Transformer model interprets raw cfRNA sequences to detect multiple cancers with high accuracy

GeneLLM, a Transformer that analyzes cell-free RNA nucleotide sequences directly, detected multiple cancers with ROC-AUCs of 0.925–0.996 in a multi-centre cohort—without relying on gene annotations, potentially enabling lower-depth, more scalable liquid biopsy screening.

• Design: Multi-centre diagnostic-accuracy study using a Transformer-based model (GeneLLM) on plasma cell-free RNA (cfRNA) reads; exact total N not reported in source.
• Key secondary: Maintained comparable classification performance using only one-sixth of the typical sequencing depth, suggesting cost-efficiency for screening programs.
• Limitation: Retrospective analysis; prospective validation and real-world screening performance not yet established.

RadPigeon summaries are original and for information only. They are not clinical advice.