Percepta: Understanding ICH M7 Classification Assignment Logic, "Class 0 / Inconclusive" Results, and Impurity Workflow Behavior

Description

Users performing mutagenicity assessments in Percepta under the ICH M7 guidelines may observe unexpected class assignments or different results between standalone calculations and batch impurity profiling workflows. The most common scenarios include:

1. Experimental Ames Negative Compounds Assigned Class 3
   • A compound with known, verified negative bacterial reverse mutation assay (Ames) data (e.g., Acyclovir) is predicted as Class 3 (alerting structure, unrelated to API; no mutagenicity data) instead of Class 5 (no alerts, or alert with concluding negative mutagenicity data).
   • Why? Percepta’s algorithm prioritizes structural alerts over experimental data to avoid false negatives.
   • Regulatory Note: ICH M7 allows Ames-negative data to override alerts, but manual review is required .
2. "ICH M7 Class 0" or "Inconclusive" Displayed in Reports
   • In Spreadsheet calculations or summary reports, certain compounds are assigned Class 0. In the detailed report, this corresponds to ICH M7: Inconclusive (with notes such as "Low probability of positive Ames test", "Alerting structure", "Insufficient evidence from similar compounds").
   • Why? "Class 0" = numeric placeholder for "Inconclusive" (insufficient evidence to distinguish Class 3 vs. 5).
   • Action: Treat as Class 3 until expert review or additional testing (e.g., Ames assay)
3. API / Parent Compound Assigned Class 4 in Impurity Assessment Workflows
   • When an API/parent compound (e.g., Octenidine Dihydrochloride, row ID 1) is evaluated independently with Parent ID = 0, it yields Class 0 / Inconclusive.
   • Why? When the API is set as the Parent ID, Percepta compares it to itself, creating a calculation artifact.
   • Action: Ignore Class 4 for the API; use standalone assessment (Parent ID = 0)
   • However, when evaluated in the Impurity Profiling workflow with Parent ID = 1 (where row ID 1 is designated as the parent), the same row ID 1 is assigned Class 4 with the justification: "Alerting structure shared with parent compound; treat as non-mutagenic impurity".

Solution

Underlying Software Logic & Root Cause

• Conservative Weight-of-Evidence (WoE) Hierarchy:
  The Percepta ICH M7 algorithm adopts a conservative approach: whenever a confident "red" structural alert is flagged in silico, the algorithm avoids automatically concluding a safe classification (Class 5), even if secondary or negative experimental data exists in related records, unless explicitly configured to let the Ames negative call override all alerts.
• Numerical Representation of "Inconclusive" as "Class 0" in Spreadsheet:
  The ICH M7 guideline defines classes 1 through 5. In Percepta Spreadsheet, the ICH M7 Class column is stored as a numerical data field to support spreadsheet sorting, filtering, and data querying. Because "Inconclusive" is text, the system represents it numerically as 0. A value of 0 does not mean non-toxic or non-alerting; it denotes that the collected computational evidence is insufficient to distinguish conclusively between Class 3 and Class 5. Regulatory best practice requires treating an Inconclusive (Class 0) compound as Class 3 until human expert review or further testing clarifies the endpoint.
• Parent-Impurity Relational Comparison Engine:
  The Impurity Profiling Suite identifies Class 4 compounds (alerting structure shared with parent compound) by comparing structural alerts of each project entry against the structural alerts detected in the designated Parent ID. In current versions, when row ID 1 is specified as the parent (Parent ID = 1), the algorithm evaluates row ID 1 against the parent alert library (which is itself). Because its alerts match the parent's alerts, it assigns Class 4 as a derivative/impurity calculation artifact rather than an independent API evaluation.

Solution & Recommended Workflows

1. Handling Compounds with Experimental Ames Negative Data

Run prediction → If Class 3 but Ames-negative:

• Manually reclassify as Class 5 in reports.
• Cite ICH M7 Section 6.2.2 (Ames test overrides alerts)
• Workaround / Regulatory Justification: Under Section 6 of the ICH M7 guideline, "an appropriately conducted negative bacterial reverse mutation assay (Ames test) overrides in silico structural alerts." Toxicologists and regulatory teams can justify classifying the compound as Class 5 during expert review by appending the verified Ames assay reference to the final report documentation.

2. Interpreting and Handling "Class 0 / Inconclusive"

• When the Spreadsheet column shows 0:
  • Cross-reference the detailed report comments (e.g., "Insufficient evidence from similar compounds").
  • Treat the compound from a risk management standpoint similarly to Class 3 unless human expert evaluation, additional orthogonal QSAR models, or laboratory Ames assays are available to reassign it.
  • If submitting summary tables to external partners or regulators, re-label 0 as Inconclusive (Treated as Class 3) to avoid confusion.

3. Recommended Workflow for APIs and Impurity Batches (Avoiding Class 4 API Artifacts)

To prevent the API/parent molecule from erroneously inheriting a Class 4 designation:

• Two-Step Calculation Workflow:
  1. Step 1 (Standalone API Assessment):
     Set Parent ID = 0 for the API compound. Run the prediction to determine its true standalone mutagenicity profile and baseline alerts (e.g., Class 1, 2, 3, 5, or Inconclusive/0).
  2. Step 2 (Impurity / Derivative Assessment):
     Set the impurities' Parent ID to the row number corresponding to the API (e.g., Parent ID = 1). Run the calculation to evaluate alert-sharing for actual impurities.
• Handling the Summary / Batch Report:
  • Disregard the Class 4 designation generated for row ID 1 itself when Parent ID = 1. Class 4 is only scientifically valid for derivatives/impurities relative to the active substance, not for the API itself.
  • In your final risk documentation, report the API’s standalone assessment (Step 1) and present the impurities relative to that baseline.

Please note, that in silico predictions are decision-support tools. Whenever there is any doubt, conflict between endpoints, an Inconclusive result, or an unexpected class assignment, a qualified toxicologist/expert review is always required to establish the final regulatory classification and justification.