cobas 8100 automated workflow series 3-D intelligence in ...
An Automated Classification Workflow to Assist Expert ...
1
An Automated Classification Workflow to Assist Expert Assessment and Regulatory Submission for ICH M7 William E. Nye, Andrew L. Shannon-Little, David J. Yeo, Laura A. Johnston, Alexander P. Harding, Richard V. Williams. Granary Wharf House, 2 Canal Wharf, Leeds, LS11 5PS, [email protected] Abstract Introduction Inconclusive predictions References Conclusion The ICH M7 Classification tool within Lhasa Limited’s Nexus platform provides a complete solution for the review of impurities under the ICH M7 guideline, which, where possible, provides a proposed ICH M7 Class for each impurity for expert review. Although the vast majority of impurities can be easily classified as Class 1, 2, 3, 4, or 5 using the ICH M7 guideline, this poster demonstrates that the outcome for some compounds is not certain. Such cases, as discussed, include known genotoxic carcinogens that are negative in the Ames assay, or have negative in silico results. A cold reading of such impurities could lead to a Class 5 call on account of the compounds’ negative Ames data/in silico results. By contrast, some experts may propose the indication of increased risk of DNA reactivity is sufficient to treat such compounds as Class 1; a known mutagenic carcinogen. This situation is therefore open to interpretation, and Lhasa Limited would be keen to understand other people’s opinion as to how such situations should be dealt with, and would be interested to hear from you at accountmanagement@lhasalimited .org. • This information is presented in an interactive results table, simplifying the expert review process • Experimental Ames and carcinogenicity data can be added to the workflow to influence the ICH M7 prediction • Results from an additional in silico model can also be included within the results table • Users can explore and review predictions • A report is generated that displays all relevant information including any edits or comments added during expert review • A submission ready ‘Classification Card’ is included within the report. ICH M7 Workflow Tool • The knowledge base is composed of publicly available data as well as proprietary data donated by Lhasa’s members. • 25% of mutagenicity alerts contain proprietary data. • Derek Nexus v 5.0.2 comprises 115 alerts for mutagenicity in vitro. 2 • The achievements of Derek Nexus recently won Lhasa a Queen’s Award for Enterprise: Innovation. The ICH M7 guideline is now in force in the United States, Europe and Japan. To fulfil the requirements for the use of in silico tools, two complementary systems must be employed: an expert rule-based system and a statistical system. The outputs of the two systems should then be reviewed using expert judgement and a reasonable, supported conclusion presented. Derek Nexus fulfils the criteria for the expert rule-based in silico system and Sarah Nexus fulfils the statistical system requirement of the ICH M7 guideline. A classification workflow tool has been developed by Lhasa Limited to aid in the submission of impurities under the ICH M7 guideline. The workflow uses two in silico tools, Derek and Sarah Nexus, alongside experimental Ames and carcinogenicity data to predict an ICH M7 class. This workflow tool presents the classifications in a submission ready format, while still allowing the user to supplement the workflow with their own data. The workflow currently provides an ‘Inconclusive’ classification for chemicals that are not well defined by the guideline. This poster describes how impurities are classified using the workflow, in order to facilitate discussion regarding whether those compounds with a predicted class of ‘inconclusive’ are being treated appropriately. “The focus of this guideline is on DNA reactive substances that have a potential to directly cause DNA damage when present at low levels leading to mutations and therefore, potentially causing cancer. This type of mutagenic carcinogen is usually detected in a bacterial reverse mutation (mutagenicity) assay.” Ames UNKNOWN CPDB UNKNOWN Derek INACTIVE Sarah POSITIVE Ames UNKNOWN CPDB UNKNOWN Derek INACTIVE Sarah NEGATIVE Ames UNKNOWN CPDB UNKNOWN Derek INACTIVE Sarah EQUIVOCAL Ames UNKNOWN CPDB UNKNOWN Derek PLAUSIBLE* Sarah OoD *Alert also in the API Ames UNKNOWN CPDB UNKNOWN Derek PLAUSIBLE* Sarah POSITIVE *Alert not in the API Figure 1 : How Derek Nexus predicts Figure 2 : How Sarah Nexus predicts Figure 3 : Example of Results Table generated by Nexus Figure 6 : 2-methylimidazole, with its associated