Network for knowledge Organization (NEKO): An AI knowledge mining workflow for synthetic biology research

Area of Science:

• Bioinformatics
• Artificial Intelligence in Science
• Scientific Literature Mining

Background:

• Large language models (LLMs) offer general scientific Q&A but are limited by outdated information and lack specific, cited knowledge.
• Existing AI tools often require substantial computational resources, limiting accessibility for many researchers.

Purpose of the Study:

• To introduce Network for Knowledge Organization (NEKO), a novel workflow designed to overcome LLM limitations in scientific knowledge retrieval.
• To enhance LLMs' ability to provide specific, cited, and up-to-date scientific information through literature text mining.

Main Methods:

• NEKO utilizes the Qwen LLM for extracting knowledge from scientific literature.
• The workflow generates knowledge graphs to connect bioinformatics entities and provides summaries from PubMed searches based on user keywords.
• Case studies were conducted on yeast fermentation and cyanobacterial biorefinery to demonstrate NEKO's applicability.

Main Results:

• NEKO generates informative, specific, and actionable outputs, outperforming GPT-4's general Q&A.
• The system successfully links bioinformatics entities and produces comprehensive literature summaries.
• Demonstrated applicability in complex biological research areas like fermentation and biorefinery.

Conclusions:

• NEKO significantly enhances LLM capabilities for scientific applications, including education, literature review, and hypothesis generation.
• The workflow provides a more informative and actionable alternative to standard LLM Q&A for scientific inquiry.
• NEKO promotes AI accessibility in research through flexible, lightweight local deployment options, democratizing advanced scientific foundation models.