PharmaToxAI / Evidence architecture

From molecular predictions to experimental evidence.

Explore an example of an integrated, traceable compound assessment spanning structural alerts, statistical QSAR, ADMET, chemical analogues and targeted in vitro endpoints.

DEMONSTRATION · SYNTHETIC DATA. All compound identifiers, assay results, predicted metrics and analogue entries below are illustrative. No training, chemical calculation, laboratory measurement, independent validation or regulatory submission was performed to produce these examples.

00 / Report provenance

Assessment record ILLUSTRATIVE · NOT CERTIFIED

Report ID
Issue date22 Oct 2026
Compound ID / role
Input identity
Computational provenanceExample onlyNo executable model, weights, hashes or training set provided.
Experimental provenanceExample onlyNo raw data, batch record, SOP or lab signatures.

Comparison reference: . Intended context: hypothetical small-molecule drug-development screening. Regulatory classification requires confirmation of chemical identity, impurity context and source evidence.

01 / In silico hazard assessment

ICH M7-oriented mutagenicity assessment

Structural rules (expert system)Independent statistical QSARApplicability domainAnalogue review

Complementary evidence streams

StreamDemonstration findingInterpretation

Applicability and classification

Maximum Morgan/Tanimoto similarity (illustrative)
ICH M7 interpretation (provisional demonstration)

Read-across candidate records

Source identifierStructural relationshipEndpoint observationProvenance

Example similarity assumes radius-2 Morgan fingerprints and Tanimoto scoring. Thresholds and confidence require endpoint-specific calibration with a defined training set, nearest-neighbour distributions and external error analysis. Structural similarity alone does not substantiate mechanistic read-across.

02 / Predicted exposure and liability

ADMET · quantitative screening

Outputs are hypothetical model results, not empirical measurements. Kinetic and thermodynamic solubility are distinct experimental concepts and cannot be interchanged. Endpoint-specific units, assays, model versions, calibration and uncertainty must accompany real reports.

Risk profile · normalized display

Normalized scores are demonstration-only visualizations, not calibrated probabilities or regulatory thresholds. Higher indicates greater hypothetical liability.

03 / Targeted experimental follow-up

Cross-check in human cell and bacterial systems

Proposed validation plan: HepG2 and human dermal fibroblasts; orthogonal metabolic-activity (MTT) and ATP-based (CellTiter-Glo) viability readouts; exposure–response curves; intracellular DCFH-DA oxidation with interference controls; glutathione depletion where mechanistically relevant; exploratory microplate bacterial reverse-mutation screening in TA98/TA100 ±S9.

Prediction-to-assay evidence alignment

EndpointIn silicoIn vitro (synthetic)Concordance / caveat
A two-strain microplate screen is not a full OECD TG 471 bacterial reverse mutation test. A fold-change is not sufficient for a positive Ames classification; controls, dose response, cytotoxicity, reproducibility, appropriate strain coverage and test-design-specific criteria are needed. DCFH-DA is not ROS-species-specific; MTT is a metabolic readout, not a direct cell count.

04 / Evidence integration

Technical interpretation and next decision

Current illustrative disposition
Residual uncertainty
Regulatory statusNot submittedIndependent expert review required.

Report authentication

Generated report state: demonstration only. Digital signature: not applied. A production dossier must include accountable reviewer identity, signed record, audit trail, validated methods where required, raw data references and verified hashes.

PharmaToxAI · Integrated evidence demonstration · © 2026Controlled-example template · no regulatory endorsement

Methods / references

Framework and scope

ICH M7(R2), Assessment and Control of DNA Reactive (Mutagenic) Impurities in Pharmaceuticals: ICH guideline (PDF). Two complementary (Q)SAR methodologies and expert review support impurity assessment; classifications depend on context and evidence. Nitrosamines may require compound-specific approaches and additional regulatory guidance.

OECD Principles for the Validation of (Q)SAR Models: OECD guidance. Applicability domain must be defined for the specific model. Bacterial reverse mutation: OECD Test Guideline 471.

Read-across: EFSA guidance (2025; food/feed-specific scope). This reference informs scientific documentation and does not imply automatic acceptance under other jurisdictions.

Reporting levels: rule/ML outputs = predictions; microplate and cellular results = experimental observations only when actually measured; final risk decision = expert evidence integration. This page contains synthetic demonstrations at every level.