LitSumm: large language models for literature summarization of noncoding RNAs.

Area of Science:

• Life Sciences
• Bioinformatics
• Computational Biology

Background:

• Life sciences literature is rapidly expanding, overwhelming manual curation efforts for biomedical knowledgebases.
• Limited curator resources necessitate innovative solutions to manage the increasing publication volume.
• Prioritization of literature is essential for knowledgebase development due to resource constraints.

Purpose of the Study:

• To develop an automated method for summarizing scientific literature on noncoding RNAs (ncRNAs).
• To alleviate the bottleneck of manual curator time in RNA science.
• To leverage large language models (LLMs) for efficient literature summarization.

Main Methods:

• Utilized a commercial large language model (LLM) for text summarization.
• Implemented a chain of prompts and automated checks to ensure summary quality and factual accuracy.
• Applied the developed tool to generate summaries for over 4600 ncRNAs.

Main Results:

• Demonstrated the generation of high-quality, factually accurate summaries with correct references.
• Manual assessment confirmed the majority of generated summaries were of extremely high quality.
• Generated summaries for >4600 ncRNAs are now available through the RNAcentral resource.

Conclusions:

• Automated literature summarization using current LLMs is a viable solution for managing scientific publications.
• Careful prompt engineering and automated checking are crucial for successful LLM-based summarization.
• This methodology offers a scalable approach to support biomedical knowledgebases and RNA science research.