Quantitative Structure-Activity Relationships (QSAR) for Pesticide Regulatory Purposes Book

Forewords     xi
The DEMETRA Project: An Innovative Contribution to Regulatory QSAR     xi
Reference     xiv
Preface   Emilio Benfenati   Mose Casalegno     xv
The Pesticides and their Ecotoxicological Properties     xv
Moving Forwards the use of QSAR to Predict Toxicological Properties     xvi
The DEMETRA Project     xvii
The Book Chapters     xix
Acknowledgement     xx
Disclaimer     xx
References     xxi
QSARs for regulatory purposes: the case for pesticide authorization   Emilio Benfenati   Mark Clook   Steven Fryday   Andy Hart     1
Overview of the Current Pesticide Authorization Procedure     1
Description of the current pesticide legislation (EU Directive 91/414/EEC)     1
Outline of the ecotoxicology tests required for pesticide authorization under 91/414/EEC     3
How frequently are certain studies submitted and how many studies are submitted to address an Annex point?     3
What changes are likely to occur that could alter the frequency and number of toxicity studies submitted?     17
Introduction on QSARS for Pesticides     19
Regulatory Perspectives in the use of QSARs     22
Current use of QSARs in regulation     22
Potential barriers for using QSARs in the pesticide authorization procedure     27
End-user criteria for the use of QSARs in regulatory assessment     29
Quality Criteria for Modelling Ecotoxicity Data     30
Data quality and precision required     30
Quality criteria to be applied to ecotoxicity data used in a QSAR     30
Degree of precision required of QSARs for pesticide assessments     39
Toxicity End-Points with a High Potential to be Replaced with a QSAR Approach     47
Data availability     47
Number of animals tested     48
Study costs     50
End-points with high potential for replacement with a QSAR     51
Priority end-points     54
References     54
Databases for pesticide ecotoxicity   Emilio Benfenati   Elena Boriani   Marian Craciun   Ladan Malazizi   Daniel Neagu   Alessandra Roncaglioni     59
Introduction     59
Data Availability     60
The EPA-OPP database     61
The SEEM database     62
The BBA database     64
Other databases     64
Selection of the Data      65
Key features in the choice of the database     66
Comparison of the data internally to the database     67
Data Representation for Predictive Toxicology     70
A public database example: DSSTox     72
Current toxicity database limitations     72
XML-based standards in chemistry and toxicology     73
PToxML - a simple XML-based description in predictive toxicology     73
The Characteristics of the Final Data Sets     78
Conclusions     78
Acknowledgments     80
References     80
Characterization of chemical structures   Emilio Benfenati   Mose Casalegno   Jane Cotterill   Nick Price   Morena Spreafico   Andrey Toropov     83
Introduction     83
Characterization of Bi-dimensional Structures     85
Preprocessing of compounds in the data set     86
Geometrical isomers     88
Tautomers     88
Characterization of Tri-dimensional Structures     89
Crystallographic data     89
Conformational searching and energy minimization     90
Stereoisomers     93
Procedure for the quality control of the chemicals and chemical structures     94
Chemical Structure File Formats     94
Bi-dimensional descriptors     95
Tri-dimensional descriptors     99
Fragments and Residues in DEMETRA     102
References     107
Algorithms for (Q)SAR model building   Qasim Chaudhry   Jacques Chretien   Marian Craciun   Gongde Guo   Frank Lemke   Johann-Adolf Muller   Daniel Neagu   Nadege Piclin   Marco Pintore   Paul Trundle     111
Introduction     111
Methods for Data Pre-Processing and Selecting Descriptors     112
Models with Classifiers     114
FISs     114
Adaptive fuzzy partition     116
k-NN methods     120
Models with Regression Systems     125
Traditional linear regression QSAR models     125
ANNs and fuzzy neural networks     128
Self-organizing statistical-learning networks     134
Conclusions     143
References     144
Hybrid systems   Nicolas Amaury   Emilio Benfenati   Severin Bumbaru   Antonio Chana   Marian Craciun   Jacques R. Chretien    Giuseppina Gini   Gongde Guo   Frank Lemke   Viorel Minzu   Johann-Adolf Muller   Daniel Neagu   Marco Pintore   Silviu Augustin Stroia   Paul Trundle     149
Introduction: Goals of the Hybrid Systems     149
Our Hybrid Approach for Quantitative Structure-Activity Relationship     151
