Agentomics: an agentic system that autonomously develops novel state-of-the-art solutions for biomedical machine learning tasks.

Agentomics autonomously develops state-of-the-art biomedical machine learning models, outperforming existing agentic systems and matching human expertise on 11 of 20 datasets.

• Why it matters: Automating biomedical ML is essential due to the rapidly evolving data quality and quantity, yet current tools lack flexibility and reproducibility, hindering progress in understanding biological systems and therapeutics.
• What they did: The system employs an LLM-powered agentic approach, implementing diverse modeling strategies with validation checkpoints, supporting biomedical foundation models, and generating ready-to-use models across various datasets and LLMs, evaluated on 20 datasets from Protein Engineering, Drug Discovery, and Genomics.
• The result: Agentomics consistently outperformed other agentic systems and produced novel, high-performing models comparable to human experts, enabling more efficient and reliable biomedical data analysis and model development.