PREDICTIVE TOXICOLOGY - CHAPTER 1 docx

Toxicological Information for Use in Predictive Modeling: Quality, Sources, and Databases.. 98 High Quality Data Sources for Predictive Modeling.. The Use of Expert Systems for Toxicolog

PREDICTIVE TOXICOLOGY

edited by

Christoph Helma

University of Freiburg, Germany

PREDICTIVE TOXICOLOGY

Published in 2005 by

Taylor & Francis Group

6000 Broken Sound Parkway NW

Boca Raton, FL 33487–2742

#2005 by Taylor & Francis Group, LLC

No claim to original U.S Government works

Printed in the United States of America on acid-free paper

10 9 8 7 6 5 4 3 2 1

International Standard Book Number-10: 0–8247–2397–X (Hardcover)

This book contains information obtained from authentic and highly regarded sources Reprinted material is quoted with permission, and sources are indicated A wide variety

of references are listed Reasonable efforts have been made to publish reliable data and information, but the author and the publisher cannot assume responsibility for the validity

of all materials or for the consequences of their use.

No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access

www.copyright.com ( http:==www.copyright.com= ) or contact the Copyright Clearance Center, Inc (CCC) 222 Rosewood Drive, Danvers, MA 01923, 978–750–8400 CCC is

a not-for-profit organization that provides licenses and registration for a variety of users For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trade-marks, and are used only for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data Catalog record is available from the Library of Congress

Visit the Taylor & Francis Web site at

http :==www.taylorandfrancis.com

Contributors ix

1 A Brief Introduction to Predictive

Toxicology 1

Christoph Helma

What Is Predictive Toxicology? 1

Ingredients of a Predictive Toxicology System 3 Concluding Remarks 7

2 Description and Representation

of Chemicals 11

Wolfgang Guba

Introduction 11

Fragment-Based and Whole Molecule Descriptor

Schemes 13

Fragment Descriptors 14

Topological Descriptors 19

3D Molecular Interaction Fields 23

Other Approaches 27

iii

3 Computational Biology

and Toxicogenomics 37

Kathleen Marchal, Frank De Smet, Kristof Engelen, and Bart De Moor

Introduction 37

Microarrays 41

Analysis of Microarray Experiments 46

Conclusions and Perspectives 74

4 Toxicological Information for Use

in Predictive Modeling: Quality,

Sources, and Databases 93

Mark T D Cronin

Introduction 93

Requirements for Toxicological Data for Predictive Toxicity 98

High Quality Data Sources for Predictive

Modeling 104

Databases Providing General Sources of

Toxicological Information 104

Databases Providing Sources of Toxicological Information for Specific Endpoints 110

Sources of Chemical Structures 119

Sources of Further Toxicity Data 121

Conclusions 123

5 The Use of Expert Systems for Toxicology Risk Prediction 135

Simon Parsons and Peter McBurney

Introduction 136

Expert Systems 137

Expert Systems for Risk Prediction 147

Systems of Argumentation 153

Summary 167

6 Regression- and Projection-Based Approaches

in Predictive Toxicology 177

Lennart Eriksson, Erik Johansson, and

Torbjo¨rn Lundstedt

Introduction 178

Characterization and Selection of Compounds:

Statistical Molecular Design 179

Data Analytical Techniques 182

Results for the First Example—Modeling and Predicting

In Vitro Toxicity of Small Haloalkanes 190 Results for the Second Example—Lead Finding and QSAR-Directed Virtual Screening of

Hexapeptides 203

Discussion 211

7 Machine Learning and Data Mining 223

Stefan Kramer and Christoph Helma

Introduction 223

Descriptive DM 231

Predictive DM 239

Literature and Tools=Implementations 246 Summary 249

8 Neural Networks and Kernel Machines for Vector and Structured Data 255

Paolo Frasconi

Introduction 255

Supervised Learning 258

The Multilayered Perceptron 268

Support Vector Machines 279

Learning in Structured Domains 288

Conclusion 299

9 Applications of Substructure-Based SAR in

Toxicology 309

Herbert S Rosenkranz and Bhavani P Thampatty

Introduction 309

The Role of Human Expertise 311

Model Validation: Characterization and

Interpretation 316

Congeneric vs Non-congeneric Data Sets 335 Complexity of Toxicological Phenomena and Limitations

of the SAR Approach 343

Mechanistic Insight from SAR Models 345

Application of SAR to a Dietary Supplement 348 SAR in the Generation of Mechanistic

