[Paper Review] Explainable Artificial Intelligence for Drug Discovery and Development -- A Comprehensive Survey

Roohallah Alizadehsani, Oyelere, Solomon Sunday+6 more|arXiv (Cornell University)|Sep 21, 2023

Computational Drug Discovery Methods4 citations

TL;DR

This comprehensive survey proposes a systematic integration of Explainable AI (XAI) into drug discovery and development to enhance transparency and trust in complex ML models. By analyzing XAI techniques across target identification, compound design, and toxicity prediction, the paper demonstrates how interpretability can improve decision-making, reduce bias, and support regulatory approval—offering a roadmap for future research in dynamic, ethical, and domain-specific XAI applications.

ABSTRACT

The field of drug discovery has experienced a remarkable transformation with the advent of artificial intelligence (AI) and machine learning (ML) technologies. However, as these AI and ML models are becoming more complex, there is a growing need for transparency and interpretability of the models. Explainable Artificial Intelligence (XAI) is a novel approach that addresses this issue and provides a more interpretable understanding of the predictions made by machine learning models. In recent years, there has been an increasing interest in the application of XAI techniques to drug discovery. This review article provides a comprehensive overview of the current state-of-the-art in XAI for drug discovery, including various XAI methods, their application in drug discovery, and the challenges and limitations of XAI techniques in drug discovery. The article also covers the application of XAI in drug discovery, including target identification, compound design, and toxicity prediction. Furthermore, the article suggests potential future research directions for the application of XAI in drug discovery. The aim of this review article is to provide a comprehensive understanding of the current state of XAI in drug discovery and its potential to transform the field.

Motivation & Objective

• Address the critical need for interpretability in AI/ML models used in drug discovery, where complex predictions lack transparency.
• Systematically review XAI techniques and their applications across key drug discovery stages, including target identification, compound design, and toxicity prediction.
• Identify limitations and challenges of current XAI methods in the context of biomedical data and regulatory requirements.
• Propose future research directions to advance XAI, including dynamic explanation adaptation, ethical considerations, and hybrid modeling.
• Bridge the gap between technical AI outputs and practical decision-making for researchers, clinicians, and regulators through domain-specific explanation frameworks.

Proposed method

• Conduct a comprehensive survey of state-of-the-art XAI methods, including post-hoc explanation techniques (e.g., LIME, SHAP), inherently interpretable models, and attention-based methods.
• Categorize and analyze XAI applications in drug discovery across three core tasks: target identification, compound design, and toxicity prediction.
• Evaluate XAI frameworks based on their ability to provide faithful, actionable, and user-understandable explanations for diverse stakeholders.
• Integrate insights from cognitive psychology and visualization science to design human-centered XAI interfaces that support decision-making.
• Propose future methodological directions such as dynamic explanation adaptation to model and data shifts, and hybrid models combining black-box performance with interpretable reasoning.
• Explore domain-specific explanation languages to tailor AI outputs for non-technical users, including regulators and patients.

a Explainability vs. Prediction Accuracy

Experimental results

Research questions

• RQ1 How can XAI techniques improve the interpretability and trustworthiness of AI-driven predictions in drug discovery?
• RQ2 What are the key challenges and limitations of applying XAI methods to complex, high-dimensional biological and chemical data in drug development?
• RQ3 How can XAI be adapted to evolving models and streaming data in real-time drug discovery pipelines?
• RQ4 What ethical considerations—such as bias, fairness, and data privacy—arise when deploying XAI in healthcare-focused drug development?
• RQ5 How can XAI frameworks be designed to effectively communicate complex AI insights to multidisciplinary stakeholders, including clinicians and regulators?

Key findings

• XAI significantly enhances trust and transparency in AI-driven drug discovery by providing interpretable justifications for model predictions.
• Post-hoc explanation methods like LIME and SHAP are widely used but may lack fidelity when applied to complex deep learning models in molecular data.
• Incorporating XAI into target identification and toxicity prediction enables researchers to detect model biases and improve biological plausibility of predictions.
• Dynamic explanation adaptation is essential for maintaining interpretability as models are retrained or updated with new data.
• Ethical considerations such as bias, fairness, and data privacy are critical and must be proactively addressed in XAI design for healthcare applications.
• Future XAI frameworks should integrate domain-specific languages and visualization strategies to make AI insights accessible to non-technical stakeholders in drug development.

b Explainability vs. Learning Performance