AI agents in oncology

This Collection supports and amplifies research related to SDG 3: Good Health & Wellbeing and SDG 4: Learning & Education.

Artificial Intelligence (AI) is rapidly transforming cancer research and clinical oncology. Over the past decade, deep learning models have demonstrated human, or above-human-level performance across a wide range of specific tasks, including image interpretation, genomics analysis, and risk prediction. However, these models are typically single-purpose, i.e. they are trained to perform one predefined task at a time, and require substantial manual effort to integrate into real-world biomedical workflows. As oncology becomes increasingly data rich and multimodal, there is a growing need for more flexible, interactive, and scalable AI solutions.

This Collection focuses on an emerging field: AI agents in oncology. AI agents are autonomous systems, often powered by large language models (LLMs) or multimodal foundation models, that can reason, plan, and act across complex sequences of tasks. Ideally, these agents can orchestrate diverse tools, synthesize information from heterogeneous data sources, and support multi-step clinical or research workflows. From literature review and trial design to precision treatment recommendations and real-time decision support, AI agents offer the potential to optimize and enhance oncologic practice and research.

Recent studies have demonstrated early evidence that autonomous AI agents can successfully support decision-making in oncology scenarios by combining capabilities such as LLMs, radiology and pathology image interpretation, genomic variant analysis, and guideline retrieval. These agents can potentially mimic the reasoning steps of a clinical expert, automatically select and apply relevant tools, interpret complex patient cases, and cite supporting evidence. Moreover, they provide a flexible alternative to general-purpose foundation models by using domain-specific components that are independently verifiable and updatable, which is a key consideration in clinical settings.

Despite their promise, AI agents in oncology are still in their early stage. Key challenges remain, including robust evaluation metrics, domain-specific benchmarking datasets, the need for clinical-grade trustworthiness, and clarity around regulatory and ethical frameworks. Questions about oversight, explainability, and human-in-the-loop design are especially relevant for agents that may operate semi-autonomously or in multiagent collaborative systems. Additionally, the integration of generative capabilities raises new possibilities and concerns related to hallucination, bias propagation, and patient safety.

This Collection invites original research, reviews, perspectives and comments on the development, evaluation, and frameworks for clinical implementation of AI agents in oncology. We welcome submissions on topics including but not limited to:

• AI agent architectures for clinical decision support
• Multi-step reasoning, planning, and tool orchestration
• Vision-language and generalist multimodal foundation models in oncology
• Integration of generative AI in cancer research workflows
• Clinical validation, regulatory strategies, and benchmarking
• Trustworthiness, transparency, and explainability in agentic systems
• Case studies demonstrating clinical or preclinical use of autonomous agents
• Collaborative multi-agent frameworks in cancer research

By selecting this rapidly emerging area, we aim to promote cross-disciplinary dialogue between AI developers, oncologists, computational biologists, and regulatory scientists. The goal is to accelerate the translation of autonomous and semi-autonomous AI systems into real-world oncology settings—supporting more personalized, efficient, and evidence-based cancer care.