Back

Contextual Science and Genome Analysis for Air-Gapped AI Research

Lazo, G. R.; Ayyappan, D.; Sharma, P. K.; Tiwari, V. K.

2025-03-24 bioinformatics
10.1101/2025.03.21.644606 bioRxiv
Show abstract

Provided here is a study of large language models (LLMs) and retrieval augmented generation (RAG) frameworks in air-gapped environments for genome research on small grain crops. We developed two main applications: (1) a RAG-based system for contextual analysis of scientific literature, collecting over 5,000 PDFs on wheat pathogens, and (2) a GFF3 file analysis tool called Genoma that enables exploration of genome annotation through an interactive interface. Using the open-source framework Ollama, we compared the performance of multiple LLMs including Llama3.1, Deepseek-r1, and Qwen2.5 for biological data analysis. A LightRAG approach provided semantic visualization of document relationships, while the Genoma tool offered chromosome-level insights into genome annotations. These tools demonstrate the viability of powerful AI assistance for sensitive research environments, with potential applications for pangenome analysis and gene discovery. The code provides researchers with practical solutions for implementing AI in secure settings without sacrificing analytical power.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.