improving toxicity predictions using data and knowledge
TRANSCRIPT
Improving Toxicity
Predictions using Data
and Knowledge Sharing
Dr Liz Covey-Crump
In Silico Prediction of Toxicity
• Toxicity Risk assessment of chemicals is required in
several areas:-
• Deliberate exposure – e.g. Drugs and Personal Care
• ‘Accidental’ human exposure – agrochemical,
manufacturing, packaging etc
• In Silico predictions can be used to:-
• Predict, support and explain experimental results
• In some cases in lieu of testing (e.g. ICH M7 – mutagenic
impurity assessment)
In Silico Prediction of Toxicity
• Advantages of an In Silico prediction?
• Cost/Time effective
• No need to synthesise compound
• Reproducible
• Reduces animal testing
• Can provide mechanistic information
• Can be used to help inform next steps as part of intelligent
testing strategy and/or defined approach
• Different In Silico methodologies
• Statistical
• Knowledge Base Focus for this presentation
Knowledge Based Expert Systems
Qualitative
Predictions
Transparent(What underlying data
has been used to make
the prediction)
Structure Activity Relationship (SAR)
methodology
Broad Range of endpoints
Alert 884: alpha-Dihalo or trihalo ketone or aldehyde
This alert describes the mutagenicity of alpha-dihalo or alpha-trihalo ketones and aldehydes.
Alpha-dihalo or alpha-trihalo ketones and aldehydes globally exhibit mutagenicity…….both in the preseence
and in the absence of metabolic activation……. A Lhasa Limited member donated alpha-dihalo ketone shows
positive Ames results in strains …………………..Alpha-dihalo or alpha-trihalo ketones and aldehydes are
electrophilic species that are capable of directly alkylating DNA. The electrophilicity of the carbon atom alpha
to the carbonyl is enhanced by both the carbonyl group and the halogen atoms.
The scope of this alert has been defined by the available Ames test data for alpha-dihalo or alpha-trihalo
ketones and aldehydes, including also a compound donated by a Lhasa Limited member…………………..
Why Share Data?
• Cover gaps in chemical space within in silico models• e.g. 25% of Derek Nexus alerts for mutagenicity have been
built using proprietary data
• Donation of proprietary data can fill the gaps to allow:-• Modelling of chemical space unique for an organisation
• Improve predictivity for chemical space of highest relevance
• Generalise models for mutual benefit
• Encouraging collaboration which benefits the scientific
community• Standardisation of methodologies
• Standardisation of approach and workflows
Highly relevant proprietary data
A representation of chemical coverage in the eTOX database when
compared to marketed drugs
Visualisation produced in StarDrop (Optibrium Ltd, Cambridge, UK)
eTOX 25/10/16 DrugBank Approved v5.0.3
Mutagenicity in Derek Nexus
Metrics (%) Results
Data
setSe Sp PP NP Acc TP FP TN FN Total
Public 83 75 79 79 79 2908 762 2247 595 6512
• 140 mutagenicity alerts
• 25% of alerts contain proprietary data
• Comprehensive coverage of endpoint• Aromatic amines and boronic acids are still of significant interest
and require refinement
• Derek Nexus performance against public data is very good
Member data set - Performance
• 1261 compounds
• Mainly negative results• Bias = 77% negative
• 114 FP
• 117 FN
Mutagenicity
Metrics (%) Results
Data
setSe Sp PP NP Acc TP FP TN FN Total
Public 83 75 79 79 79 2908 762 2247 595 6512
Member 59 88 60 88 82 168 114 862 117 1261
Member data set – New/Modified alert summary
• 5 new alerts
• Amine (x4)
• Boronic acid
• 4 modifications to existing alerts
• Azide, hydrazoic acid or azide salt
• Alkyl aldehyde
• Arylhydrazine
• Arylboronic acid or derivative
• 4 potential new alerts/alert modifications
• require more data/mechanistic support
Alkylhydrazines are tricky to predict…
• Alkylhydrazines tend to be at best weakly positive in the
Ames test
• Conflicting evidence as to strain activity for compounds
of this class
• Inconsistencies in published Ames test results…• weak activity
• high toxicity
• facile oxidation
• The mechanism of mutagenic activity has yet to be fully
elucidated• Different mechanistic pathways may furthermore contribute
to the differences in observed strain specificity between
alkylhydrazines
The problem with alert 28 (alkyl hydrazine)
• A pharmaceutical company found that alert 28 was over-
firing for their proprietary data.