Ames, carcinogenicity data and ICH M7 class Figure 8 : methyl eugenol, with its associated Ames, carcinogenicity data and ICH M7 class Figure 7 : ICH M7 guideline Figure 4 : A workflow to demonstrate the logic behind the ICH M7 classification tool, to demonstrate what information is considered to make predictions. ‘Other’ in Sarah Nexus is a result of Equivocal or Out of Domain. Figure 5: Examples of impurities that can be classified by the workflow tool In order to aid the expert review of in silico predictions for impurities and their submission under the ICH M7 guideline, Lhasa Limited has integrated an ICH M7 Classification tool within the Nexus platform. The workflow considers Derek and Sarah predictions, alongside experimental Ames and carcinogenicity data to arrive at a predicted ICH M7 Class. The source for the experimental Ames data is the ‘Lhasa Summary Call’ table within Lhasa Limited’s database management tool, Vitic Nexus. The source of the experimental carcinogenicity data is the Carcinogenicity Potency Database (CPDB). 4 How likely is it? What endpoint could occur? Rule base Knowledge base Matches toxicophore to Alerts Alert and Likelihood Significance and consistency of supporting examples of hypotheses? Prediction derived from relevant hypotheses Calculates confidence Sarah Model Generate hypotheses from training set Overall Prediction • The Sarah training set comprises 9,507 compounds in the public domain. • Hypotheses are structural moieties that suggest activity in the Ames assay and are supported by the most similar examples from the training set. • The predictions are interpretable and provide easy access to the molecules from the training set that were used to form the prediction. 3 1. ICH HARMONISED TRIPARTITE GUIDELINE, http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Multidisciplinary/M7/M7_Step_4.pdf . 2. R. V. Williams, A. Amberg, A. Brigo, L. Coquin, A. Giddings, S. Glowienke, N. Greene, R. Jolly, R. Kemper, C. O’Leary-Steele, A. Parenty, HP. Spirkl, S. A. Stalford, S. K. Weiner, J. Wichard, Regul. Toxicol. Pharm., 2016, 76, 79-86. 3. C. Barber, A. Amberg, L. Custer, K. L. Dobo, S. Glowienke, J. Van Gompel, S. Gutsell, J. Harvey, M. Honma, M. O. Kenyon, N. Kruhlak, W. Muster, L. Stavitskaya, A. Teasdale, J. Vessey, J. Wichard, Toxicol. Pharm., 2015, 73, 367-377. 4. The Carcinogenicty Potency Project, https://toxnet.nlm.nih.gov/cpdb/ . 5. http://www.ncbi.nlm.nih.gov/pubmed/23074021 Ames ACTIVE CPDB UNKNOWN Derek POSITIVE Sarah POSITIVE Ames ACTIVE CPDB ACTIVE Derek POSITIVE Sarah POSITIVE API: Ames Negative Derek alert fires (alkyl nitro compound) Ames NEGATIVE CPDB POSITIVE Derek INACTIVE Sarah NEGATIVE Ames NEGATIVE CPDB POSITIVE Derek INACTIVE Sarah NEGATIVE ICH M7 class = Inconclusive ICH M7 class = Inconclusive • The ICH M7 guideline are clear on how to classify non-carcinogens irrespective of their mutagenicity (class 5) but are less clear on how to interpret Ames carcinogens. In these circumstances, the interpretation of the guideline can potentially lead to differing classifications. • The ICH M7 classification tool will assign compounds with negative Ames results with positive carcinogenicity data as ‘Inconclusive’. While some would consider this situation as class 5, the justification behind this inconclusive prediction is that further expert review may be required in order to assign the classification accurately. • Contrasting examples that fall within the described classification are 2-methylimidazole and methyl eugenol, which can be seen in figures 6 and 8. They first appear to be compounds that would be classified equally and the workflow tool has assigned them both as ‘inconclusive’. However, further analysis of methyl eugenol reveals that it is positive in several other genotoxic in vitro assays, such as the unscheduled DNA synthesis assay, as well as an in vivo mutagenicity assay 5 and it could therefore be considered inaccurate to classify this compound as class 5. All the genotoxic in vitro data for 2-methylimidazole in Vitic Nexus is negative, so a prediction of class 5 is accurate. • These challenging examples demonstrate how the ICH M7workflow tool provides relevant information and guides the user to focus on compounds that demand further expert review. Lhasa would be interested to hear your feedback on how you may classify these compounds and if you think the workflow is correct in its prediction. Feedback should be emailed to accountmanagement@lhasalimited .org. Information Considered in the ICH M7 Workflow
Transcript of An Automated Classification Workflow to Assist Expert ...
