A knowledge-based approach to the in silico assessment of toxicity Carol Marchant carol.marchant@lhasalimited.org
About Lhasa Limited • Not-for-profit company and educational charity • Interests in knowledge and data sharing in chemistry and the life sciences Booth 70
Some Lhasa Limited Products and Projects Derek for Windows & Derek Nexus Expert systems for the assessment of toxicity Meteor An expert system for the assessment of xenobiotic metabolism Vitic & Vitic Nexus Structure-searchable toxicity databases Zeneth An expert system for the assessment of chemical degradation pathways
Scope of Presentation • In silico assessment of toxicity from chemical structure • Single component, organic chemicals of low to medium molecular weight • Human health-related
Outline • Overview of expert systems in the in silico assessment of toxicity • Approaches to overcoming data availability issues
I n Silico Toxicity Assessment Statistical model Expert system Molecular model System which makes use of rules compiled by human experts and stored in a knowledge base Derek HazardExpert OncoLogic
Derek Knowledge Base Structure Comments Rules References Typical rules describe: Presence of structural alerts Physicochemical property dependencies Known toxicity data Endpoint extrapolations Comments Species-specific effects Validation comments Structural alerts Examples References Evidence might include consideration of: chemistry, mechanism of action, metabolism, pharmacology, physicochemical properties, toxicology etc
Derek Reasoning Process • Each rule that is applied to a query chemical provides a qualitative indication of the likelihood of toxicity for a particular toxicological endpoint • A typical rule might take the form: If [Mutagenicity alert] is [certain] then [Mutagenicity] is [plausible] • Arguments for and against toxicity from such rules are weighed up against each other to arrive at an overall assessment Judson et al, Journal of Chemical Information and Computer Sciences 43 1364-1370 2003
Advantages of an Expert System Approach • Human experts are able to work with diverse data types, also allowing for inconsistent or incomplete data � Use can be made of all available information • Assessments are meaningful and transparent � Suitable for both prediction and interpretation • Knowledge can be expanded and updated without being completely rebuilt � Readily adapted to incorporate proprietary knowledge in-house
Sharing of Data and Knowledge Share knowledge Share data Vitic Derek Generate knowledge Enter data Enter knowledge Vitic editor Derek editor
Major Endpoint Categories in Derek • Carcinogenicity • Chromosome damage • Genotoxicity • Hepatotoxicity • HERG channel inhibition • Irritation • Mutagenicity • Ocular toxicity • Reproductive toxicity • Respiratory sensitisation • Skin sensitisation • Thyroid toxicity
I mpact of High Throughput Screening at Hoffmann-La Roche 60 50 Introduction of in silico testing 40 Percentage Positive Ames GLP 30 Positive Ames screen 20 10 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Year Müller et al , Computational Toxicology (ed. S Ekins) pp 545-579 2007
I mpurity Assessment Starting materials Starting materials Process through Intermediates Intermediates APIs Derek APIs + SciFinder, TOXNET & database search + Synthetic route Categorisation & Categorisation & management review management 75% of chemicals can be assessed on the basis of structure alone Dobo et al, Regulatory Toxicology and Pharmacology 44 282-293 2006
EMEA Guidance on I mpurities European Medicines Agency (2006) Guideline on the Limits of Genotoxic Impurities http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2009/09/WC500002903.pdf European Medicines Agency (2009) Q & A on the CHMP Guideline on the Limits of Genotoxic Impurities http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2009/09/WC500002907.pdf
DRAFT FDA Guidance on I mpurities US Food and Drug Administration (2008) DRAFT Guidance for Industry, Genotoxic and Carcinogenic Impurities in Drug Substances and Products: Recommended Approaches http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm079235.pdf
Continuous I mprovement Application Consensus modelling Integrated testing strategies Performance Availability of data Structure of knowledge Presentation OECD Principles for (Q)SAR validation
