Computer-based Predictive Toxicology: Advances and Impact of Cheminformatics on the Safety-oriented Design of New Products
Current advances in computer-based predictive toxicology offer the potential to create more advanced environments for the screening and prediction of safety issues due to chemical and drug adverse side effects, drug-drug and chemical-system interactions, and chemical and drug toxicologies in the environment and the human body. Advances in this growing field also offer the potential to replace or reduce the need for animal testing and to reduce later stage clinical trial failures or new product development rejection. Acceleration of progress in practical applications requires the creation of interoperable environments, knowledge sharing, data integration, algorithm development, and extensive validation and testing.
Numerous opportunities exist in this field for scientific advances, but also for innovation, service and product development, and value creation. Additionally, significant collaboration approaches are a scientific, industry and society imperative to advance this field and the safety of new products and all society members. Collaborative approaches need to support the multidisciplinary networking and collaboration between computer scientists, biologists, chemists, toxicologists, product development and clinical and environmental researchers, and to network groups, centers, initiatives, projects and data into interoperable semantic frameworks, systems, knowledge bases and virtual organisations.
At our Predictive Toxicology session chaired by Artem Cherkasov (University of British Columbia) running 17 October 2008 at Bryn Mawr recent developments in the field of predictive toxicology will be presented and discussed.
The session will be preceded the evening of October 16 by a Knowledge Café to discuss Collaboration Opportunities in Predictive ADME & Predictive Toxicology.
A description of the session with presentation abstracts follows. Please add your comments, discussion or questions at the end of the post.
Predictive Toxicology
http://echeminfo.com/COMTY_confprog08predtox
(Please follow continuation here to read abstracts. Comments can be made at the end.)
Abstracts
A Hard Look at Predictive Modeling: How Much Data is Enough?
Curt M. Breneman, Director, RECCR Center, RPI Department of Chemistry, 110 8th St, Troy, NY 12180, USA
A frequent concern within the predictive cheminformatics community revolves around knowing for any given situation when sufficient data exists to create a usable model with quantifiable reliability when applied to specific test sets. This problem is particularly acute in modeling toxicity, since assay data is often confounded by high in-vivo error bars and multiple mechanisms of action. Coupled with a proliferation of easily available molecular property descriptors and non-linear modeling methods, this creates an environment conducive to the creation of highly "local" models which are only valid within the confined chemical space of the training set. This phenomenon frequently defies objective measurement, since the size and scope of a model's applicability domain depends on a number of factors, including the assay endpoint, the descriptor set and details of the machine learning method used, as well as other factors.
This presentation will summarize our efforts to predict model performance on a number of cases using different descriptor sets through an analysis of its stability to a loss of training information, coupled with an analysis of similarity of the training data with test cases. This analysis involved the development of the "Rank Order Entropy" (ROE) metric as well as an equivalent floating point metric for determining if enough data is present to support the production of a predictive model. Examples of its use in training and test cases will be provided in the presentation.
The Use of Conventional Drug Design Technologies for Identification of Potential Endocrine Disruptors Interacting with Sex-Hormone Binding Globulin in Zebra Fish
Artem Cherkasov, University of British Columbia
Anthropogenic compounds with the capacity to interact with the steroid-binding site of sex hormone binding globulin (SHBG) pose health risks to humans and other vertebrates including fish. Building on studies of human SHBG, we have applied in silico drug discovery methods to identify potential binders for SHBG in zebrafish (Danio rerio) as a model aquatic organism. Computational methods, including; homology modeling, molecular dynamics simulations, virtual screening, and 3D QSAR analysis, successfully identified 6 non-steroidal substances from the ZINC chemical database that bind to zebrafish SHBG (zfSHBG) with low-micromolar to nanomolar affinities, as determined by a competitive ligand-binding assay. We also screened 80,000 commercial substances listed by the European Chemicals Bureau and Environment Canada, and 6 non-steroidal hits from this in silico screen were tested experimentally for zfSHBG binding. All 6 of these compounds displaced the [3H]5α-dihydrotestosterone used as labeled ligand in the zfSHBG screening assay when tested at a 33 μM concentration, and 3 of them (hexestrol, 4-tert-octylcatechol, and dihydrobenzo(a)pyren-7(8H)-one) bind to zfSHBG in the micromolar range. The study demonstrates the feasibility of large scale in silico screening of anthropogenic compounds that may disrupt or highjack functionally important protein:ligand interactions. Such studies could increase the awareness of hazards posed by existing commercial chemicals at relatively low cost.
