Comparative Assessment of QSAR Models for Aquatic Toxicity

This report describes a comparative assessment of QSAR models for aquatic toxicity, which was performed to evaluate the possibility of using QSAR predictions for regulatory purposes. To this end, six literature-based QSAR models for acute fish toxicity on Pimephales promelas were analyzed with respect to their ability to predict OECD Screening Information Data Set (SIDS) data for 177 High Production Volume (HPV) chemicals. The first two models are QSARs recommended by the EU Technical Guidance Document on chemical risk assessment for the polar and non-polar narcotic mechanisms of action. The third model was developed by ECB to represent the narcosis mechanism of action, including both non-polar and polar action. The fourth model is more general than the previous ones since by including an electrophilicity descriptor it is supposed to describe potentially bioreactive (electrophilic) chemicals. The fifth model is a more recently proposed model based on hydrophobic and polar atom-type electrotopological state (E-state) indices. The sixth model is a commercially available neural network software program, developed by the TerraBase Inc., for the computation of acute (96hr) median lethal concentrations (LC50) of organic substances. The SIDS substances were classified according to expected Mode of Action according to a consensus classification scheme based on three different schemes. The six models were also assessed according to the extent to which they meet OECD (Organisation for Economic Cooperation and Development) principles for the validation of (Q)SARs for regulatory purposes. The applicability domain was visualised by means of the William’s plot. For each model, a comparison between predictions and experimental fish toxicity was performed by calculating the number of chemicals with predicted effect concentrations within factors of 10, 100 and 1000 of the corresponding SIDS test data. For each model the ratio was calculated first by using the entire SIDS data set and then by using only the chemicals falling with the model applicability domain. The results show that when the model domain was taken into account, the ratio was always near one and in the range from 0.1 to 10. The range became much larger when the applicability domain constraints were ignored. The results of this study support the view that the regulatory application of a QSAR model should be based on a suitable definition of the model applicability domain in order to identify reliable predictions. . CONTENTS LIST OF ABBREVIATIONS 1

[1]  Hxugo Kubiny.  Variable Selection in QSAR Studies. I. An Evolutionary Algorithm , 1994 .

[2]  Gilman D. Veith,et al.  Structure–Toxicity Relationships for the Fathead Minnow, Pimephales promelas: Narcotic Industrial Chemicals , 1983 .

[3]  D. M. Allen.  Mean Square Error of Prediction as a Criterion for Selecting Variables , 1971 .

[4]  J. Hermens,et al.  Classifying environmental pollutants , 1992 .

[5]  T W Schultz,et al.  Probabilistic neural network modeling of the toxicity of chemicals to Tetrahymena pyriformis with molecular fragment descriptors , 2002, SAR and QSAR in environmental research.

[6]  Matthew Clark,et al.  The Probability of Chance Correlation Using Partial Least Squares (PLS) , 1993 .

[7]  Gilman D. Veith,et al.  A QSAR Approach for Estimating the Aquatic Toxicity of Soft Electrophiles [QSAR for Soft Electrophiles] , 1993 .

[8]  C. Russom,et al.  Predicting modes of toxic action from chemical structure: Acute toxicity in the fathead minnow (Pimephales promelas) , 1997 .

[9]  S. A. bano C. D. nn W. I. i Wold,et al.  Pattern recognition: finding and using regularities in multivariate data Food research, how to relate sets of measurements or observations to each other , 1983 .

[10]  Gilman D. Veith,et al.  Structure-Toxicity Relationships for Industrial Chemicals Causing Type (II) Narcosis Syndrome , 1987 .

[11]  Gerrit Schüürmann,et al.  Feed Forward Backpropagation Neural Networks and their Use in Predicting the Acute Toxicity of Chemicals to the Fathead Minnow , 1997 .

[12]  Ş. Niculescu,et al.  Modeling acute toxicity of chemicals to Daphnia magna: A probabilistic neural network approach , 2001, Environmental toxicology and chemistry.

[13]  Worth Andrew,et al.  Preliminary Analysis of an Aquatic Toxicity Dataset and Assessment of QSAR Models for Narcosis , 2005 .

[14]  J. Topliss,et al.  CHANCE FACTORS IN STUDIES OF QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIPS , 1980 .

[15]  Scott D. Kahn,et al.  Current Status of Methods for Defining the Applicability Domain of (Quantitative) Structure-Activity Relationships , 2005, Alternatives to laboratory animals : ATLA.

[16]  B. Efron.  The jackknife, the bootstrap, and other resampling plans , 1987 .

[17]  J. Huuskonen.  QSAR modeling with the electrotopological state indices: predicting the toxicity of organic chemicals. , 2003, Chemosphere.

[18]  Ş. Niculescu,et al.  On the PNN Modeling of Estrogen Receptor Binding Data for Carboxylic Acid Esters and Organochlorine Compounds , 2001 .

[19]  B. Efron.  Better Bootstrap Confidence Intervals , 1987 .