- Email: [email protected]
An improved workflow to perform in silico mutagenicity assessment of impurities as per ICH M7 guideline
S164
Abstracts / Toxicology Letters 229S (2014) S40–S252
public domain (PubMed articles and FDA reports), describing cellular processes activated by these compounds. Moreover, we grouped the compounds based on their chemical structure in addition to their MOA, by applying newly developed bioinformatics analysis tools for assessing chemical structure similarity. Finally, for other structurally similar skin toxicant and/or know biological downstream partners we were able to predict skin toxicity of these compounds and suggest possible MOA. http://dx.doi.org/10.1016/j.toxlet.2014.06.562 P-3.53 An improved workflow to perform in silico mutagenicity assessment of impurities as per ICH M7 guideline Roustem Saiakhov ∗ , Suman Chakravarti, Aleks Sedykh MultiCASE Inc., Beachwood, OH, USA Purpose: Use of in silico tools is one of the central points of ICH M7 guideline. However computational assessment of DNA reactive impurities in real life is still challenging. Purpose of this study is to develop and apply an effective workflow using a QSAR statistical system and a quantitative read across modeling methodology to obtain reliable genotoxicity assessments adhering to the ICH M7 recommendations. The novel multi-tier methodology provides the detailed information of toxicity alerts and reduces false positives, false negatives while increasing the test coverage. This in silico approach can be successfully used for assessment of mutagenicity of new drug candidates, impurities and metabolites. Method and data: CASE Ultra is a QSAR statistics based computer program that automatically extracts structure-activity knowledge from chemical databases and applies the knowledge for predicting activity in test chemicals. Several models, developed by MultiCASE Inc. alone and within Research Cooperation Agreement with Center for Drug Evaluation and Research of FDA were utilized, as well as a read across modeling technique for quantitative prediction of toxicity that uses the neighborhood profiles of chemicals. A multi-tier approach in combining several models and read across engine for bacterial mutagenicity evaluation resulted in significant improvements in the sensitivity, specificity and coverage. Results of the study: The improvements in the predictive performance as a result of the effective workflow is demonstrated for various scenarios of mutagenic impurity assessments, using an external validation set and real life case studies as examples. http://dx.doi.org/10.1016/j.toxlet.2014.06.563 P-3.54 Computational prediction of off-target related safety liabilities of molecules: Cardiotoxicity, hepatotoxicity and reproductive toxicity Friedemann Schmidt 1,∗ , Alexander Amberg 1 , Denis Mulliner 1 , Manuela Stolte 1 , Hans Matter 2 , Gerhard Hessler 2 , Axel Dietrich 2 , Nikita Remez 3 , David Vidal 3 , Jordi Mestres 3 , Andreas Czich 1 1
Preclinical Safety, Sanofi, Frankfurt, Germany, 2 Structure, Design and Informatics, Frankfurt, Germany, 3 Chemotargets, Barcelona, Spain Off-target liabilities of molecules can lead to drug toxicity. Clinically relevant examples include, e.g. inhibition of the hERG potassium channel, or activation of the serotonin receptor 5HT2B,
which can both lead to cardiosafety risks, the latter also to impaired cardiac development. While in vitro profiling is often limited to few targets and compounds due to experimental constraints, computational approaches are aiming to overcome these limitations. We have developed a tiered strategy for in silico off-target profiling.
(1) Off-target prediction employs the PredictFX method originating from Chemotargets. Prospective validation for our in vitro assay data revealed a predictive accuracy of 81%. The model applicability domain could be substantially enhanced by retraining with Sanofi in vitro data. (2) Quantitative structure-activity models (QSAR) were built for a subset of >400 important off-targets (200 Kinases, >200 GPCRs, ion channels, transporters and enzymes) in alignment to existing screening panels at Sanofi. These models were extensively validated and the best have good r2 /r2 (cv)/q2 ≥ 0.6. (3) Toxicity can be inferred by interaction with key protein targets implicated in ‘adverse outcome pathways’ (AOPs). To this end, both open and commercial pathway analysis databases were implemented, that relate off-target interaction at a given compound potency threshold to a toxicity phenotype.
