Agentomics

Agentomics autonomously develops novel machine learning solutions for biomedical datasets through an LLM-powered agentic system that implements various modeling strategies with strict validation checkpoints to produce ready-to-use models.

Key Features:

• Autonomous ML Experimentation: Agentomics performs end-to-end machine learning experimentation given a biomedical dataset without human intervention.
• Validation Checkpoints: Strict validation checkpoints for standard ML development steps enable gradual development on top of working code with defined interfaces and validated artifacts.
• Biomedical Foundation Model Support: Native support for biomedical foundation models that can be leveraged during experimentation.
• Generic Architecture: Supports creation of ML solutions for a large variety of datasets and use of various large language models.

Scientific Applications:

• Protein Engineering: Agentomics generated novel state-of-the-art models for established benchmark datasets in protein engineering.
• Drug Discovery: Agentomics outperformed other agentic systems when evaluated on drug discovery datasets.
• Regulatory Genomics: Agentomics was evaluated and benchmarked on regulatory genomics datasets.

Methodology:

Agentomics implements various ML modeling strategies with strict validation checkpoints for standard ML development steps, enabling gradual development on top of working code with defined interfaces and validated artifacts; it was evaluated across 20 datasets from Protein Engineering, Drug Discovery, and Regulatory Genomics and benchmarked against other agentic systems and human expert solutions.