This review evaluates the emerging paradigm of multi-agent systems (MASs) for biomedical and clinical data analysis, focusing on their ability to overcome the reasoning and reliability limitations of standalone large language models (LLMs). We synthesize findings from recent architectural frameworks, specifically LangGraph, CrewAI, and the Model Context Protocol (MCP), to examine how specialized agent teams divide labor, utilize precision tools, and cross-verify outputs. We find that MAS architectures yield significant performance gains in various domains: recent implementations improved oncology decision-making accuracy from 30.3% to 87.2% and reached a peak of 93.2% accuracy on USMLE-style benchmarks through simulated clinical evolution. In clinical trial matching, multi-agent frameworks achieved 87.3% accuracy and enhanced clinician screening efficiency by 42.6% (p < 0.001). However, we also highlight critical operational challenges, including an unreliability tax of 15–50× higher token consumption compared to standalone models and the risk of cascading errors where initial hallucinations are amplified across the agent collective. We conclude that while MAS enables a shift toward collaborative intelligence in biomedicine, its clinical and research adoption requires the development of deterministic orchestration and rigorous cost-utility frameworks to ensure safety and expert-centered oversight.
Maintenance of fissed mitochondria is viewed as a defining feature of Kirsten rat sarcoma viral oncogene homolog (KRAS)-mutant cancers. However, regulation of this process by accompanying comutations or environmental factors is not clearly defined. Here, by analyzing a subset of pancreatic cancer lesions driven by concurrent Kras G12D and GNAS complex locus gene (GNAS R201C/H ) mutations, we found that despite the presence of mutant Kras, hyperactive Gnas R201C maintains mitochondria predominantly in a fused state, which is necessary for tumor growth. Multiplex proteomics, super-resolution microscopy, loss- and gain-of-function studies, coupled with metabolite rescue experiments, revealed that Gnas R201C -regulated branched-chain amino acid (BCAA) pathway is a previously unidentified regulator of mitochondrial morphology. Mechanistically, the BCAA pathway, the associated tricarboxylic acid cycle, and aspartate metabolism converge on nicotinamide adenine dinucleotide (NADH-NAD + ) metabolites to promote mitochondrial elongation. NAD + availability is crucial for mitochondrial fusion, as facilitating NAD + generation through alternative means promotes fusion. Collectively, we unraveled a new mechanism that drives mitochondrial fusion and showed that the combination of oncogenic signaling and metabolism can maintain distinct mitochondrial morphology within genetic subsets of KRAS-mutant pancreatic cancer.
Multi-agent AI systems, where multiple specialized agents collaborate, are emerging as a powerful approach in biomedicine to tackle complex analytical and clinical tasks that exceed the scope of any single model. Background: This review outlines how orchestrating large language model (LLM) based agents can improve performance and reliability in biomedical data analysis. It surveys new frameworks that coordinate agent teams and highlights state-of-the-art applications in domains such as drug discovery, clinical trial matching, and decision support, where early multi-agent prototypes have achieved higher accuracy or more robust results compared to lone LLMs. Methods: We synthesize findings from recent studies and architectures, categorizing applications and examining how agents divide labor, use tools, and cross-verify each other’s outputs. Results: The review finds that multi-agent strategies yield notable advantages – for example, reducing errors via inter-agent checking and providing more explainable reasoning through transparent dialogues. We also catalog available orchestration platforms and benchmarks driving this field. Conclusions: While multi-agent AI shows promise in augmenting biomedical research and healthcare (by integrating diverse knowledge sources and simulating collaborative problem-solving), ensuring its reliable and ethical deployment will require addressing challenges in verification, scalability, continual learning, and safety. The paper concludes that with careful design and rigorous evaluation, AI agent teams could significantly enhance biomedical intelligence without replacing human experts.