• The member provided Lhasa with the relevant structures.
?
‘Alert 28 – alkyl hydrazine’ modification
Anonymised
structure
Ames positive
compounds
Ames negative
compounds
Mechanistic
rationalOld alert New alert
3 4 Hydrolysable Positive Positive
0 3
Sufficient
hydrolysis not
expected
Positive Negative
Skin Sensitisation Data Sharing
7 new alerts
5 modifications to
existing alerts
Public data (n = 2611) BMS (n = 467)
History of Toxicity Data/Knowledge Sharing at Lhasa
Time
2018
Knowledge Shared
Structures + Toxicity Data
Preclinical
ClinicalDegree of Sharing/Value of Data
Structures +Ames Data
Ongoing Initiatives
• 12 Japanese Pharmaceutical companies have shared
mutagenicity data – analysis is underway
• Japanese NIHS (Regulatory) Collaboration continues
(14th year)
• Several other data sharing initiatives ongoing at Lhasa
Data and Knowledge Sharing
• Data sharing groups benefits:-
• Data can be used to prevent repeat testing
• Data can be used to make decisions/help prioritise
• Verification of the results of a model
• Improvement of models
• Set common quality standards
• Standardisation of workflows etc
• Changing regulatory guidance has led to the need to build
knowledge of EI levels in excipients
http://www.ich.org/products/guidelines/quality/article/quality-
guidelines.html
Elemental Impurities Data Sharing Initiative (EI)
Elemental Impurities Data Sharing Initiative (EI)
• Q3D :-
• Section 5 - Information for this risk assessment includes but is
not limited to: data generated by the applicant, information
supplied by drug substance and/or excipient manufacturers
and/or data available in published literature.
• Section 5.5 - The data that support this risk assessment can
come from a number of sources that include, but are not limited
to:
• Prior knowledge;
• Published literature;
• Data generated from similar processes;
• Supplier information or data;
• Testing of the components of the drug product;
• Testing of the drug product.
A proactive action from the pharma industry regarding the
compliance with the regulatory guidelines ICH Q3D for elemental
impurities.
Elemental Impurities Data Sharing Initiative
Facilitate more scientifically driven elemental impurities risk
assessments under ICH Q3D, and reduce unnecessary
testing as part of the elemental impurities risk assessment
efforts.
Elemental Impurities Data Sharing Initiative
• The data to be shared is the analytical data generated to
establish the levels of trace metals within batches of excipients
used in the manufacture of pharmaceuticals
• Aims to save time and provide further evidence/robustness
when partners are setting the limits for the elemental impurities
in the API
• Lhasa acts as the ‘honest broker’ and hosts the data within a
custom version of Vitic and facilitates the data sharing group
How will the EI database be used?
• The database will provide a better assessment of which
materials represent a more significant risk than others
• Indicate where the risk is real and where it is negligible
• Reduce the amount of testing that is needed
• The EI database can be used as key supportive information in
conjunction with some product specific test data for a risk
assessment
• Various case studies have been presented in a Lhasa vICGMhttps://www.lhasalimited.org/publications/?custom_in_RelatedEvent=
4032&orderby=Name
Publication on the database pending
• What the Consortium do?
Discussing and agreeing upon the scientific direction of the
project
Contributing expertise and knowledge
Monitoring the data provided by the member organisations
and ensuring it meets predefined quality standards
Recommending priorities for work on the project
Elemental Impurities Data Sharing Consortium