Читать книгу Mutagenic Impurities - Группа авторов - Страница 94

2.4.6.1 Question 6.1

In Section 6 of the guideline, it says that (emphasis on the critical phrase): “A computational toxicology assessment should be performed using (Q)SAR methodologies that predict the outcome of a bacterial mutagenicity assay (Ref. 6). Two (Q)SAR prediction methodologies that complement each other should be applied. One methodology should be expert rule‐based and the second methodology should be statistical‐based. (Q)SAR models utilizing these prediction methodologies should follow the general validation principles set forth by the Organisation for Economic Co‐operation and Development (OECD).”

Question	Answer
What information and/or documentation should be provided to regulatory agencies to sufficiently demonstrate validation of (Q)SAR models that are developed in‐house or are not commonly used?	Section 6 of ICH M7 states that “(Q)SAR models utilizing these prediction methodologies should follow the general validation principles set forth by the Organization for Economic Co‐operation and Development (OECD)” (OECD Validation 2007). In the context of ICH M7, the OECD Principles of (Q)SAR Validation are:A defined end point – the model should be trained using experimental data generated according to the standard OECD protocol for the in vitro bacterial reverse mutation assay.An unambiguous algorithm – the algorithm used to construct the model should be disclosed. It should be clear whether the model is considered statistical (constructed via machine learning) or expert rule‐based (created from human expert‐derived knowledge).A defined domain of applicability – describe whether a test chemical falls within the model's applicability domain and how it is calculated. It should warn the user when the model does not have enough information to make a reliable prediction on a chemical.Appropriate measures of goodness‐of–fit, robustness, and predictivity – the model should be evaluated and shown to be sufficiently predictive of bacterial reverse mutagenicity. Standard validation techniques that should be used are recall, cross‐validation, and external validation. Evidence that the model has not been over‐fit should also be provided.A mechanistic interpretation – is there adequate information to allow an assessment of mechanistic relevance to be made (e.g. specific descriptors)? For any system developed in‐house or not commonly used, to demonstrate how each model follows these principles and to understand how a (Q)SAR model was developed and validated, submission of the OECD (Q)SAR Model Reporting Format (QMRF) (OECD QRMF 2017) for each model used should accompany each regulatory submission. A harmonized template for the QMRF was developed by the Joint Research Centre (JRC) and EU Member State authorities. This template summarizes and reports key information on (Q)SAR models, including the results of any validation studies, as well as provides supplementary information on applicability of the model to a given chemical.

Question

Answer

What information and/or documentation should be provided to regulatory agencies to sufficiently demonstrate validation of (Q)SAR models that are developed in‐house or are not commonly used?

Section 6 of ICH M7 states that “(Q)SAR models utilizing these prediction methodologies should follow the general validation principles set forth by the Organization for Economic Co‐operation and Development (OECD)” (OECD Validation 2007). In the context of ICH M7, the OECD Principles of (Q)SAR Validation are:A defined end point – the model should be trained using experimental data generated according to the standard OECD protocol for the in vitro bacterial reverse mutation assay.An unambiguous algorithm – the algorithm used to construct the model should be disclosed. It should be clear whether the model is considered statistical (constructed via machine learning) or expert rule‐based (created from human expert‐derived knowledge).A defined domain of applicability – describe whether a test chemical falls within the model's applicability domain and how it is calculated. It should warn the user when the model does not have enough information to make a reliable prediction on a chemical.Appropriate measures of goodness‐of–fit, robustness, and predictivity – the model should be evaluated and shown to be sufficiently predictive of bacterial reverse mutagenicity. Standard validation techniques that should be used are recall, cross‐validation, and external validation. Evidence that the model has not been over‐fit should also be provided.A mechanistic interpretation – is there adequate information to allow an assessment of mechanistic relevance to be made (e.g. specific descriptors)? For any system developed in‐house or not commonly used, to demonstrate how each model follows these principles and to understand how a (Q)SAR model was developed and validated, submission of the OECD (Q)SAR Model Reporting Format (QMRF) (OECD QRMF 2017) for each model used should accompany each regulatory submission. A harmonized template for the QMRF was developed by the Joint Research Centre (JRC) and EU Member State authorities. This template summarizes and reports key information on (Q)SAR models, including the results of any validation studies, as well as provides supplementary information on applicability of the model to a given chemical.

Most stakeholders currently use commercial software that have been validated and approved by the regulatory agencies. However, if one were to develop their own in‐house software, then the criteria for validation, as described above, need to be rigorously followed.

Подняться наверх