The importance of outlier detection and training set selection for reliable environmental QSAR predictions

Empirical QSAR models are only valid in the domain they were trained and validated. Application of the model to substances outside the domain of the model can lead to grossly erroneous predictions. Partial least squares (PLS) regression provides tools for prediction diagnostics that can be used to decide whether or not a substance is within the model domain, i.e. if the model prediction can be trusted. QSAR models for four different environmental end-points are used to demonstrate the importance of appropriate training set selection and how the reliability of QSAR predictions can be increased by outlier diagnostics. All models showed consistent results; test set prediction errors were very similar in magnitude to training set estimation errors when prediction outlier diagnostics were used to detect and remove outliers in the prediction data. Test set prediction errors for substances classified as outliers were much larger. The difference in the number of outliers between models with a randomly and systematically selected training illustrates well the need of representative training data.     

A protocol to select high quality datasets of ecotoxicity values for pesticides

The key to any QSAR model is the underlying dataset. In order to construct a reliable dataset to develop a QSAR model for pesticide toxicity, we have derived a protocol to critically evaluate the quality of the underlying data. In developing an appropriate protocol that would enable data to be selected in constructing a QSAR, we concentrated on one toxicity end point, the 96 h LC50 from the acute rainbow trout study. This end point is key in pesticide regulation carried out under 91/414/EEC. The dataset used for this exercise was from the US EPA-OPP database.     

A Protocol to Select High Quality Datasets of Ecotoxicity Values for Pesticides

Abstract

The key to any QSAR model is the underlying dataset. In order to construct a reliable dataset to develop a QSAR model for pesticide toxicity, we have derived a protocol to critically evaluate the quality of the underlying data. In developing an appropriate protocol that would enable data to be selected in constructing a QSAR, we concentrated on one toxicity end point, the 96 h LC50 from the acute rainbow trout study. This end point is key in pesticide regulation carried out under 91/414/EEC. The dataset used for this exercise was from the US EPA-OPP database.     

The effect of nitroaromatics' composition on their toxicity in vivo: novel, efficient non-additive 1D QSAR analysis

Novel 1D QSAR approach that allows analysis of non-additive effects of molecular fragments on toxicity has been proposed. Twenty-eight nitroaromatic compounds including some well-known explosives have been chosen for this study. The 50% lethal dose concentration for rats (LD50) was used as the estimation of toxicity in vivo to develop 1D QSAR models on the framework of Simplex representation of molecular structure. The results of 1D QSAR analysis show that even the information about the composition of molecules provides the main trends of toxicity changes. The necessity of consideration of substituents' mutual impacts for the development of adequate QSAR models of nitroaromatics' toxicity was demonstrated. Statistic characteristics for all the developed partial least squares QSAR models, except the additive ones are quite satisfactory (R2=0.81-0.92; Q2=0.64-0.83; R2 test=0.84-0.87). A successful performance of such models is due to their non-additivity i.e. possibility of taking into account the mutual influence of substituents in benzene ring which plays the governing role for toxicity change and could be mediated through the different C-H fragments of the ring. The correspondence between observed and predicted by these models toxicity values is good. This allowing combine advantages of such approaches and develop adequate consensus model that can be used as a toxicity virtual screening tool.     

Acute toxicity of benzene derivatives to the tadpoles (Rana japonica) and QSAR analyses

Acute lethal toxicity (the negative logarithm of molar concentrations of 12 h acute median lethal, expressed as 12 h-log1/LC50) of 46 benzene derivatives to Rana japonica tadpoles was determined. 1-octanol/water partition coefficient (logKow)-dependent models were developed to study the toxicity of different categories chemicals. In an effort to model all chemicals, response surface analyses and stepwise multiple regression analyses were performed and successful models were obtained. A general and robust QSAR model was achieved with the combined application of variables reflecting hydrophobicity, electric property, and molecular size respectively (12h-log1/LC50 = 0.393logKow - 0.428Elumo + 0.0110Vol. + 1.362 n = 51, r2 = 0.834) using stepwise multiple regression analyses. Because of strong dissociation of carboxyl group greatly decreasing their observed toxicity, using logDow in instead of logKow the quality of the models is greatly improved. The conventional r2 and cross-validation r2(CV) were 0.914 and 0.785, respectively, indicating that QSAR was both internally consistent and highly predictive.     

Macroinvertebrate response to acid mine drainage: community metrics and on-line behavioural toxicity bioassay

The hypothesis is tested that toxicity of acid mine drainage can be detected by a selection of existing macroinvertebrate community and bioindicator metrices supplemented by toxicity tests with the local mosquitofish Gambusia holbrooki Girard and the shrimp Atyaephyra desmaresti Millet. The behavioural responses of A. desmaresti to acid mine drainage were recorded in the Multispecies Freshwater Biomonitor, based on behaviour and survival as parameters. Bioassessment methods were based on community diversity, structure, function, and bioindicators and supplemented by chemical analysis (temperature, pH, metals). The Biological Monitoring Working Party adapted for the Iberian Peninsula, the number of predators (Coleoptera, Hemiptera) and the number of Ephemeroptera and Trichoptera taxa differentiated the sites well. The on-line toxicity test revealed pH-dependent acute toxicity of the acid mine drainage for the shrimp (LC(50)-48 h: pH-AMD=5.8) and a pH- dependent decrease in locomotory activity with the lowest-observed-response-times (LORTs) within 5 h of exposure. Shrimp were more sensitive to acid mine drainage than fish (LC(50)-48 h: pH-AMD=4.9). A new multimetric index combining toxicity testing and bioassessment methods is proposed.     

