LLM Agent Based Protein Function Prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Artificial intelligence in biology

Background:

• Protein function prediction is a critical but challenging task in computational biology.
• Existing methods often struggle with integrating diverse biological data sources.
• Accurate protein function annotation is essential for understanding biological systems and disease mechanisms.

Purpose of the Study:

• To develop and evaluate a Large Language Model (LLM) agent-based system for improved protein function prediction.
• To leverage knowledge-augmented reasoning and multi-source evidence synthesis for enhanced prediction accuracy.
• To provide transparent and explainable predictions by documenting the reasoning process.

Main Methods:

• Integration of computational predictions with structured protein metadata, scientific literature, and ontological knowledge.
• A multi-stage reasoning process utilizing an LLM agent with specialized tools for querying, cross-referencing, and plausibility checks.
• Systematic evaluation against baseline methods across Gene Ontology sub-ontologies using multiple performance metrics.

Main Results:

• The LLM agent system demonstrated superior performance in threshold-dependent measures compared to established baseline methods.
• Achieved the lowest Smin scores across all evaluated Gene Ontology sub-ontologies.
• Attained the best Fmax scores for the Molecular Function and Cellular Component ontologies.

Conclusions:

• The proposed LLM agent-based system significantly enhances protein function prediction accuracy.
• Knowledge-augmented reasoning and multi-source evidence synthesis are effective strategies for improving biological predictions.
• The system offers a promising approach for advancing computational biology research and applications.