An Automated Classification Workflow to Assist Expert Assessment and Regulatory Submission for ICH M7
William E. Nye, Andrew L. Shannon-Little, David J. Yeo, Laura A. Johnston, Alexander P. Harding, Richard V. Williams.
Granary Wharf House, 2 Canal Wharf, Leeds, LS11 5PS, [email protected]
Abstract
Introduction
Inconclusive predictions
References
ConclusionThe ICH M7 Classification tool within Lhasa Limited’s Nexus platform provides a complete solution for the review of impurities under the ICH M7 guideline, which, where possible, provides a proposed ICH M7 Class for each impurity for expertreview. Although the vast majority of impurities can be easily classified as Class 1, 2, 3, 4, or 5 using the ICH M7 guideline, this poster demonstrates that the outcome for some compounds is not certain. Such cases, as discussed, include knowngenotoxic carcinogens that are negative in the Ames assay, or have negative in silico results. A cold reading of such impurities could lead to a Class 5 call on account of the compounds’ negative Ames data/in silico results. By contrast, someexperts may propose the indication of increased risk of DNA reactivity is sufficient to treat such compounds as Class 1; a known mutagenic carcinogen. This situation is therefore open to interpretation, and Lhasa Limited would be keen tounderstand other people’s opinion as to how such situations should be dealt with, and would be interested to hear from you at [email protected].
• This information is presented in an interactive results table, simplifying the expert review process• Experimental Ames and carcinogenicity data can be added to the workflow to influence the ICH M7 prediction• Results from an additional in silico model can also be included within the results table• Users can explore and review predictions• A report is generated that displays all relevant information including any edits or comments added during expert
review• A submission ready ‘Classification Card’ is included within the report.
ICH M7 Workflow Tool
• The knowledge base is composed ofpublicly available data as well asproprietary data donated by Lhasa’smembers.
• 25% of mutagenicity alerts containproprietary data.
• Derek Nexus v 5.0.2 comprises 115alerts for mutagenicity in vitro.2
• The achievements of Derek Nexusrecently won Lhasa a Queen’s Awardfor Enterprise: Innovation.
The ICH M7 guideline is now in force in the United States, Europe and Japan. To fulfil the requirements for the use of insilico tools, two complementary systems must be employed: an expert rule-based system and a statistical system. Theoutputs of the two systems should then be reviewed using expert judgement and a reasonable, supported conclusionpresented. Derek Nexus fulfils the criteria for the expert rule-based in silico system and Sarah Nexus fulfils the statisticalsystem requirement of the ICH M7 guideline.
A classification workflow tool has been developed by Lhasa Limited to aid in the submission of impurities under the ICH M7 guideline. The workflow uses two in silico tools, Derek and Sarah Nexus, alongside experimental Ames andcarcinogenicity data to predict an ICH M7 class. This workflow tool presents the classifications in a submission ready format, while still allowing the user to supplement the workflow with their own data. The workflow currently provides an‘Inconclusive’ classification for chemicals that are not well defined by the guideline. This poster describes how impurities are classified using the workflow, in order to facilitate discussion regarding whether those compounds with a predictedclass of ‘inconclusive’ are being treated appropriately.
“The focus of this guideline is onDNA reactive substances thathave a potential to directly causeDNA damage when present at lowlevels leading to mutations andtherefore, potentially causingcancer. This type of mutageniccarcinogen is usually detected in abacterial reverse mutation(mutagenicity) assay.”
Ames UNKNOWN
CPDB UNKNOWN
Derek INACTIVE
Sarah POSITIVE
Ames UNKNOWN
CPDB UNKNOWN
Derek INACTIVE
Sarah NEGATIVE
Ames UNKNOWN
CPDB UNKNOWN
Derek INACTIVE
Sarah EQUIVOCAL
Ames UNKNOWN
CPDB UNKNOWN
Derek PLAUSIBLE*
Sarah OoD
*Alert also in the API
Ames UNKNOWN
CPDB UNKNOWN
Derek PLAUSIBLE*
Sarah POSITIVE
*Alert not in the API
Figure 1: How Derek Nexus predicts
Figure 2: How Sarah Nexus predicts
Figure 3: Example of Results Table generated by Nexus
Figure 6: 2-methylimidazole, with its associated Ames, carcinogenicity data and ICH M7 class Figure 8: methyl eugenol, with its associated Ames, carcinogenicity data and ICH M7 classFigure 7: ICH M7 guideline
Figure 4: A workflow to demonstrate the logic behind the ICH M7 classification tool, to demonstrate what information is considered to make predictions. ‘Other’ in Sarah Nexus is a result of Equivocal or Out of Domain.
