.png){kind=link}
An international research team from Google DeepMind presents in Nature of 28 January a new model capable of predicting the specific regulatory functions of DNA sequences: AlphaGenome. The model has been trained on DNA from mice and humans, and the researchers are enthusiastic. They claim it is currently one of the best models for predicting variations in DNA.
AlphaGenome can analyse regulatory DNA sequences of up to 1 million bases. Whilst other models such as SpliceAI or ProCapNet can also predict DNA functions, AlphaGenome is the first to do so for such large stretches of DNA at once. This is indeed necessary, as AlphaGenome analyses stretches of non-coding DNA, which accounts for up to 98 per cent of our genome.
In some cases, AlphaGenome can determine to the level of a single base whether a given DNA sequence is likely to be involved in switching a gene on or off, in RNA splicing, or in making that stretch of DNA more accessible. Researchers hope that AlphaGenome will thus become a valuable tool in unravelling the functions of the largest part of the genome.