All model cards

Reproductive Toxicity

classification · LightGBM (classification)

Excluded from the integrated risk score. Underpowered (n=127) — scaffold AUC=0.588, not significantly better than random on novel scaffolds. Reported for information only and excluded from the integrated risk score.

Purpose

Predicts reproductive/developmental toxicity potential. Reported for information only.

Prediction Output

Binary classification (reproductive-toxic / non-toxic) + probability.

Input Requirements

A single valid SMILES string, standardized server-side before featurization.

Training Methodology

Algorithm: LightGBM (classification) · Optuna, TPE sampler, 100 trials, random_state=42

Feature representation: 2,048-bit Morgan ECFP4 (radius=2) + 167 MACCS keys + 25 RDKit physicochemical descriptors → 2,240 raw features → variance threshold (>0.01) → Pearson correlation filter (|r|<0.95) → top 500 features → StandardScaler (fit on the training fold only).

  • Sources: curated teratogenicity and reproductive toxicity literature. N = 127 compounds — explicitly documented as severely underpowered.
  • Train/test: approximately 100 / 26 (scaffold split).

Validation Methodology

  • Scaffold-split GroupKFold (k=5). Random-split hold-out reported separately for comparison.

Scaffold-split AUC = 0.588 (near-random — the primary, reported figure). Random-split AUC = 0.929, which the model card explicitly calls "misleadingly high" and not the number to use.

Known Limitations

  • EXCLUDED from the integrated risk score (docs/risk_score.md §3.5) — with N=127 and scaffold AUC=0.588, the model provides no reliable generalization signal for novel scaffolds.
  • Target dataset size before this endpoint could be reconsidered for the integrated score: ≥1,000 compounds (documented roadmap item, CERAPP/CoMPARA dataset integration).
  • Do not use this endpoint's output for any safety-critical purpose.

Applicability Domain

Leverage-based applicability domain: h* = 3×(p+1)/n (Williams plot). in_domain=true → confidence 1.0; in_domain=false → confidence 0.0, and the prediction should be treated as exploratory only. Supplementary checks: Tanimoto k-NN distance > 0.6 or descriptor values outside [min−3σ, max+3σ].

Interpretability Support

SHAP (SHapley Additive exPlanations) feature attribution is available via /predict/interpret and /predict/integrated, returning per-feature contribution values and a global/local breakdown by feature category (physicochemical / structural fingerprints / substructure keys).

Research-Use Disclaimer

Research use only. Not validated for regulatory submissions or clinical decision-making. Predictions must not be the sole basis for safety decisions without independent experimental validation.

View this endpoint's dataset card →