The OpenTox Predictive Toxicology Project
Barry Hardy, Douglas Connect, Switzerland
The goal of the OpenTox project is to develop a predictive toxicology framework, that provides a unified access to toxicological data, (Q)SAR models and toxicological information.
The OpenTox framework will provide tools for the integration of data from various sources (public and confidential), for the generation and validation of (Q)SAR models for toxic effects, libraries for the development and seamless integration of new (Q)SAR algorithms, and scientifically sound validation routines. OpenTox will attract users from a variety of research areas:
• Toxicological and chemical experts (e.g. risk assessors, drug designers, researchers)
• (Q)SAR model developers and algorithm developers
• Non-(Q)SAR specialists requiring access to Predictive Toxicology models and data
The OpenTox project will move beyond existing attempts to solve individual research issues within this area, by providing a flexible, extensible, and user friendly framework that integrates existing solutions as well as providing easy access to new developments.
OpenTox will be relevant for the implementation of REACH as it allows regulatory and industrial risk assessors to access experimental data, (Q)SAR models and toxicological information from a unified, simple-to-use interface, that adheres to European and international regulatory requirements (e.g. OECD Guidelines for (Q)SAR validation, QSAR Model Reporting Formats (QMRF)). For maximum transparency OpenTox will be published as an open source project. This will allow a critical evaluation of the implemented algorithms, ensure a widespread dissemination and will attract external developers. Facilities for the inclusion of confidential in-house data and for accessing and integrating commercial prediction systems will be included.
The OpenTox framework will be populated initially with high-quality data and (Q)SAR models for chronic, genotoxic and carcinogenic effects. These are the endpoints, where computational methods promise the greatest potential reduction in animal testing, that would be required for the implementation of REACH. The impact of OpenTox will however go beyond REACH, industrial chemicals and long-term effects, because reliable toxicity estimates are also needed for other products (e.g., pharmaceuticals, cosmetics, food-additives) and endpoints (e.g,. sensitisation, liver-toxicity, cardio-toxicity).
The proposed framework will actively support the development of new (Q)SAR models by automating routine tasks, providing a testing and validation environment and allowing the easy addition of new data. It will also support the development of new algorithms and avoid duplicated work by providing easy access to common components, validation routines and an easy comparison with benchmark techniques. For this reason we expect, that OpenTox will lead to (Q)SAR models for further toxic endpoints and generally improve the acceptance and reliability of (Q)SAR models.
Project Coordinator: Barry Hardy, Douglas Connect, Switzerland
Principal Investigators
Christoph Helma, Andreas Karwath, Stefan Kramer, David Gallagher, Romualdo Benigni, Nina Jeliazkova, Haralambos Sarimveis, Vladimir Poroikov, Indira Ghosh, Sylvia Escher
Project Partners
Douglas Connect, In Silico Toxicology, Ideaconsult, Istituto Superiore di Sanita', Technical University of Munich, Albert Ludwigs University Freiburg, National Technical University of Athens, David Gallagher, Institute of Biomedical Chemistry of the Russian Academy of Medical Sciences, Seascape Learning and the Fraunhofer Institute for Toxicology & Experimental Medicine
Advisory Board
European Center for the Validation of Alternative Methods, European Chemicals Bureau, U.S Environmental Protection Agency, U.S. Food & Drug Administration, Nestle, Roche, AstraZeneca, LHASA, University North Carolina, EC Environment Directorate General, Organisation for Economic Co-operation & Development, CADASTER and Bayer Healthcare
Project Web Site: OpenTox.org
OpenTox is a European Commission funded FP7 Research Project (1 September 2008 – 2011)
New Lazar Developments and Data Mining Techniques for the Identification of Structural Alerts
Andreas Maunz and Christoph Helma, Freiburg Center for Data Analysis and Modelling
This talk focuses on the extension of the Lazar system for regression problems and on feature mining techniques for the efficient identification of statistically significant substructures.
