[論文レビュー] Robin: A multi-agent system for automating scientific discovery

Ali Ghareeb, Benjamin Chang+8 more|ArXiv.org|May 19, 2025

Machine Learning in Bioinformatics被引用数 4

ひとこと要約

Robin is a multi-agent AI system that automates hypothesis generation, experimental design, data analysis, and hypothesis refinement to drive AI-assisted drug discovery, demonstrated by identifying ripasudil as a potential dry AMD treatment and revealing ABCA1-related mechanisms.

ABSTRACT

Scientific discovery is driven by the iterative process of background research, hypothesis generation, experimentation, and data analysis. Despite recent advancements in applying artificial intelligence to scientific discovery, no system has yet automated all of these stages in a single workflow. Here, we introduce Robin, the first multi-agent system capable of fully automating the key intellectual steps of the scientific process. By integrating literature search agents with data analysis agents, Robin can generate hypotheses, propose experiments, interpret experimental results, and generate updated hypotheses, achieving a semi-autonomous approach to scientific discovery. By applying this system, we were able to identify a novel treatment for dry age-related macular degeneration (dAMD), the major cause of blindness in the developed world. Robin proposed enhancing retinal pigment epithelium phagocytosis as a therapeutic strategy, and identified and validated a promising therapeutic candidate, ripasudil. Ripasudil is a clinically-used rho kinase (ROCK) inhibitor that has never previously been proposed for treating dAMD. To elucidate the mechanism of ripasudil-induced upregulation of phagocytosis, Robin then proposed and analyzed a follow-up RNA-seq experiment, which revealed upregulation of ABCA1, a critical lipid efflux pump and possible novel target. All hypotheses, experimental plans, data analyses, and data figures in the main text of this report were produced by Robin. As the first AI system to autonomously discover and validate a novel therapeutic candidate within an iterative lab-in-the-loop framework, Robin establishes a new paradigm for AI-driven scientific discovery.

研究の動機と目的

• Automate the full cycle of scientific discovery from literature mining to experimental data interpretation.
• Integrate hypothesis generation with autonomous experimental planning within a lab-in-the-loop framework.
• Identify novel therapeutic candidates for a target disease and validate them with in vitro experiments and analyses.
• Reveal mechanistic insights and potential molecular targets through AI-driven data analysis.

提案手法

• A multi-agent architecture integrating literature search agents (Crow and Falcon) with a data-analysis agent (Finch).
• Hypothesis generation driven by literature synthesis and an LLM-based judge to rank candidate hypotheses.
• Automated selection of in vitro disease models and corresponding assays guided by literature reviews.
• Experimental execution by researchers followed by autonomous data analysis with Finch, including multiple analysis trajectories and meta-analysis for consensus.
• RNA-seq and flow cytometry analyses to characterize transcriptional changes and phagocytosis outcomes, respectively.
• Iterative lab-in-the-loop workflow where results update subsequent rounds of hypotheses and experimental plans.

実験結果

リサーチクエスチョン

• RQ1 Can Robin autonomously generate disease-relevant therapeutic hypotheses from literature?
• RQ2 Can Robin design and interpret experiments to test these hypotheses and refine them based on results?
• RQ3 What molecular mechanisms or targets emerge from AI-driven analyses of experimental data?
• RQ4 Is the system capable of identifying and validating novel therapeutic candidates with a favorable mechanistic rationale?
• RQ5 How well does an AI-driven ranking/judgment align with expert human judgment in selecting promising hypotheses?

主な発見

• Robin identified 151 papers to propose ten disease mechanisms and corresponding in vitro models for dry AMD.
• Robin proposed testing compounds that increase RPE phagocytosis and generated 30 candidate drugs for the phagocytosis assay.
• Ripasudil, a ROCK inhibitor, outperformed Y-27632 and increased RPE phagocytosis about 7.5-fold vs DMSO controls in the phagocytosis assay.
• RNA-seq analysis revealed ABCA1 upregulation in ROCK inhibitor–treated RPE cells, linking phagocytosis enhancement to lipid transport pathways.
• Robin’s iterative loop refined hypotheses, enabling generation and validation of mechanistic insights such as actin remodeling and autophagy pathways.
• The automated analysis by Finch and the human analysis showed concordant findings, supporting the system’s reliability in data interpretation.