Explainability Methods from Machine Learning Detect Important Drugs' Atoms in Drug-Target Interactions
Clinical Snapshot
PICO Framework
| P — Population | Graph neural network models trained on kinase and G-protein-coupled receptor (GPCR) drug-target interaction datasets |
| I — Intervention | Four explainable artificial intelligence (XAI) attribution methods applied to GNN models |
| C — Comparator | Comparison between different XAI methods and validation against 3D protein-ligand structural data |
| O — Outcomes | Consistency of XAI methods (atom-level intersection over union), biological relevance of attributed atoms, enrichment for binding pocket contact atoms |
Bottom Line
This methodological study demonstrates that explainable AI methods can identify biologically meaningful features in machine learning models predicting drug-target interactions, despite modest agreement between different XAI approaches. The finding that consensus attributions are highly enriched for atoms directly contacting protein binding pockets (76% within 2Å for kinases) provides confidence that these methods capture genuine chemical interactions rather than spurious correlations. While the modest consistency between methods suggests caution in relying on any single XAI approach, the biological validation against 3D structures offers a pathway for developing more interpretable AI systems in drug discovery. This work addresses a critical need for transparency in AI-driven pharmaceutical research, though further validation across diverse target classes and larger datasets would strengthen the conclusions.
Key Findings
P Value: Not reported in abstract
Effect Size: 76% enrichment for atoms within 2Å of binding pocket in kinase-inhibitor complexes
Primary Outcome: Biological relevance of XAI-attributed atoms in drug-target interactions
Nnt Or Sensitivity: Atom-level intersection over union (IoU) used for consistency assessment
Confidence Interval: Not reported in abstract
Clinical Application
High - computational methods can be readily implemented in existing drug discovery pipelines Relevant to Australian pharmaceutical research institutions and biotech companies developing novel therapeutics, particularly those using AI/ML approaches supported by government innovation programs Pharmaceutical researchers and computational biologists working on drug discovery
Abstract
Predicting drug-target interactions (DTI) with graph neural networks (GNNs) is hindered by their lack of interpretability. To address this, we benchmark four explainable artificial intelligence (XAI) attribution methods on GNN models trained for kinase and G-protein-coupled receptors (GPCR) targets. We assess the methods' consistency through atom-level intersection over union (IoU) and validate their biological relevance by mapping attributed atoms to three-dimensional (3D) protein-ligand structures. While consistency across methods was modest, consensus attributions were highly enriched for atoms directly contacting the binding pocket─up to 76% within 2 Å in the kinase-inhibitor complexes. Notably, these attributed atoms were frequently found contacting experimentally important regulatory residues such as those in the DFG motif. This indicates that XAI methods, despite their disagreements, can identify chemically meaningful ligand features, providing a foundation for developing more interpretable GNNs in drug discovery.
References
- 1.Mahindran, M., Liu, Q., Kadambalithaya, V. M., & Kalinina, O. V. (2026). Explainability Methods from Machine Learning Detect Important Drugs' Atoms in Drug-Target Interactions. Journal of Chemical Information and Modeling. https://doi.org/10.1021/acs.jcim.6c00037
Related Research
CPT: pharmacometrics & systems pharmacology
Structured Schemas for Provenance-Rich, LLM-Assisted QSP Model Calibration
2 Aug 2026
Journal of pharmacokinetics and pharmacodynamics
Diffusion models for virtual populations and pharmacometric simulations
1 Aug 2026
Journal of chemical information and modeling
HyperDC: A Non-Uniform Hypergraph Framework for Dual- and Higher-Order Drug Combination Recommendation Across Diverse Complex Diseases
29 July 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service