DeepMind Creates AI Map of Human Genome: Direct Search for 9 Billion DNA Variations

Beating AI Newsflash: Google DeepMind has released the AlphaGenome Atlas, which precomputes approximately 9 billion possible single-letter DNA variations in the human genome and organizes them into a searchable database. The entire dataset is about 1PB in size, over 30 times larger than the AlphaFold Database. The human genome consists of roughly 3 billion DNA letters, with each position potentially substituting with 3 other options, resulting in approximately 9 billion single-letter variations. Previously, researchers had to use AlphaGenome to analyze variations individually, but now they can directly search for results in the Atlas without rerunning the model each time. The Atlas also provides an AVI score for each variation, combining predictions from AlphaGenome and AlphaMissense to help researchers identify the variations most likely to have an impact. It covers approximately 2% of protein-coding regions as well as the more challenging 98% non-coding regions. Research teams have already used it to identify previously overlooked rare disease-related variations and confirmed through experiments that one of these variations indeed causes RNA splicing abnormalities. [Original Link]