Overview
- The model processes stretches of DNA up to about one million bases to reveal long‑range interactions between distant regulatory switches and genes.
- It predicts eleven molecular features, including gene boundaries, splicing, RNA output, chromatin accessibility, 3D proximity, and protein binding across diverse cell types.
- In tests focused on variant effects, it matched or outperformed leading specialized tools in 25 of 26 evaluations and often delivered broader accuracy than existing methods.
- Scientists expect it to guide hypothesis generation and experiment planning, but they caution that its precision is not yet sufficient for clinical interpretation of specific variants.
- DeepMind offers free API access for non‑commercial research and says it plans a full release, as experts flag concerns about dependence on private infrastructure, access, and reproducibility.