On the evening of June 24, Google DeepMind launched a new AI model **AlphaGenome**, specifically designed for DNA sequence analysis and prediction of variant effects.

This general-purpose model can process DNA sequences up to 1 million base pairs, with prediction accuracy at the level of individual base pairs, marking a significant breakthrough in genomic research. AlphaGenome can directly predict the location and expression levels of RNA splicing sites from DNA sequences, which is of great significance for studying rare genetic diseases caused by RNA splicing errors, such as spinal muscular atrophy.

AlphaGenome provides a "panoramic view" of predictions, comprehensively covering various molecular characteristics such as gene expression, RNA splicing, chromatin features, and DNA-binding proteins, revealing the complex landscape of gene regulation. In 22 out of 24 DNA sequence prediction tasks, it surpassed the best specialized models; in 24 out of 26 variant regulatory effect predictions, it reached or exceeded top-level performance, demonstrating unparalleled capabilities.

Cell

By integrating large-scale public data, AlphaGenome provides researchers with tools to gain deep insights into genome function and disease mechanisms, helping to discover new therapeutic targets. Currently, the model is freely available via API for non-commercial research, and it is expected to promote the rapid development of rare genetic disease research and personalized medicine in the future.