Explainable and Auditable AI for Translational Drug Discovery

Artificial intelligence (AI) is rapidly transforming translational drug discovery by accelerating target identification, compound screening, biomarker discovery, and preclinical modelling. However, much of the current AI landscape still relies on opaque deep learning models whose decision-making processes are largely uninterpretable. This “black box” problem restricts clinical translation, limits regulatory acceptance, and raises concerns about reproducibility and bias. Recent advances in explainable AI (XAI), neuro-symbolic architectures, and auditable frameworks are addressing some of these limitations by promoting transparency, interpretability, and accountability in computational drug discovery pipelines. Yet, the field lacks agreed-upon methodologies and benchmarks for systematically evaluating model interpretability and auditability, especially in biomedical contexts where life-critical decisions are at stake.This Research Topic aims to bring together interdisciplinary efforts to develop and evaluate explainable and auditable AI systems that enhance the reliability of translational drug discovery workflows. It seeks to explore fundamental questions about how transparency, symbolic reasoning, and interpretability can be embedded in AI models while maintaining robust predictive performance. Key objectives include identifying frameworks for model auditing, determining how explainability and knowledge-guided AI contribute to hypothesis refinement in drug development, and assessing how transparent AI tools align with regulatory and ethical standards. Contributions are encouraged to address both methodological innovations and their translational applications in pharmacology, toxicology, and systems biology.To gather further insights into explainable and auditable AI applications within translational drug discovery, we welcome articles addressing, but not limited to, the following themes:• Development of explainable machine learning and neuro-symbolic AI methods for target identification, hit discovery, and lead optimization• Neuro-symbolic approaches integrating prior biological knowledge, biomedical ontologies, and knowledge graphs into deep learning models• Auditable frameworks for AI model validation, versioning, and reproducibility in drug pipelines• Integration of explainable AI in multi-omics data analysis and biomarker discovery• Regulatory, ethical, and reproducibility considerations for AI transparency in drug R&D• Visualization and interpretability methods for chemical space of small molecules and peptides, protein–ligand interactions, synergistic drug combinations, cellular permeability, pharmacokinetic, and toxicology modelling• Human–AI collaboration and user trust in AI-driven biomedical research• Benchmark datasets and performance metrics for explainability and auditability evaluation• Automated and interpretable design of machine learning models and algorithms for molecular representation and property prediction• Rigorous data splitting, validation strategies, and benchmarking practices for machine learning in drug discovery and development• Mitigation and detection of data leakage in AI models for molecular property prediction, target identification, virtual screening, and clinical applications• Bias identification, quantification, and correction in biomedical and chemical datasets used for AI-driven drug discovery• Assessing the impact of scaffold splitting, temporal splitting, cluster-based splitting, assay-based splitting, and external validation on model generalizabilityWe welcome original research, methods papers, reviews, perspectives, and brief reports that advance understanding of transparent, auditable, and trustworthy AI in drug discovery and translational applications. Topic editors: Alex de Sa, Kamya Goyal, Pinyi Lu Source: https://www.frontiersin.org/research-topics/83178/explainable-and-auditable-ai-for-translationaldrugdiscovery