We present a novel activity-specific kernel to obtain predictions from a training set with a modified k-nearest-neighbor approach. Endpoints modeled include Fathead Minnow Acute Toxicity, Maximum Recommended Therapeutic Dose and IRIS Lifetime Cancer Risk. The new kernel provides results superior to the well-established Tanimoto kernel and individual predictions are interpretable for toxicological experts without a data mining background.
The identification of structural alerts has a long tradition in toxicological research. While initial approaches relied on expert knowledge alone, recent techniques incorporate data mining techniques together with statistical criteria. We present a novel objective feature mining technique that can be used without prior knowledge of toxicological mechanisms. This makes it especially useful for the investigation of endpoints that are poorly understood and for the generation of hypotheses about toxicological mechanisms.
EPA DSSTox and ToxCastTM Project Updates: Generating New Data and Linkages in Support of Public Toxico-Cheminformatics Efforts
Ann Richard (1), Maritja Wolf (2), Thomas Transue (2), ClarLynda Williams-Devane (3), and Richard Judson (1)
(1) National Center for Computational Toxicology, US EPA, RTP, NC 27711; (2) Lockheed Martin – Contractor to the US EPA, RTP, NC 27711; (3) NC State Univ. Bioinformatics Graduate Program, US EPA Student COOP, RTP, NC 27711
EPA’s National Center for Computational Toxicology is generating data and capabilities to support a new paradigm for toxicity screening and prediction. The DSSTox project is improving public access to quality structure-annotated chemical toxicity information in less summarized forms than traditionally employed in SAR modeling, and in ways that facilitate data-mining and data read-across. The DSSTox Structure-Browser provides structure searchability across the full published DSSTox toxicity-related inventory, enables linkages to and from previously isolated toxicity data resources (soon to include public microarray resources GEO, ArrayExpress, and CEBS), and provides link-outs to cross-indexed public resources such as PubChem, ChemSpider, and ACToR. The published DSSTox inventory and bioassay information also have been integrated into PubChem allowing a user to take full advantage of PubChem structure-activity and bioassay clustering features. Phase I of the ToxCastTM project has generated high-throughput screening (HTS) data from several hundred biochemical and cell-based assays for a set of 320 chemicals, mostly pesticide actives, with rich toxicology profiles. DSSTox and ACToR are providing the primary cheminformatics support for ToxCastTM and collaborative efforts with the National Toxicology Program’s HTS Program and the NIH Chemical Genomics Center. DSSTox will also be a primary vehicle for publishing ToxCastTM ToxRef summarized bioassay data for use by modelers. Incorporating and expanding traditional SAR concepts into this new high-throughput and data-rich world pose conceptual and practical challenges, but also offer great promise for improving predictive capabilities. This work was reviewed by EPA and approved for publication, but does not necessarily reflect EPA policy.
