textToKnowledgeGraph: generation of molecular interaction knowledge graphs using large language models for...

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence in Biology

Background:

• Knowledge graphs (KGs) are vital for biological data analysis but manual construction from literature is costly and time-consuming.
• Existing text-mining methods struggle with contextual understanding and inferring complex biological relationships.
• Large Language Models (LLMs) offer enhanced contextual knowledge for more accurate information extraction.

Purpose of the Study:

• To develop an automated method for extracting biological interactions from scientific literature.
• To represent extracted biological relationships in the structured Biological Expression Language (BEL).
• To overcome limitations of traditional text-mining approaches in capturing complex biological context.

Main Methods:

• Utilized Large Language Models (LLMs) to process scientific articles and extract biological interactions.
• Developed the open-source Python package `textToKnowledgeGraph` for automated extraction.
• Integrated an interactive application within Cytoscape Web for simplified extraction and exploration.

Main Results:

• Successfully extracted biological interactions directly into Biological Expression Language (BEL) format.
• Created an open-source tool and an interactive application for knowledge graph construction.
• Generated a reviewed dataset of extractions to facilitate future model fine-tuning.

Conclusions:

• The `textToKnowledgeGraph` tool enhances the automated construction of biological knowledge graphs.
• LLM-powered extraction into BEL provides a structured and computationally accessible representation of biological relationships.
• This work facilitates more efficient and accurate biological data analysis and knowledge discovery.