In a preprint, researchers presented DeepD, a genome model that works with DNA sequences up to one million tokens while using less active computation. It predicts variant effects across species and generates controlled DNA, from core promoters to whole yeast chromosomes.
The model could help researchers interpret disease-linked DNA changes, including variants in BRCA1 and BRCA2, and test how regulatory DNA affects gene activity. It also offers a way to design promoters with chosen strengths and generate long yeast chromosomes or plant DNA segments while preserving gene order.
✦ Long-range DNA modeling could sharpen variant analysis and genome design.