Google’s AlphaGenome wants to do for DNA what AlphaFold did for proteins

Google’s new deep learning model, AlphaGenome, can predict the impact of small changes in DNA sequences up to one million base pairs, especially focusing on non-coding DNA.
AlphaGenome can handle mutations across whole DNA sequences, making it superior to its predecessor AlphaMissense, which only focused on coding regions.
Trained on data from human and mouse genomes, AlphaGenome can predict gene expression, DNA accessibility, and gene splicing.
The model has matched or outperformed other state-of-the-art models in 25 out of 26 tasks predicting genetic variation effects, including simulated mutations linked to leukaemia.
It aims to enhance fundamental genome knowledge, aid in understanding rare diseases and cancers, and help design new DNA sequences for treating specific conditions.
Though a powerful tool, AlphaGenome has limitations, struggling to predict distant genetic alterations and can only make predictions about human and mouse DNA sequences.
Since its release, it has been used by nearly 3000 scientists in 160 countries, assisting in over many requests daily.
Google DeepMind hopes AlphaGenome will advance our understanding of the genome, disease biology, and lead to new treatments.

Google’s AlphaGenome wants to do for DNA what AlphaFold did for proteins | Research