A new hybrid system of QSAR models for predicting bioconcentration factors (BCF)

The aim was to develop a reliable and practical quantitative structure–activity relationship (QSAR) model validated by strict conditions for predicting bioconcentration factors (BCF). We built up several QSAR models starting from a large data set of 473 heterogeneous chemicals, based on multiple linear regression (MLR), radial basis function neural network (RBFNN) and support vector machine (SVM) methods. To improve the results, we also applied a hybrid model, which gave better prediction than single models. All models were statistically analysed using strict criteria, including an external test set. The outliers were also examined to understand better in which cases large errors were to be expected and to improve the predictive models. The models offer more robust tools for regulatory purposes, on the basis of the statistical results and the quality check on the input data.     

A new hybrid system of QSAR models for predicting bioconcentration factors (BCF)

The aim was to develop a reliable and practical quantitative structure-activity relationship (QSAR) model validated by strict conditions for predicting bioconcentration factors (BCF). We built up several QSAR models starting from a large data set of 473 heterogeneous chemicals, based on multiple linear regression (MLR), radial basis function neural network (RBFNN) and support vector machine (SVM) methods. To improve the results, we also applied a hybrid model, which gave better prediction than single models. All models were statistically analysed using strict criteria, including an external test set. The outliers were also examined to understand better in which cases large errors were to be expected and to improve the predictive models. The models offer more robust tools for regulatory purposes, on the basis of the statistical results and the quality check on the input data.     

Progressive study and robustness test of QSAR model based on quantum chemical parameters for predicting BCF of selected polychlorinated organic compounds (PCOCs)

Systematic analyses on the effects of chemical structures of 31 polychlorinated organic compounds (PCOCs) on their bioconcentration behavior in rainbow trout (Oncorhynchus mykiss) were conducted using quantitative structure-activity relationship (QSAR) techniques. The cluster analyses of individual variables as well as the quality control chart of QSAR model implies the existence of outliers, while the simulation model excluding such samples showed an extreme robustness even if it was tested with different methods. Furthermore, the quantum chemical parameters entering into QSAR model were used to describe the bioconcentration pathways, and the results indicated that bioconcentration behaviors of selected compounds were complicated processes involving permeation stages as well as bio-chemical reaction stages.     

Acute toxicity of substituted phenols to Rana japonica tadpoles and mechanism-based quantitative structure-activity relationship (QSAR) study.

Acute 12 h and 24 h lethal toxicity (12 h-LC50 and 24 h-LC50) of 31 substituted phenols to Rana japonica tadpoles was determined. Results indicate that toxicity of phenols to tadpoles varied only slightly with length of exposure and the 12-h test could serve as surrogate of the 24-h test. A mechanism-based quantitative structure-activity relationship (QSAR) method was employed and 1-octanol/water partition coefficient (log K(ow))-dependent models were developed to study different modes of toxic action. Most phenols elicited their response via a polar narcotic mechanism and an excellent logK(ow)-dependent model was obtained. Soft electrophilicity and pro-electrophilicity were observed for some phenols and a good log K(ow)-dependent model was also achieved. Additionally, the significant dissociation of carboxyl on benzoic acid derivatives sharply reduced their toxicity. A statistically robust QSAR model was developed for all studied compounds with the combined application of log K(ow), energy of lowest unoccupied orbital (E(lumo)), heat of formation (HOF) and the first-order path molecular connectivity dices (1chi(p)).     

A novel approach to predict aquatic toxicity from molecular structure

The main aim of the study was to develop quantitative structure-activity relationship (QSAR) models for the prediction of aquatic toxicity using atom-based non-stochastic and stochastic linear indices. The used dataset consist of 392 benzene derivatives, separated into training and test sets, for which toxicity data to the ciliate Tetrahymena pyriformis were available. Using multiple linear regression, two statistically significant QSAR models were obtained with non-stochastic (R2=0.791 and s=0.344) and stochastic (R2=0.799 and s=0.343) linear indices. A leave-one-out (LOO) cross-validation procedure was carried out achieving values of q2=0.781 (scv=0.348) and q2=0.786 (scv=0.350), respectively. In addition, a validation through an external test set was performed, which yields significant values of Rpred2 of 0.762 and 0.797. A brief study of the influence of the statistical outliers in QSAR's model development was also carried out. Finally, our method was compared with other approaches implemented in the Dragon software achieving better results. The non-stochastic and stochastic linear indices appear to provide an interesting alternative to costly and time-consuming experiments for determining toxicity.     