Hypotheses 354

Mechanisms: Data Mining Approach 355

An SAR-Based Data Mining Approach to Toxicological Discovery 357

Conclusion 361

10 OncoLogic: A Mechanism-Based Expert System for Predicting the Carcinogenic Potential of

Chemicals 385

Yin-Tak Woo and David Y Lai

Introduction 385

Mechanism-Based Structure–Activity Relationships Analysis 387

The OncoLogic Expert System 390

11 META: An Expert System for the Prediction of Metabolic Transformations 415

Gilles Klopman and Aleksandr Sedykh

Overview of Metabolism Expert Systems 415 The META Expert System 416

META Dictionary Structure 417

META Methodology 418

META_TREE 419

12 MC4PC—An Artificial Intelligence Approach to the Discovery of Quantitative Structure–Toxic Activity Relationships 423

Gilles Klopman, Julian Ivanov, Roustem Saiakhov, and Suman Chakravarti

Introduction 423

The MCASE Methodology 427

Recent Developments: The MC4PC Program 433 BAIA Plus 438

Development of Expert System Predictors Based on MCASE Results 443

Conclusion 451

13 PASS: Prediction of Biological Activity Spectra for Substances 459

Vladimir Poroikov and Dmitri Filimonov

Introduction 459

Brief Description of the Method for Predicting Biological Activity Spectra 461

Application of Predicted Biological Activity Spectra

in Pharmaceutical Research and

Development 471

Future Trends in Biological Activity Spectra

Prediction 474

14 lazar: Lazy Structure–Activity Relationships for Toxicity Prediction 479

Christoph Helma

Introduction 479

Problem Definition 482

The Basic lazar Concept 484

Detailed Description 485

Results 491

Learning from Mistakes 493

Conclusion 495

Suman Chakravarti Case Western Reserve University,

Cleveland, Ohio, U.S.A.

Mark T D Cronin School of Pharmacy and Chemistry,

John Moores University, Liverpool, U.K.

Bart De Moor ESAT-SCD, K.U Leuven, Leuven, Belgium Frank De Smet ESAT-SCD, K.U Leuven, Leuven, Belgium Kristof Engelen ESAT-SCD, K.U Leuven, Leuven, Belgium Lennart Eriksson Umetrics AB, Umea˚, Sweden

Dmitri Filimonov Institute of Biomedical Chemistry of Russian Academy of Medical Sciences, Moscow, Russia

Paolo Frasconi Dipartimento di Sistemi e Informatica,

Universita` degli Studi di Firenze, Firenze, Italy

Wolfgang Guba F Hoffmann-La Roche Ltd, Pharmaceuticals Division, Basel, Switzerland

Christoph Helma Institute for Computer Science, Universita¨t Freiburg, Georges Ko¨hler Allee, Freiburg, Germany

ix

Julian Ivanov MULTICASE Inc., Beachwood, Ohio, U.S.A Erik Johansson Umetrics AB, Umea˚, Sweden

Gilles Klopman MULTICASE Inc., Beachwood, Ohio, and Department of Chemistry, Case Western Reserve University, Cleveland, Ohio, U.S.A.

Stefan Kramer Institut fu¨r Informatik, Technische Universita¨t Mu¨nchen, Garching, Mu¨nchen, Germany

David Y Lai Risk Assessment Division, Office of Pollution Prevention and Toxics, U.S Environmental Protection Agency, Washington, D.C., U.S.A.

Torbjo¨rn Lundstedt Acurepharma AB and BMC, Uppsala, Sweden

Peter McBurney Department of Computer Science,

University of Liverpool, Liverpool, U.K.

Kathleen Marchal ESAT-SCD, K.U BMC, Leuven, Leuven, Belgium

Simon Parsons Department of Computer and Information Science, Brooklyn College, City University of New York, Brooklyn, New York, U.S.A.

Vladimir Poroikov Institute of Biomedical Chemistry of Russian Academy of Medical Sciences, Moscow, Russia

Herbert S Rosenkranz Department of Biomedical Sciences, Florida Atlantic University, Boca Raton, Florida, U.S.A.

Roustem Saiakhov MULTICASE Inc., Beachwood, Ohio, U.S.A Aleksandr Sedykh Department of Chemistry, Case Western Reserve University, Cleveland, Ohio, U.S.A.

Bhavani P Thampatty Department of Environmental and Occupational Health, Graduate School of Public Health, University

of Pittsburgh, Pittsburgh, Pennsylvania, U.S.A.

Yin-Tak Woo Risk Assessment Division, Office of Pollution Prevention and Toxics, U.S Environmental Protection Agency, Washington, D.C., U.S.A.

