Variantscape: Large Language Model-Driven Mining of Biomedical Literature for Clinical Interpretation of Cancer Variants

Variantscape is an automated tool and web platform designed for interpreting gene variants in cancer. The system combines traditional natural language processing methods with advanced language models to extract information from biomedical articles. From over 3 million abstracts, 335,817 articles containing gene names were processed. Among these, 7,423 articles (2.2%) simultaneously mentioned a variant, cancer type, and therapeutic agent, representing 3,902 unique variants across 98 cancer types and 388 therapeutic agents. Network analysis revealed 14,831 statistically significant associations. The tool supports clinical decision-making in oncology by rapidly linking information about variants, treatments, and cancer types from scientific literature.