Figure 5: Examples of impurities that can be classified by the workflow tool
In order to aid the expert review of in silico predictions for impurities and their submission under the ICH M7 guideline, Lhasa Limited has integrated an ICH M7 Classification tool within the Nexus platform. The workflow considers Derek and Sarah predictions, alongside experimental Ames and carcinogenicity data to arrive at a predicted ICH M7 Class. The source for the experimental Ames data is the ‘Lhasa Summary Call’ table within Lhasa Limited’s database management tool, Vitic Nexus. The source of the experimental carcinogenicity data is the Carcinogenicity Potency Database (CPDB).4
How likely is it?
What endpoint could
occur?
Rule base
Knowledge base
Matches toxicophore
to Alerts
Alert and
Likelihood
Significance and consistency of
supporting examples of
hypotheses?
Prediction derived from
relevant hypotheses
Calculates confidence
Sarah Model
Generate hypotheses
from training set
Overall
Prediction
• The Sarah training set comprises9,507 compounds in the publicdomain.
• Hypotheses are structural moietiesthat suggest activity in the Amesassay and are supported by the mostsimilar examples from the trainingset.
• The predictions are interpretable andprovide easy access to the moleculesfrom the training set that were usedto form the prediction.3
1. ICH HARMONISED TRIPARTITE GUIDELINE, http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Multidisciplinary/M7/M7_Step_4.pdf . 2. R. V. Williams, A. Amberg, A. Brigo, L. Coquin, A. Giddings, S. Glowienke, N. Greene, R. Jolly, R. Kemper, C. O’Leary-Steele, A. Parenty, HP. Spirkl, S. A. Stalford, S. K. Weiner, J. Wichard, Regul. Toxicol. Pharm., 2016, 76, 79-86. 3. C. Barber, A. Amberg, L. Custer, K. L. Dobo, S. Glowienke, J. Van Gompel, S. Gutsell, J. Harvey, M. Honma, M. O. Kenyon, N. Kruhlak, W. Muster, L. Stavitskaya, A. Teasdale, J. Vessey, J. Wichard, Toxicol. Pharm., 2015, 73, 367-377. 4. The Carcinogenicty Potency Project, https://toxnet.nlm.nih.gov/cpdb/ . 5. http://www.ncbi.nlm.nih.gov/pubmed/23074021
Ames ACTIVE
CPDB UNKNOWN
Derek POSITIVE
Sarah POSITIVE
Ames ACTIVE
CPDB ACTIVE
Derek POSITIVE
Sarah POSITIVE
API: Ames NegativeDerek alert fires (alkyl nitro compound)
Ames NEGATIVE
CPDB POSITIVE
Derek INACTIVE
Sarah NEGATIVE
Ames NEGATIVE
CPDB POSITIVE
Derek INACTIVE
Sarah NEGATIVE
ICH M7 class = Inconclusive ICH M7 class = Inconclusive
• The ICH M7 guideline are clear on how to classify non-carcinogens irrespective of their mutagenicity (class 5) but are less clear on how to interpret Ames carcinogens. In these circumstances, the interpretation of the guideline can potentiallylead to differing classifications.
• The ICH M7 classification tool will assign compounds with negative Ames results with positive carcinogenicity data as ‘Inconclusive’. While some would consider this situation as class 5, the justification behind this inconclusive prediction isthat further expert review may be required in order to assign the classification accurately.
• Contrasting examples that fall within the described classification are 2-methylimidazole and methyl eugenol, which can be seen in figures 6 and 8. They first appear to be compounds that would be classified equally and the workflow tool hasassigned them both as ‘inconclusive’. However, further analysis of methyl eugenol reveals that it is positive in several other genotoxic in vitro assays, such as the unscheduled DNA synthesis assay, as well as an in vivo mutagenicity assay5 andit could therefore be considered inaccurate to classify this compound as class 5. All the genotoxic in vitro data for 2-methylimidazole in Vitic Nexus is negative, so a prediction of class 5 is accurate.
• These challenging examples demonstrate how the ICH M7 workflow tool provides relevant information and guides the user to focus on compounds that demand further expert review. Lhasa would be interested to hear your feedback onhow you may classify these compounds and if you think the workflow is correct in its prediction. Feedback should be emailed to [email protected].
Information Considered in the ICH M7 Workflow