For further characterization of lead series and support of discovery projects, advice can be provided on the priorization of molecules and the selection of samples for experimental validation. This is illustrated by case studies. The main analysis scope is on organ toxicities (e.g. cardiotoxicity, hepatotoxicity and nephrotoxicity) and reproductive toxicities. http://dx.doi.org/10.1016/j.toxlet.2014.06.564 P-3.55 A text-mining approach for chemical risk assessment and cancer research Ilona Silins 1,2,∗ , Anna Korhonen 2 , Yufan Guo 3 , Ulla Stenius 1 1 Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden, 2 Computer Laboratory, University of Cambridge, Cambridge, UK, 3 Computer Science and Engineering, University of Washington, Seattle, USA
The identification and understanding of chemicals’ “mode of action” (MOA) can both improve cancer risk assessment and reduce uncertainties. The term MOA is defined as a sequence of key events, starting with the interaction of an agent with a cell, proceeding through cellular changes ultimately resulting in cancer formation. We are developing a computerized text-mining tool, CRAB, for cancer research and risk assessment. The tool is based on the current understanding of the MOA and mechanisms relevant for cancer development. It can automatically analyze scientific data on chemicals of interest and classifies the literature according to the type, amount and strength of the evidence it provides for risk assessment. Chemical-specific toxicological literature profiles are generated by the tool. Several areas where cancer risk assessment and research could be further developed with the aid of text-mining have been recognized, e.g. identification of gender-specific mechanisms for carcinogens and defining critical signaling pathways for cancer development. Another important application is to identify chemicals that share the same MOA that can cause additivity or interactions in the context of mixed exposure. The tool provides both a qualitative and quantitative overview of existing scientific data. In addition, it can help to find patterns
Abstracts / Toxicology Letters 229S (2014) S40–S252
public domain (PubMed articles and FDA reports), describing cellular processes activated by these compounds. Moreover, we grouped the compounds based on their chemical structure in addition to their MOA, by applying newly developed bioinformatics analysis tools for assessing chemical structure similarity. Finally, for other structurally similar skin toxicant and/or know biological downstream partners we were able to predict skin toxicity of these compounds and suggest possible MOA. http://dx.doi.org/10.1016/j.toxlet.2014.06.562 P-3.53 An improved workflow to perform in silico mutagenicity assessment of impurities as per ICH M7 guideline Roustem Saiakhov ∗ , Suman Chakravarti, Aleks Sedykh MultiCASE Inc., Beachwood, OH, USA Purpose: Use of in silico tools is one of the central points of ICH M7 guideline. However computational assessment of DNA reactive impurities in real life is still challenging. Purpose of this study is to develop and apply an effective workflow using a QSAR statistical system and a quantitative read across modeling methodology to obtain reliable genotoxicity assessments adhering to the ICH M7 recommendations. The novel multi-tier methodology provides the detailed information of toxicity alerts and reduces false positives, false negatives while increasing the test coverage. This in silico approach can be successfully used for assessment of mutagenicity of new drug candidates, impurities and metabolites. Method and data: CASE Ultra is a QSAR statistics based computer program that automatically extracts structure-activity knowledge from chemical databases and applies the knowledge for predicting activity in test chemicals. Several models, developed by MultiCASE Inc. alone and within Research Cooperation Agreement with Center for Drug Evaluation and Research of FDA were utilized, as well as a read across modeling technique for quantitative prediction of toxicity that uses the neighborhood profiles of chemicals. A multi-tier approach in combining several models and read across engine for bacterial mutagenicity evaluation resulted in significant improvements in the sensitivity, specificity and coverage. Results of the study: The improvements in the predictive performance as a result of the effective workflow is demonstrated for various scenarios of mutagenic impurity assessments, using an external validation set and real life case studies as examples. http://dx.doi.org/10.1016/j.toxlet.2014.06.563 P-3.54 Computational prediction of off-target related safety liabilities of molecules: Cardiotoxicity, hepatotoxicity and reproductive toxicity Friedemann Schmidt 1,∗ , Alexander Amberg 1 , Denis Mulliner 1 , Manuela Stolte 1 , Hans Matter 2 , Gerhard Hessler 2 , Axel Dietrich 2 , Nikita Remez 3 , David Vidal 3 , Jordi Mestres 3 , Andreas Czich 1 1
Preclinical Safety, Sanofi, Frankfurt, Germany, 2 Structure, Design and Informatics, Frankfurt, Germany, 3 Chemotargets, Barcelona, Spain Off-target liabilities of molecules can lead to drug toxicity. Clinically relevant examples include, e.g. inhibition of the hERG potassium channel, or activation of the serotonin receptor 5HT2B,
which can both lead to cardiosafety risks, the latter also to impaired cardiac development. While in vitro profiling is often limited to few targets and compounds due to experimental constraints, computational approaches are aiming to overcome these limitations. We have developed a tiered strategy for in silico off-target profiling.
(1) Off-target prediction employs the PredictFX method originating from Chemotargets. Prospective validation for our in vitro assay data revealed a predictive accuracy of 81%. The model applicability domain could be substantially enhanced by retraining with Sanofi in vitro data. (2) Quantitative structure-activity models (QSAR) were built for a subset of >400 important off-targets (200 Kinases, >200 GPCRs, ion channels, transporters and enzymes) in alignment to existing screening panels at Sanofi. These models were extensively validated and the best have good r2 /r2 (cv)/q2 ≥ 0.6. (3) Toxicity can be inferred by interaction with key protein targets implicated in ‘adverse outcome pathways’ (AOPs). To this end, both open and commercial pathway analysis databases were implemented, that relate off-target interaction at a given compound potency threshold to a toxicity phenotype.
For further characterization of lead series and support of discovery projects, advice can be provided on the priorization of molecules and the selection of samples for experimental validation. This is illustrated by case studies. The main analysis scope is on organ toxicities (e.g. cardiotoxicity, hepatotoxicity and nephrotoxicity) and reproductive toxicities. http://dx.doi.org/10.1016/j.toxlet.2014.06.564 P-3.55 A text-mining approach for chemical risk assessment and cancer research Ilona Silins 1,2,∗ , Anna Korhonen 2 , Yufan Guo 3 , Ulla Stenius 1 1 Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden, 2 Computer Laboratory, University of Cambridge, Cambridge, UK, 3 Computer Science and Engineering, University of Washington, Seattle, USA
The identification and understanding of chemicals’ “mode of action” (MOA) can both improve cancer risk assessment and reduce uncertainties. The term MOA is defined as a sequence of key events, starting with the interaction of an agent with a cell, proceeding through cellular changes ultimately resulting in cancer formation. We are developing a computerized text-mining tool, CRAB, for cancer research and risk assessment. The tool is based on the current understanding of the MOA and mechanisms relevant for cancer development. It can automatically analyze scientific data on chemicals of interest and classifies the literature according to the type, amount and strength of the evidence it provides for risk assessment. Chemical-specific toxicological literature profiles are generated by the tool. Several areas where cancer risk assessment and research could be further developed with the aid of text-mining have been recognized, e.g. identification of gender-specific mechanisms for carcinogens and defining critical signaling pathways for cancer development. Another important application is to identify chemicals that share the same MOA that can cause additivity or interactions in the context of mixed exposure. The tool provides both a qualitative and quantitative overview of existing scientific data. In addition, it can help to find patterns