Talos: Open-source AI cuts rare disease diagnosis time to 32 days

Talos transforms rare disease diagnosis by automating the reanalysis of stored genomic data as scientific knowledge evolves. Developed through a collaboration between the Centre for Population Genomics, Australian Genomics, the Broad Institute, and Microsoft, Talos is an open-source pipeline that re-interprets existing variant calls against continuously updated resources like PanelApp Australia and ClinVar. It prioritizes variants likely to meet clinical reporting criteria while deliberately returning a small set of high-confidence candidates to minimize expert review time. Talos handles single-nucleotide variants, indels, copy number variants, and structural variants from exome or genome data.

In validation across nearly 1,100 patients, Talos recovered 90% of diagnoses while flagging only 1.3 variants per patient. When deployed on a prospective cohort of almost 5,000 undiagnosed patients, it produced 241 new diagnoses—a 5.1% additional yield—with an average of just 32 days between new evidence becoming public and the diagnosis. On monthly iterative cycles, analysts reviewed only one new variant per 200 patients, proving that frequent systematic reanalysis is sustainable. This addresses the critical problem that over half of rare disease patients remain undiagnosed after initial testing due to incomplete genomic knowledge.