preclinical-safety-assessment

1. When to use, and when not

Use this skill when you need to interpret preclinical toxicology data and compute safety margins. Entry points include:

• Ingesting and normalising repeat-dose, safety pharmacology, or genotoxicity study data for downstream analysis.
• Computing therapeutic index (TI) from NOAEL exposure and PK NCA data — cross-artifact margin calculation requiring both tox and PK artifacts.
• Assessing a genotoxicity battery using ICH S2(R1) weight-of-evidence classification.
• Computing measured hERG IC50 safety margins from safety pharmacology data and PK exposure.
• Applying the tox-safety-package threshold set to produce verdicts on margins, genotoxicity, or safety pharmacology artifacts.

Do not use when:

• You need predicted ADMET endpoints (metabolic stability, permeability, predicted hERG structural flag) -> admet-property-prediction.
• You need in vivo PK parameters, NCA analysis, or allometric scaling -> in-vivo-pk-analysis (though PK NCA artifacts are a required input for margin calculation here).
• You need compound drug-likeness rules or structural alerts -> compound-property-profile.

2. Preconditions

• Input schemas: canonical JSON matching one of:
  • oase.tox-repeat-dose.v1 — repeat-dose study data (for ingest, margins).
  • oase.tox-safety-pharm.v1 — safety pharmacology data with measured hERG IC50 (for ingest, margins --herg, analyze).
  • oase.tox-genotox.v1 — genotoxicity battery results (for ingest, genotox).
• Cross-artifact dependency: tox margins requires BOTH a tox repeat-dose artifact AND a PK NCA artifact (oase.pk-nca.v1, produced by in-vivo-pk-analysis). Optionally takes a safety-pharm artifact via --herg for measured hERG IC50 margin calculation. PK data is a precondition for margin computation — run in-vivo-pk-analysis first.
• oase doctor: run before first use. It ends with a verdict line: STOP means fix or report; PROCEED means work.
• No authentication needed — all operations are local.
• No network — all subcommands are offline.

3. Tool invocations

Run phase 1 (ingest, genotox, margins) before phase 2 (analyze). analyze reads from disk and can be re-run with different thresholds without recomputing.

ingest dispatches on the schema field in the input JSON — one command handles all three study types.

4. Interpretation contract

Verdicts

analyze applies the tox-safety-package@1.0 threshold set and emits per-artifact-type verdicts:

Margins analysis (oase.tox-margins.v1):

• acceptable — TI and hERG margins (when present) meet thresholds.
• flagged — one or more margins below threshold.
• incomplete — NOAEL exposure data not available; TI cannot be computed.

Genotox analysis (oase.tox-genotox-assessment.v1):

• acceptable — battery negative, no genotoxic concern.
• flagged_equivocal — mixed results, in vitro positive / in vivo negative pattern (equivocal).
• flagged_concern — positive or mixed results indicating genotoxic concern (includes all-positive, in vivo positive, or preponderance positive batteries).
• unknown — battery verdict not classifiable.

Safety-pharm analysis (oase.tox-safety-pharm.v1):

• recorded — safety pharmacology data recorded. Quantitative hERG margin requires running tox margins with both PK and safety-pharm artifacts.

Genotox battery classification

tox genotox classifies the battery result per ICH S2(R1):

• negative — all assays negative.
• positive — all assays positive.
• equivocal — in vitro positive / in vivo negative (WoE needed).
• concern — in vivo micronucleus positive with mixed results (WoE needed).
• positive_concern — 2+ positive results in a mixed battery (WoE needed).

Thresholds

The tox-safety-package@1.0 threshold set provides:

• ti_minimum — minimum acceptable therapeutic index fold-margin (ICH M3(R2)).
• herg_safety_margin — minimum hERG safety margin fold (ICH S7B).
• herg_marginal — operational boundary below the full hERG margin.
• noael_exposure_margin — UNRESOLVED. This threshold has no defensible default value. It is entirely program-specific. Do not invent a value — set it via .oase/thresholds.yaml per program and record the justification.

Thresholds are cited by name. The values in force are in the artifact: every .analysis.json carries threshold_set, thresholds_applied, threshold_sources, and threshold_provenance. Quote values from the analysis being cited, never from prose.

