Researchers developed Fragmentia-AI-COAD, a transformer-based AI model that analyzes cell-free DNA (cfDNA) fragmentomic patterns in blood to detect colorectal cancer (CRC) and advanced adenomas. Published in Cell Communication and Signaling, the study used low-pass whole-genome sequencing data to identify cancer-associated DNA fragmentation patterns without relying on predefined molecular features or substantial sequencing depth.
In an independent test cohort, the model achieved an area under the curve (AUC) of 0.971 for CRC detection and 0.914 for advanced adenoma detection. Overall sensitivity was 92.9%, with specificity of 93.3%. Sensitivity for stage I CRC was 90.5%, while advanced adenoma sensitivity was 76.2%.
The model combines transformer-based sequence encoding with attention-based multiple instance learning to analyze millions of cfDNA fragments. Further analysis identified patterns including shorter DNA fragments, increased sequence complexity, and specific sequence motifs. Performance was largely maintained in analyses using fewer fragments from existing sequencing libraries.