic |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116030905A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116541785A (en) * | 2023-07-05 | 2023-08-04 | 北京建工环境修复股份有限公司 | Toxicity prediction method and system based on deep integration machine learning model |
-
2023
- 2023-02-16 CN CN202310122078.XA patent/CN116030905A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116541785A (en) * | 2023-07-05 | 2023-08-04 | 北京建工环境修复股份有限公司 | Toxicity prediction method and system based on deep integration machine learning model |
CN116541785B (en) * | 2023-07-05 | 2023-09-12 | 北京建工环境修复股份有限公司 | Toxicity prediction method and system based on deep integration machine learning model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Pillar et al. | A framework for metacommunity analysis of phylogenetic structure | |
Bastille‐Rousseau et al. | Leveraging multidimensional heterogeneity in resource selection to define movement tactics of animals | |
Morris | Ecological scale and habitat use | |
Raymond et al. | A review of structure-based biodegradation estimation methods | |
Weiblen et al. | Phylogenetic dispersion of host use in a tropical insect herbivore community | |
Cunningham et al. | Modeling count data of rare species: some statistical issues | |
Dray et al. | Testing the species traits–environment relationships: the fourth‐corner problem revisited | |
Ai et al. | QSAR modelling study of the bioconcentration factor and toxicity of organic compounds to aquatic organisms using machine learning and ensemble methods | |
Daskin et al. | Does primary productivity modulate the indirect effects of large herbivores? A global meta‐analysis | |
Horn et al. | Landscape composition, patch size, and distance to edges: interactions affecting duck reproductive success | |
Smith et al. | Are sage‐grouse fine‐scale specialists or shrub‐steppe generalists? | |
CN116030905A (en) | Integrated learning method for predicting short-term exposure lethal effect of neurotoxic | |
Chen et al. | Ecotoxicological QSAR study of fused/non-fused polycyclic aromatic hydrocarbons (FNFPAHs): Assessment and priority ranking of the acute toxicity to Pimephales promelas by QSAR and consensus modeling methods | |
Zuckerberg et al. | A review of overlapping landscapes: pseudoreplication or a red herring in landscape ecology? | |
Cable et al. | Habitat suitability and connectivity modeling reveal priority areas for Indiana bat (Myotis sodalis) conservation in a complex habitat mosaic | |
Yang et al. | Development of QSAR models for evaluating pesticide toxicity against Skeletonema costatum | |
Shanahan et al. | Models based on individual level movement predict spatial patterns of genetic relatedness for two Australian forest birds | |
Dyson et al. | Microhabitat nest‐site selection by ducks in the boreal forest | |
Yan et al. | Species co‐occurrence and phylogenetic structure of terrestrial vertebrates at regional scales | |
Cao et al. | In silico estimation of chemical aquatic toxicity on crustaceans using chemical category methods | |
Jasper et al. | Estimating dispersal using close kin dyads: The kindisperse R package | |
Silva et al. | Closely related species show species-specific environmental responses and different spatial conservation needs: Prionailurus cats in the Indian subcontinent | |
Connors et al. | Conservation risk and uncertainty in recovery prospects for a collapsed and culturally important salmon population in a mixed‐stock fishery | |
Woods et al. | Testing the diversity–biomass relationship in riverine fish communities | |
Liu et al. | Prediction of Nucleophilicity and Electrophilicity Based on a Machine‐Learning Approach |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |