Pipeline to extract structured data from scientific literature using large language models (LLMs), as part of our MR-KG work (medRxiv preprint) for extracting Mendelian Randomization (MR) studies published on PubMed. The pipeline processes abstracts through multiple LLMs, validates outputs against JSON schemas, and generates manuscript-ready performance assessments.
See DEV.md for setup instructions and development workflow.
- DEV.md - Development setup, environment configuration, and task runners
- DATA.md - Data organization, file structure, and workflow outputs
- ANALYSIS.md - Complete analysis workflow from preprocessing to manuscript assets
- docs/manuscript-assets.md - Manuscript figures and tables generation
@article{liu2025mr-kg,
title = {MR-KG: A knowledge graph of Mendelian randomization evidence powered by large language models},
url = {http://dx.doi.org/10.64898/2025.12.14.25342218},
DOI = {10.64898/2025.12.14.25342218},
publisher = {openRxiv},
author = {Liu, Yi and Burton, Joshua and Gatua, Winfred and Hemani, Gibran and Gaunt, Tom R},
year = {2025},
month = dec
}