The FDA’s Endocrine Disruptor Knowledge Base (EDKB)– Lessons Learned in QSAR Modeling and Applications
Weida Tong, Director of Center for Toxicoinformatics, NCTR/FDA
Considerable scientific, regulatory and popular press attention has been devoted to the Endocrine Disrupting Chemicals (EDCs). A larger number of potential estrogenic EDCs are associated with products regulated by the Food and Drug Administration (FDA), including plastics used in food packaging, phytoestrogens, food additives, pharmaceuticals, cosmetics, etc. Given the huge number of chemicals, many commercially important, and the expense of testing, SAR/QSAR has been considered to be an important priority setting strategy for subsequent experimentation. At the U.S. FDA’s National Center for Toxicological Research (NCTR), we have conducted the Endocrine Disruptor Knowledge Base (EDKB) project, of which SAR/QSARs is a major component. We have developed predictive models for estrogen and androgen receptor binding. The strengths and weaknesses of various QSAR methods were assessed to select those most appropriate for regulatory priority setting. This presentation, rather than presenting the work and results of the EDKB program in an exhaustive manner, selectively discusses salient concepts, issues, and challenges, endeavoring to achieve a tutorial outcome. In particular, concepts such as designing training sets, living models, use of QSARs in a regulatory context, predictive model validation, QSAR applicability domain and prediction confidence estimates are among topics to highlight. The concepts are presented and discussed using EDKB program results to provide qualitative and quantitative illustrations and examples. We believe the experience and lessons learned in the EDKB program will prove valuable to practitioners of QSAR should they endeavor to extend predictive systems to real-world regulatory implementations.
Predictive Chemical Toxicity Models Using in vitro - in vivo Correlations Enriched by Cheminformatics
Hao Zhu (1,2), Ling Ye (2), Ivan Rusyn (1,3), Ann Richard (4), Alexander Golbraikh (2) and Alexander Tropsha (1,2)
(1) Carolina Environmental Bioinformatics Center; (2) Division of Medicinal Chemistry and Natural Products; (3) Department of Environmental Sciences and Engineering, UNC-Chapel Hill, Chapel Hill, NC 27599; and (4) National Center for Computational Toxicology, EPA, RTP, NC 27711
Establishing robust correlations between in vitro and in vivo toxicity of environmental chemicals or drug candidates is critical to increase the efficiency of toxicity testing. In most studied cases, the correlation has been poor or non-existent. We have developed a novel two-step modeling approach to address this challenge. The approach is illustrated with the data from the German Center for the Documentation and Validation of Alternative Methods (ZEBET), that compiled a database including 347 chemicals with experimental in vitro cytotoxicity IC50 and rodent in vivo LD50 values. We found that this dataset can be subdivided into two subsets: first includes compounds with high correlation in LD50 and IC50 values (R2>0.9), and second is comprised of compounds with poor IC50/LD50 correlation (outlier set). The classification QSAR modeling was successful in discriminating the two classes of compounds using MolConnZ chemical descriptors with the classification accuracy as high as 72% for the external validation set. In addition, k nearest neighbor (kNN) QSAR modeling method was applied to the outlier set only and 16 models with R2/Q2>0.5/0.5 cutoff were produced. Overall, the two-step QSAR approach achieved the prediction accuracy of 72% for the external validation set (25 compounds). We conclude that considerable improvements can be achieved in correlating in vitro and in vivo toxicity data when the activity-correlation based pre-clustering is applied prior to development of predictive toxicity models.
Contact Information:
Program: Dr. Barry Hardy, eCheminfo Community of Practice & Research, Douglas Connect. Tel: +41 61 851 0170. barry.hardy -[at]- douglasconnect.com
Registration Enquiries: Nicki Douglas, Douglas Connect, Baermeggenweg 14, 4314 Zeiningen, Switzerland. Tel: +41 61 851 0461. eCheminfo -[at]- douglasconnect.com
or please visit:
http://echeminfobm810.eventsbot.com/
Barry Hardy
Community of Practice & Research Manager
InnovationWell eCheminfo QSAR Toxicology Predictive Toxicology bioinformatics cheminformatics chemoinformatics Computational Biology Computational Chemistry Translational Medicine Molecular Medicine biology pharmaceutical pharma meeting workshop training Critical Path toxicology Life Sciences Drug Discovery Research and Development Drug Development Healthcare Innovation Innovation Management events
Comments