Agentic AI and the Rise of in silico Team Science in Biomedical Research

Summary

A Nature Biotechnology review framing the emerging class of agentic AI systems as teams of intelligent computational experts — an "in silico team science" — capable of taking on labor-intensive biomedical research tasks such as literature review, hypothesis formulation, data analysis, and model interpretation. The paper organises the design space around three key algorithms and seven foundational building-block characteristics that distinguish deployable agentic systems from prompt-chaining prototypes, and surveys applications in drug discovery, data analysis, and biomarker identification.

Contribution

A structured typology for agentic-AI systems in a high-stakes domain. Where most prior reviews are either too general (LLM-as-tool-use survey) or too narrow (single application paper), this one positions itself at the seam between agent architecture and biomedical workflow — naming the algorithms and characteristics that practitioners should think about before deploying. The "in silico team science" framing is itself a contribution: it reframes agentic-AI from "automated assistant" to "distributed collaborator team", with implications for credit, oversight, and integration into existing research-lab social structures.

Method

Narrative review; surveys recent agentic-AI work in biomedical research and synthesises into a structured framework of algorithms and characteristics. No primary empirical evaluation.

Relevance to RISE

A useful theoretical anchor for the RISE concept paper. Three specific points of contact:

• "In silico team science" framing parallels the sociotechnical framing RISE proposes — agentic pipelines are not just tools but collaborator teams that need integration into existing research practices.
• Domain-specific concretisation: biomedical research is one of the fields where agentic pipelines have already demonstrated end-to-end value (ghareeb2026robin is the clearest example), so a review that surveys this surface is directly applicable when discussing field-level adoption.
• Building-block taxonomy complements the architectural decomposition in yang2026aris — Yang et al. argue for three system layers (execution, orchestration, assurance); Li et al. add seven characteristics that any layer needs to satisfy. The two papers are complementary perspectives on the same design problem.

Critique / open questions

A review-paper limitation applies: the seven characteristics and three algorithms are extracted from existing systems rather than validated empirically against a benchmark of deployment outcomes. The framework risks being too high-level for engineers and too prescriptive for theorists. The "team science" metaphor is provocative but unargued — whether agentic systems actually behave like research teams (with division of labor, conflict, coordination overhead) vs. like distributed function calls is an empirical question the paper does not test. As is typical for Nature Biotech review articles, the focus is biomedical; whether the framework ports to other research domains (economics, social science) is unaddressed.

Key quotes

"Agentic artificial intelligence (AI) systems are emerging as teams of intelligent computational experts capable of rivaling human performance in labor-intensive tasks, including literature review, hypothesis formulation, data analysis and model interpretation."

"These systems are poised to accelerate labor-intensive biomedical research by making autonomous decisions based on contextual information and expert feedback."

"Here we discuss three key algorithms and seven foundational building-block characteristics that contribute to the development of agentic AI systems. We highlight their biomedical applications, design considerations and the challenges and opportunities associated with deploying agentic AI systems to advance collaborative scientific research."