HEDGEHOG: hierarchical evaluation of drug generators through rigorous filtration

Generative molecular models can support early drug discovery by proposing new candidate compounds de novo. In practice, useful candidates must balance target-relevant activity, physicochemical properties, and other multiparameter design constraints. However, standard metrics commonly used to evaluate molecular generators only weakly reflect whether the generated compounds are medicinally plausible and suitable for downstream computation. This can produce an incomplete view of model performance and inefficient use of computational resources. We introduce HEDGEHOG, a configurable six-stage evaluation framework that is constructed as a hit identification workflow: (i) preprocessing; (ii) physicochemical descriptor screening; (iii) structural alerts and graph-sanity checks; (iv) synthesis feasibility; (v) docking; and (vi) 3D pose and interaction checks. We evaluated $22$ generative models in a KRAS G12D case study, using three runs of $1000$ requested generation attempts per model. The models showed different patterns of attrition, and final survival ranged from zero to $127.0 \pm 18.1$ molecules per run. None of the standard metrics showed a significant association with final survival after correction for multiple testing. HEDGEHOG provides a reproducible framework for evaluating molecular generative models under defined screening criteria. The framework identifies stage-wise failure modes across generators and provides a practical basis for developing molecular generators better aligned with early drug discovery.