Data Availability • Mutagenicity is a relatively easy endpoint to derive knowledge for, at least in part because of the extent of available Ames test data • For other endpoints, knowledge development can become more problematic as complexity increases and data availability becomes an issue � Data may not exist � Data may be in an inaccessible format � Data may be proprietary
Strategies for I mproving Data Availability • Data may not exist � Custom testing � Precursory endpoints • Data may not be in an inaccessible format • Data may be proprietary
Chromosome Damage of Aromatic Thioamides • Ethionamide induces in S NH 2 vitro chromosomal aberrations in the absence of S9 mix N • Thioacetamide does not S NH 2 ⇒ Can we create a structural alert which associates all aromatic thioamides with the induction of in vitro chromosomal aberrations? Lhasa Limited collaboration with the National Institute of Health Sciences
Custom Testing of Aromatic Thioamides Aberrations -S9 + S9 S NH 2 Thiobenzamide + + S NH 2 Chlorthiamid Cl Cl - - H S N N-Phenyl w+ w+ thiobenzamide
I mplementation of Structural Alert for Chromosome Damage of Aromatic Thioamides • Structural alert scope based on available test data S R2 R1 N H R1 = C (aromatic) R2 = H, C # (sp3), C (aromatic) C # cannot be attached to further heteroatoms Ortho substituted aromatic rings are excluded • Postulated mechanism suggested to involve formation of electrophilic S-oxide metabolites
Precursory Endpoints • If insufficient data exist to develop knowledge for a particular endpoint, it may be possible to develop knowledge for a precursory effect • Mitochondrial dysfunction, for example, may be a precursor to hepatotoxic or other adverse events Structural alert describing Mitochondrial data mitochondrial dysfunction of bisguanides NH NH NH NH H 2 N N N H * # C H 2 N N N NH NH H H 2 N N N N # - no heteroatoms allowed H H C * - cannot be double, triple or aromatically bonded NH NH H 2 N N N H H Lhasa Limited collaboration with Pfizer; Fisk et al. Poster 206-005
Strategies for I mproving Data Availability • Data may not exist • Data may not be in an inaccessible format � Semi-automated data extraction • Data may be proprietary
Extracting Data from Legacy Reports Manual extraction Semi-automated extraction Optical character Read recognition document Schema Text I dentify Vocabularies facts mining Queries Check Enter data data Transfer Check data data Lhasa Limited collaboration with Linguamatics and Pfizer
Semi-automated Extraction of Repeat Dose Study Reports • Toxicity data can be captured using a semi- automated extraction process with consistency and standardisation • Extracted data has high accuracy and acceptable recall • Time taken is similar to manual extraction but can be expected to improve over time
Strategies for I mproving Data Availability • Data may not exist • Data may not be in an inaccessible format • Data may be proprietary � Derivation of non-proprietary knowledge � Data sharing
Structural Alerts from Proprietary Repeat Dose Data Proprietary data set (731 compounds) Clustering Expert analysis 34 structural alerts for hepatotoxicity 3-Furoic acid and derivatives Expert review structural alert O R1 R2 * 32 non-proprietary * O R3 structural alerts R4 for hepatotoxicity R1-R3 = H, C - R4 = O , OH, OC Bonds marked * can be single or aromatic Lhasa Limited collaboration with the National Institute of Health Sciences
Proprietary Data Sharing Model Organisation 1 Organisation 2 Organisation 3 Organisations 1-6 Organisation 4 Organisation 5 Organisation 6 Data Database Data
Vitic I ntermediates Project • Aims to provide data for: � Refinement and validation of in silico models � Assessment of impurities for genotoxicity • Involves sharing of proprietary Ames test data, particularly for synthetic intermediates • Consists of a consortium of 7 pharmaceutical companies • Shared database of more than 500 chemicals to date
Application of Vitic I ntermediates Data • Data has been used to: 120 100 Positive predictivity % 80 � Assess the performance of 60 existing structural alerts 40 20 0 2 7 12 15 16 19 27 28 33 39 49 69 303 304 305 315 318 326 329 344 349 351 352 353 354 471 645 Alert number R1 � Suggest new structural O O alerts R3 R2 H O R1-R3 = H, C (any) 4-Oxy-3(2H)-furanone structural alert
Recommend
More recommend
