PacBio on 4 August published a study in Nature Methods introducing the “Platinum Pedigree,” a long-read genome benchmark built from deep sequencing of a 28-member, multi-generational family. The dataset catalogues more than 37 megabases of genetic variation and adds over 200 million bases to existing reference regions, expanding benchmark coverage to 2.77 gigabases, including segmental duplications and other difficult-to-map areas. When Google’s DeepVariant AI variant-calling software was retrained with the new benchmark, erroneously called variants fell by 34 percent across the genome, with the largest gains in complex regions. PacBio developed the resource with researchers at the University of Washington, the University of Utah and other institutions, and is making the data and pipelines freely available through a public repository. The company says the Platinum Pedigree establishes a higher bar for evaluating sequencing workflows and for training AI models used in clinical and population genomics, potentially accelerating the adoption of long-read technologies in diagnostic and research settings.