Mandatory relays

Upstream relay forwarding. tox margins reads the PK NCA sidecar and forwards any upstream mandatory relays (e.g., pk.nca_assumes_linearity). Any upstream relay present in the margins .meta.json must also be satisfied in the finding. Check mandatory_relays in both the .analysis.json and in each phase-1 .meta.json.

Consequence rules

• Margins acceptable: state the verdict, the threshold set, and which margins were computed (TI Cmax, TI AUC, hERG). Proceed to downstream work.
• Margins flagged: state which margins are below threshold and by how much (from the analysis, not from recomputation). A flagged TI below the configured ti_minimum is a substantive safety concern. A flagged hERG margin between herg_marginal and herg_safety_margin warrants attention; below herg_marginal is elevated concern.
• Margins incomplete: NOAEL exposure data was absent. The NOAEL dose is recorded but TI cannot be computed without measured exposure. This is not a failure — it is a data gap that requires toxicokinetic (TK) data.
• Genotox acceptable: battery negative, no genotoxic concern.
• Genotox flagged: state the battery classification and the weight-of-evidence basis. Satisfy the tox.genotox_weight_of_evidence relay if present.
• Safety-pharm recorded: data is recorded but quantitative margin assessment requires running tox margins with both PK and safety-pharm artifacts.

Stage 3 to Stage 4 boundary

When a safety-pharm artifact with measured hERG IC50 is available, the measured hERG margin from tox margins supersedes the predicted structural flag from Stage 3 ADMET (admet.herg_structural_flag). A measured IC50 and computed safety margin are direct evidence; a pharmacophore-based prediction is indirect. Do not carry both the measured margin and the predicted flag as independent findings — the measured margin replaces the prediction.

Cross-disciplinary consequences

• The therapeutic index assumes the PK file represents exposure at the intended therapeutic dose. oase.pk-nca.v1 does not tag dose context — this assumption cannot be verified from the artifact alone.
• A margins artifact without hERG data does not imply hERG safety — it means the measurement was not provided.
• Genotox battery classification is categorical. No threshold application occurs for genotox — the verdict maps directly from the battery result pattern.
• Safety-pharm data recording does not substitute for quantitative margin calculation. The hERG IC50 value is captured but not compared to therapeutic exposure until tox margins is run with both artifacts.

5. Failure modes and anti-fabrication guard

Do not compute these values yourself. The value is whatever the tool emitted. If the tool did not run, there is no value. If a required tool is unavailable, report the task as blocked. Do not estimate, and do not proceed on an assumed result.

Named pathologies

• Computing TI without PK data. tox margins requires both a tox repeat-dose artifact and a PK NCA artifact. Do not estimate a therapeutic index from the NOAEL dose alone — exposure-based TI requires measured exposure data.
• Treating an incomplete margins verdict as acceptable. Incomplete means TI could not be computed, not that the margin is adequate. A missing TI is a data gap, not a pass.
• Reporting a mixed genotox battery as negative. When the tox.genotox_weight_of_evidence relay fires, the battery had mixed results. The verdict is the classification after WoE reasoning, not a clean negative.
• Carrying both predicted and measured hERG findings. When a measured hERG IC50 margin is available from tox margins, it supersedes admet.herg_structural_flag. Do not present both as independent lines of evidence.
• Ignoring upstream PK relays on the margins artifact. tox margins forwards upstream relays from the PK sidecar (e.g., pk.nca_assumes_linearity). Every forwarded relay must be satisfied in the finding — they are not tox-specific, but they apply to the TI calculation that depends on the PK data.
• Inventing a value for noael_exposure_margin. This threshold is UNRESOLVED and program-specific. The CLI will refuse to apply it. Do not supply a value in prose or in a finding.

6. Provenance

Therapeutic index convention references ICH M3(R2) general guidance for starting dose selection and safety margin assessment. hERG safety margin references ICH S7B (adopted 2005) for the recommended margin between therapeutic free plasma Cmax and hERG IC50. Genotoxicity battery assessment references ICH S2(R1) for weight-of-evidence evaluation of mixed battery results. Provenance details are stamped in every .analysis.json via threshold_provenance.