What question did this study set out to answer?

The aim is to identify and categorize the limitations of molecular structure in predicting drug toxicity.

June 3, 2026Open Access

What Molecular Structure Cannot Tell Us: A Taxonomy of Explainability Gaps in GNN-Based Drug Toxicity Prediction

Key Points

The aim is to identify and categorize the limitations of molecular structure in predicting drug toxicity.
Introduces a taxonomy classifying different types of explainability gaps in drug toxicity prediction.
Employs a Message Passing Neural Network (MPNN) trained on the Tox21 benchmark with GNNExplainer for analysis.
Quantifies the Missing Not At Random (MNAR) gap through a ChEMBL query across 42 assays.
Molecular structure accounts for approximately 45% (5/11) of known adverse effects of acetylsalicylic acid.
Identifies four categories of gaps (GAP-1 to GAP-4) affecting toxicity prediction.
The representation error is localized within the MPNN message passing layers.

Abstract

Not all clinically relevant adverse effects are structurally inferable from molecular graphs — regardless of model quality or architectural complexity. This study introduces an operational taxonomy of the structural information limits that prevent structure-based toxicity prediction, independent of the learning algorithm employed. A systematic case study using acetylsalicylic acid (ASA, Aspirin) serves as model compound. A Message Passing Neural Network (MPNN) is trained on the Tox21 benchmark and GNNExplainer is applied to characterize atom-level attribution. Results indicate that molecular structure explains approximately 45% (5/11) of known ASA adverse effects. A four-category Gap Taxonomy (GAP-1 through GAP-4) is introduced distinguishing between principally non-encodable effects, data gaps arising from Missing Not At Random (MNAR) mechanisms, assay panel mismatches, and representation errors. The MNAR gap is empirically quantified via a systematic ChEMBL query (42 documented assays, 0 retrievable bioactivity entries). An attention pooling experiment localizes the representation error to the MPNN message passing layers rather than the aggregation step. The Gap Taxonomy has direct implications for drug safety signal detection and regulatory frameworks including Good Pharmacovigilance Practice (GVP) guidelines and New Approach Methodologies (NAMs). Structural limits identified are confirmed in a companion DDI ablation study.

What Molecular Structure Cannot Tell Us: A Taxonomy of Explainability Gaps in GNN-Based Drug Toxicity Prediction

Key Points

Abstract

Cite This Study