Open-source large language models in action: A bioinformatics chatbot for PRIDE database.

Area of Science:

• Bioinformatics
• Proteomics
• Computational Biology

Background:

• User interaction with large biological databases like PRIDE can be complex.
• Efficiently searching documentation and datasets is crucial for researchers.

Purpose of the Study:

• To develop a chatbot assistant infrastructure to simplify user interactions with the PRIDE database.
• To enhance the discoverability of PRIDE datasets and improve documentation accessibility.

Main Methods:

• Utilized multiple Large Language Models (LLMs) including llama2, chatglm, mixtral, and openhermes.
• Implemented a web service API, web interface, and vector database for indexing and management.
• Integrated an Elo-ranking system for LLM performance benchmarking and documentation improvement.

Main Results:

• The chatbot facilitates natural language interaction with PRIDE documentation.
• An LLM-based recommendation system enables effective PRIDE dataset searching and discoverability.
• The framework demonstrated modularity and adaptability for broader applications.

Conclusions:

• The developed chatbot infrastructure offers a robust and transferable solution for enhancing user experience in bioinformatics.
• The integration of LLMs, vector databases, and benchmarking improves data accessibility and resource utilization.
• The open-source nature of the framework promotes wider adoption and development in related scientific domains.