Gating Networks     152
Introduction     152
Gating networks for predictive toxicology - a new approach based on descriptors clustering     154
Hybrid neural fuzzy systems     157
Gating networks as HISs - a data-driven approach     159
Multi-Classifier Systems     160
Approaches for multi-classifier systems     161
An architecture of MCS     162
Classifiers     163
Combination Methods     163
Distributed multi-classifier systems     165
Neural Ik- and Ek-Based Systems - Introduction of the Prototype NIKE     167
Experiment 1     173
Experiment 2     174
Rule-Based Systems     175
Self-Organizing Statistical Learning Networks     177
Conclusions     180
References     180
Validation of the models   Emilio Benfenati   Jacques R. Chretien   Giuseppina Gini   Nadege Piclin   Marco Pintore   Alessandra Roncaglioni     185
Introduction     185
Selection of the Training and Test Sets     186
Internal Validation and Robustness     187
External Validation     189
Validation Parameters for Classifiers: Matrix of Confusion     191
Graphical Evaluation of the Models: The Receiver Operating Characteristic and Regression Error Characteristic Curves     192
How to Deal with False Negatives/False Positives     196
The Applicability Domain     197
References     198
Results of DEMETRA models   Nicolas Amaury   Emilio Benfenati   Elena Boriani   Mose Casalegno   Antonio Chana   Qasim Chaudhry   Jacques R. Chretien   Jane Cotterill   Frank Lemke   Nadege Piclin   Marco Pintore   Chiara Porcelli   Nicholas Price   Alessandra Roncaglioni   Andrey Toropov     201
Overview of Results with the Regression Approach     201
Overview of the Prediction Results Obtained by Classification Methods     208
Data sets and toxicity intervals      208
Descriptors selection and classification results     208
Conclusions about classification results     214
Overview of Results of Local Models     215
Chemical classes     215
Overview of Results Obtained with the Hybrid Models     221
Hybrid model for rainbow trout     222
Outliers and the applicability domain     226
Hybrid model for water flea (Daphnia magna)     249
Hybrid model for quail: oral exposure     266
Hybrid model for quail: dietary exposure     269
Hybrid model for acute contact toxicity of honey bee     274
Conclusions     279
Acknowledgments     280
References     281
The quality criteria of the DEMETRA models for regulatory purposes   Emilio Benfenati     283
The OECD Guidelines for QSAR Models     283
Introduction     283
The identification of the regulation     284
The criteria for the endpoint selection     284
The model utility     285
The endpoint selection: identification of the guidelines     285
The accordance of the toxicity data to the guidelines     286
The check of quality data     286
The definition of the model components. OECD principle number 2: an unambiguous algorithm     286
The selection of the toxicity values of the data set     287
The characterization of the uncertainty of the experimental data     287
The chemical structures     288
The chemical descriptors     289
The algorithms     289
The performances of the model     289
The reproducibility of the models     290
The false-negative issue     290
The applicability domain     291
The quality control     292
The use of the model     292
The Specificity of the QSAR Models for Regulatory Purposes     292
The Probabilistic Meaning of the Model, the Prediction of the Effect, and the Prediction of the Mechanism     295
The probabilistic nature of the models     295
The mechanistic basis of the models     296
The final model and the ways to obtain it     297
The Benefits of the DEMETRA Models     297
Future Perspectives     298
References     301
The use of the DEMETRA models   Emilio Benfenati   Marian Craciun   Daniel Neagu     303
Introduction     303
The Users of the DEMETRA Models      303
Ownership of the Software     304
Using DEMETRA Models     306
Chemical Restrictions of the DEMETRA Models     307
The Format for Model Presentation for DEMETRA: HISML     308
References     312
Appendices     315
Summary of responses to DEMETRA survey     317
Toxicity values for five ECOTOX data sets for pesticide     323
Example procedures in molecular modelling     463
The descriptors selected for each data set     469
List of abbreviations     487
Software tool for toxicity prediction of pesticides, candidate pesticides, and their derivatives (user guide)     493
Index     505