This is a project for the 2025 NIAID BRC AI Codeathon.
Project page: https://niaid-brc-codeathons.github.io/projects/hiperrag-literature-extraction/
- Automated Knowledge Extraction and Curation
- Name: Ozan
- Affiliation: Argonne National Laboratory, BV-BRC
This project leverages HiPerRAG—a high-performance retrieval-augmented generation system optimized for large scientific corpora—to extract and curate structured data for priority pathogens. By targeting key relationship types such as protein–protein interactions (PPIs), host–pathogen interactions, and drug–protein binding data, the project aims to produce curated, machine-readable datasets for integration with BV-BRC knowledgebases.
- HiPerRAG codebase: https://github.com/ramanathanlab/distllm/tree/main
- HiPerRAG paper: https://arxiv.org/abs/2505.04846
- Goal 1: Define target data types relevant to CEPI and BV-BRC (e.g., PPIs, drug–protein interactions)
- Goal 2: Deploy HiPerRAG on relevant literature corpora to extract structured biological relationships
- Goal 3: Generate curated datasets for 1–2 CEPI priority pathogens (e.g., Nipah, Lassa)
HiPerRAG will be configured to parse biomedical literature and extract relations using fine-tuned retrieval and extraction modules. The system’s hybrid pipeline combines dense retrieval, passage re-ranking, and LLM-based summarization to produce high-confidence knowledge graphs. The team will evaluate both fully automated and human-in-the-loop curation workflows to balance scale and accuracy.
| Resource Type | Source / Link | Description / Purpose |
|---|---|---|
| Data | PubMed, BV-BRC text corpora | Literature sources for entity/relation extraction |
| Tools / Services | HiPerRAG | RAG-based extraction framework |
| LLMs / AI Models | Mistral Large, GPT-4 (Rhea) | Entity normalization and summarization |
| Compute / Storage | Argonne HPC, BRC clusters | Parallel literature processing |
- Curated datasets of structured biological relationships for CEPI priority pathogens
- Machine-readable outputs suitable for integration into BV-BRC pipelines
This project demonstrates scalable, AI-driven literature mining for infectious disease research. It enables automated knowledge enrichment and accelerates understanding of pathogen biology, supporting CEPI’s 100-day mission and BV-BRC’s informatics and data curation goals.