Comparison of in vitro and in vivo acute fish toxicity in relation to toxicant mode of action

Acute toxicity to fish hepatoma cell line PLHC-1 and to juvenile rainbow trout was examined for 18 plant protection products. The main objective was to explore whether hepatoma cells could be used to predict acute toxicity in fish taking into account the mode of toxic action and compound properties. Acute fish toxicity was determined using the OECD guideline test 203 and compared to predicted baseline LC50 of acute fish toxicity calculated with a quantitative structure-activity relationship (QSAR) derived for guppy fish. Cytotoxicity was determined through the inhibition of neutral red uptake (NR(50)) into lysosomes and compared to predicted baseline cytotoxicity derived for goldfish GFS cells. In general, NR50 values were higher by a factor ranging from 3 to 3000 than the corresponding acute LC50. A weak correlation between NR50 and LC50 values was found (log/log: r2=0.62). Also the lipophilicity (log K(ow)) was not a good predictor for cytotoxicity (r2=0.43) and lethality (r2=0.57) of these pesticides. The neutral red assay is detecting general baseline toxicity only. Comparing LC50 data to QSAR results, the compounds can be classified to act as narcotics or reactive compounds with a specific mode of toxic action in fish. The results indicate that limitation of the neutral red assay in predicting acute fish toxicity. A promising alternative might be the assessment of toxicity in a set of in vitro systems addressing also cell-specific functions which are related to the mode of toxic action of the compound.     

Mechanism-based quantitative structure-activity relationships for the inhibition of substituted phenols on germination rate of Cucumis sativus

Comparative inhibition activity (GC50) of 42 structurally diverse substituted phenols on seed germination rate of Cucumis sativus was investigated. Quantitative structure-activity relationships (QSARs) were developed by using hydrophobicity (1-octanol/water partition coefficient, logKow) and electrophilicity (the energy of the lowest unoccupied molecule orbital, Eluma) for the toxicity of phenols according to their modes of toxic action. Most phenols elicited their response via a polar narcotic mechanism and a highly significant log Kow-based model was obtained (GC50 = 0.92 log Kow + 1.99, r2 0.84, n = 29). The inclusion of E(lumo) greatly improved the predictive power of the polar narcotic QSAR (GC50 = 0.88 log Kow - 0.30E(lumo) + 1.99, r2 = 0.93, n = 29). pKa proved to be an insignificant influencing factor in this study. Poor correlation with hydrophobicity and strong correlation with electrophilicity were observed for the nine bio-reactive chemicals. Their elevated toxicity was considerably underestimated by the polar narcotic logKow-dependent QSAR. The nine chemicals consist of selected nitro-substituted phenols, hydroquinone, catechol and 2-aminophenol. Their excess toxic potency could be explained by their molecular structure involving in vivo reaction with bio-macromolecules. Strong dissociation of carboxyl group of the four benzoic acid derivatives greatly decreased their observed toxicity. In an effort to model all chemicals including polar narcotics and bio-reactive chemicals, a response-surface analysis with the toxicity, logKow and E(lumo) was performed. This resulted in a highly predictive two-parameter QSAR for most of the chemicals (GC50 = 0. 70 logKow - 0.66E(lumo) + 2.17, r2 = 0.89, n = 36). Catechol and 2,4-dinitrophenol proved to be outliers of this model and their much high toxicity was explained.     

Effect of binary combination of deltamethrin+MGK-264 on the levels of phospholipid and lipid peroxidation in the snail Lymnaea acuminata

Effect of sublethal treatment of (40% and 60% of 48 h LC50) of deltamethrin+MGK on phospholipid level and rate of lipid peroxidation in nervous and foot tissue of Lymnaea acuminata were studied. Maximum reduction in phospholipid (24.10%) level and increase in rate of lipid peroxidation (586.8%) were observed in foot tissue of snail exposed to 60% of 48 h LC50 of deltamethrin+MGK 264 for 96 h. Alterations in the levels of phospholipids and rate of lipid peroxidation were time and concentration dependent. Use of MGK-264 with deltamethrin increases the toxicity of deltamethrin and their action on membrane phospholipids and rate of lipid peroxidation.     

METIER (Modular Ecotoxicity Tests Incorporating Ecological Relevance) for difficult substances--III. Effects of medium renewal and use of a carrier on the bioavailability of parathion

In this work we compared the effects of medium renewal and use of a carrier (ethanol) in the lethal toxicity of parathion to Daphnia magna Straus. Actual concentrations measured in test solutions with and without ethanol were not significantly different, indicating that there is no need for a carrier solution to assist dissolution of parathion, up to a concentration of 5 mg litre(-1). LC50 values at 48 h from bioassays with and without medium renewal were identical, indicating that in our experimental conditions, degradation of parathion was not a significant consideration. Similar 48-h LC50 values were obtained in all tests, with or without ethanol present. Thus, potential interactions between carrier and toxicant had no effect on the lethal toxicity of parathion to D. magna.