HKU unveils ClairS-TO and Clair3-RNA deep-learning tools for cancer genomics

ClairS‑TO and Clair3‑RNA are good examples of how advances at the “applied” edge of AI quietly push the frontier forward. Long‑read sequencing data is messy and error‑prone, and getting robust small‑variant calls out of tumor‑only samples or RNA reads is a brutal pattern‑recognition task. HKU’s team is essentially distilling the tacit knowledge of expert bioinformaticians into models that outperform traditional pipelines and even prior AI tools, and doing so in open source.([miragenews.com](https://www.miragenews.com/ai-tools-revolutionize-cancer-diagnosis-genomic-1607422/))

For the race to AGI, this matters in two ways. First, medical genomics is one of the highest‑value domains for advanced pattern recognition; demonstrating that deep nets can safely replace hand‑tuned heuristics in this space expands the set of tasks we entrust to AI. Second, better genomic tools generate cleaner, richer biological datasets, which in turn feed into foundation models for biology and drug discovery. Those models are already pushing into territory—like protein design—that used to be purely human‑driven research.