[论文解读] Revisiting Gene Ontology Knowledge Discovery with Hierarchical Feature Selection and Virtual Study Group of AI Agents

Cen Wan, Alex Rodrigues de Freitas|arXiv (Cornell University)|Mar 20, 2026

Biomedical Text Mining and Ontologies被引用 0

一句话总结

该论文提出一个具有代理功能的 AI 虚拟学习小组，用以从分层特征选择中提取与衰老相关的 Gene Ontology 知识，并在四种模式生物中验证发现。它表明许多 AI 生成的观点与现有文献一致，并分析框架的内部机制。

ABSTRACT

Large language models have achieved great success in multiple challenging tasks, and their capacity can be further boosted by the emerging agentic AI techniques. This new computing paradigm has already started revolutionising the traditional scientific discovery pipelines. In this work, we propose a novel agentic AI-based knowledge discovery-oriented virtual study group that aims to extract meaningful ageing-related biological knowledge considering highly ageing-related Gene Ontology terms that are selected by hierarchical feature selection methods. We investigate the performance of the proposed agentic AI framework by considering four different model organisms' ageing-related Gene Ontology terms and validate the biological findings by reviewing existing research articles. It is found that the majority of the AI agent-generated scientific claims can be supported by existing literatures and the proposed internal mechanisms of the virtual study group also play an important role in the designed agentic AI-based knowledge discovery framework.

研究动机与目标

• Motivate and design an agentic AI framework to extract ageing-related biological knowledge from GO terms selected by hierarchical feature selection.
• Show how a multi-layer AI agent system can interpret GO terms and produce literature-grounded reports.
• Evaluate the correctness and usefulness of the AI-driven knowledge discovery process in ageing biology.

提出的方法

• Introduce a bottom-up Virtual Study Group (VSG) framework with four model-organism-specific junior researchers, four senior researchers, and a principal investigator.
• Use hierarchical feature selection methods (HIP, MR, HIP-MR) to select informative GO terms for ageing-related genes.
• Employ multiple LLMs (via CrewAI and Ollama) as agent cores to generate and critique reports and integrate findings.
• Validate AI-generated claims by literature review and highlight statements supported or not by existing studies.
• Describe tasks for each GO term set across Worm, Fruit Fly, Mouse, and Yeast to assess associations with ageing processes.

实验结果

研究问题

• RQ1 Can an agentic AI-based virtual study group reliably interpret GO terms linked to ageing across multiple model organisms?
• RQ2 Are the AI-generated biological claims verifiable by existing literature?
• RQ3 Does the bottom-up multi-agent structure help reduce hallucinations and improve interpretability in GO-based ageing knowledge discovery?
• RQ4 What role do hierarchical GO term selections play in guiding reliable knowledge extraction by AI agents?

主要发现

• Most junior-researcher generated claims align with known ageing biology and can be validated by existing literature.
• Senior researchers’ critiques help identify overgeneralizations and point to tissue- and context-dependent nuances in ageing mechanisms.
• The virtual ageing professor highlights the need for generalisability across species and acknowledges ROS pathways and reproductive associations as complex, context-dependent factors.
• The framework demonstrates that multiple AI agents can collaboratively review, critique, and synthesize GO-term–based knowledge, reducing single-model hallucination risk.
• The study notes some claims from AI outputs may lack existing support, illustrating the importance of literature validation within the VSG process.