Researchers led by Haoyang Liu released GenoMAS, a team of language‑model agents that work together to clean and interpret transcriptomic data. The agents follow a guided‑planning framework that lets them break tasks into small actions and adjust their steps when data are messy. On the GenoTEX benchmark, GenoMAS reached 89.13% correctness for preprocessing and an F1 score of 60.48% for finding trait‑linked genes, beating previous methods by 10.61% and 16.85% respectively.
Why it matters
If the reported gains hold, biologists could rely on fewer manual steps to discover gene‑phenotype links.
