An interoperable research agent network for scientific discovery

Recent breakthroughs in agentic systems have facilitated the autonomous execution of research tasks across various scientific domains. However, the swift emergence of specialized scientific agents for fields like computational pathology, microbiome research, gene editing, materials science, organic chemistry, and drug discovery has resulted in a fragmented ecosystem of scientific capabilities.

While these agents often exhibit exceptional performance within their specific domains, their limited interoperability hampers the integration of expertise across platforms and the coordination of intricate interdisciplinary workflows. In this paper, we present GUIA (Guided-research Utilizing Intelligent Agents), an interoperable research-agent network constructed upon a flexible Agent-to-Agent (A2A) communication architecture.

GUIA facilitates collaboration among in-house and third-party agents within shared workflows, enabling the accumulation of scientific capabilities through the integration of complementary expertise. To assess GUIA's effectiveness, we conducted four evaluations encompassing baseline benchmarking, integration of third-party single-agent systems, integration of third-party multi-agent systems, and cross-server agent collaboration.

Additionally, we illustrate the practical utility of GUIA through real-world applications in therapeutic target discovery, drug discovery, and spatial proteomics analysis. Our findings indicate that interoperable research-agent networks can synergize specialized expertise from independently developed systems, offering a scalable framework for augmenting scientific capabilities through collaboration.