Variantscape: Retrospective Study of Large Language Model-Based Extraction of Variant-Treatment-Cancer Associations from Published Oncology Abstracts

Variantscape extracted and verified variant–treatment–cancer associations from 3.7 million oncology abstracts, building a literature graph with 10,571 nodes and 44,644 edges.

• Why it matters: Precision oncology needs contextual interpretation of tumour variants, but relevant information is often buried in unstructured abstracts and hard to retrieve consistently.
• What they did: The team built an automated pipeline and open-access web tool combining traditional NLP with large language models to extract, standardise, and analyse co-associations across eligible abstracts.
• The result: From 886,195 eligible articles, 37,895 (4.3%) contained all three elements, and LLM verification confirmed at least one association in 35,224 papers.

AI summary of the abstract · may contain errors · not medical advice · read the paper.