A Brief Introduction to Predictive Toxicology

CHRISTOPH HELMA Institute for Computer Science, Universita¨ t Freiburg, Georges Ko¨ hler Allee, Freiburg, Germany

1 WHAT IS PREDICTIVE TOXICOLOGY?

The public demand for the protection of human and environ-mental health has led to the establishment of toxicology as the science of the action of chemicals on biological systems Toxicological research is focused presently very much on the elucidation of the cellular and molecular mechanisms of toxi-city and the application of this knowledge in safety evalua-tion and risk assessment This is essentially a predictive strategy (Fig 1): Toxicologists study the action of chemicals

in simplified biological systems (e.g., cell cultures, laboratory animals) and try to use these results to predict the potential impact on human or environmental health

1

Predictive toxicology, as we understand it in this book, does something very similar (Fig 1): In predictive toxicology, we try to develop procedures (algorithms in computer science terms) that are capable to predict toxic effects (the output) from chemical and biological information (the input)

Figure 1 summarizes also the key ingredients of

a predictive toxicology system First, we need a description

of chemicals and biological systems as input for predi-ctions This information is processed by the prediction algorithm, to generate a toxicity estimation as output We can also distinguish between data (input and output) and algorithms

Figure 1 Abstraction of the predictive toxicology process.

2 INGREDIENTS OF A PREDICTIVE

TOXICOLOGY SYSTEM

2.1 Chemical, Biological, and

Toxicological Data

Most of the research in predictive toxicology has been devoted

to the development of algorithms, but for a good performance, the data aspect is at least equally important It is in principle possible to use many different types of information to describe chemical and biological systems The key problem in predic-tive toxicology is to identify the parameters that are relevant for a particular toxic effect The situation is relatively easy, if the underlying biochemical mechanisms are well known In this case, we can determine a rather limited set of para-meters, that might be relevant for our purpose In practice, however, biochemical mechanisms are frequently unknown and=or too complex, to determine a suitable set

of parameters a priori Methods for parameter selection are therefore an important research topic in predictive toxicology

Toxicity data are needed for two purposes: First of all, we need to validate prediction methods, and this can be done by comparing the predictions with realworld measurements But

we can use toxicity data also as input to one of the data driven approaches that are capable of generating prediction models automatically from empirical data (Fig 2) In this case, the quality of the prediction model is largely determined by the quality of the input data

Despite many possibilities, practical applications have focused on a relatively small set of chemical and biological features The most popular chemical features are closely related to the chemical structure (e.g., presence= absence of certain substructures) or to properties, that can be calculated from the chemical structure (e.g., physico-chemical properties) As no experimental work is needed to obtain this type of data, the rationale for their choice is obvious, but other substance-related information (e.g., biological activities in screening assays, IR-spectra) can be used as well

A Brief Introduction to Predictive Toxicology 3

Up to now information about biological systems has been rarely considered in predictive toxicology Biological systems have been treated as ensembles of uniform members (e.g., equal individuals), without any biological variance The expli-cit consideration of the biological part of the equation will be

an interesting research topic of the next years.a

Chemical, biological and toxicological data and their repre-sentation are the topics of the first section of this book It con tains the chapters Description and Representation of Chemicals by Guba (1), Computational Biology and Toxicogenomics by Marcha l et al (2), and Toxicological Informa-tion for Use in Predictive Modeling: Quality, Sources, and Data-bases by C ronin (3 )

a

The chapter from Marchal et al (2) provides some examples how to use biological information for predictive purposes.

Figure 2 Abstraction of a data driven approach in predictive toxicology.

2.2 Prediction Algorithms

For the prediction algorithm, we have the choice between two strategies We can try to mimic a human expert by building

an expert system, or we can try to deduce a prediction model from empirical data by a data-driven approach as in Fig 2 The basics of expert systems and some exemplary appli-cations are the topic of Parson and McBurney’s chapter, The Use of Expert Systems for Toxicology Risk Prediction (4) Two of the programs [META (5) and OncoLogic (6)] discussed

in the section Implementations of Predictive Toxicology Sys-tems are also expert sysSys-tems

If we intend to generate a prediction model from experimen-tally determined toxicity data as in Fig 2, we have the choice between many different methods Statistical methods, for exam-ple, have been successfully applied in quantitative structureac-tivity relationships (QSAR) for decades Eriksson et al.(7) describe statistical techniques in t he chapter entitled Regres-sion- and Projection-Based Approaches in Predictive Toxicology

More recently, techniques originating from artificial intelligence research have been used in predictive toxicology These computer-science oriented developments are summar-ized in two chapters: Machine Learning and Data Mining

by Kramer and Helma (8) and Neural Networks and Kernel Machines for Vector and Structured Data by Frasconi (9) Three programs of the section Implementations of Predictive Toxicology Systems [MC4PC (10), PASS (11), lazar (12)] use such a data-driven approach

I want to stress the point that similar predictions can be obtained with a variety of methods The choice of the method for a particular purpose will depend largely on the scope

of the application, present research trends and the personal preferences of the individual researcher

2.3 Application Areas

The primary aim of predictive toxicology is, of course, the pre-diction of toxic activities of untested compounds This enables chemical and pharmaceutical companies, for example, to eval-uate potential side effects of candidate structures even without

A Brief Introduction to Predictive Toxicology 5