A cutting-edge AI tool, DOLPHIN (published by McGill researchers in Nature Communications), has set a new standard for discovering cancer biomarkers in single-cell data. DOLPHIN’s innovation is analyzing RNA sequencing data at exon and splice-junction levels, allowing it to detect subtle transcriptomic variations that gene-level methods overlook. In a pivotal test with pancreatic cancer patients, DOLPHIN identified over 800 cancer-associated genetic markers missed by traditional approaches and differentiated between high-risk and less aggressive patient groups.

By generating richer and finer-grained single-cell profiles, DOLPHIN not only supports earlier, more accurate cancer diagnosis but also lays a foundation for “virtual cell” modeling to simulate drug responses in silico. These simulations may revolutionize future drug development, enabling faster and more precise therapy optimization without laborious preclinical testing.

The DOLPHIN team’s next challenge is scaling up: applying its deep learning framework to millions of cells across multiple datasets, moving toward a universally applicable platform for personalized medicine—not just in oncology, but potentially in autoimmune and neurodegenerative diseases as well.

Source.