“深层思维”公司说其AI模型可解码人类“暗基因组”

Core Insights - The article discusses Google's DeepMind's introduction of the AlphaGenome deep learning model, which decodes 98% of the "dark genome" crucial for health, potentially aiding in understanding genetic diseases, improving genetic testing, and informing new therapies [1][2]. Group 1: Model Capabilities - AlphaGenome can predict the function of DNA sequences up to 1 million base pairs long, breaking traditional trade-offs between sequence length and prediction accuracy [1]. - The model was trained using human and mouse genomes, learning how DNA sequences influence various biological processes [1]. - It can predict the impact of single nucleotide changes in the genetic code, as well as the effects of the dark genome on gene expression and splicing [1]. Group 2: Performance Evaluation - The research team conducted 26 benchmark tests, with AlphaGenome achieving or exceeding the performance of existing state-of-the-art models in 25 tasks [2]. - The tool has been made available for non-commercial use since last year, with 3,000 scientists having utilized it [2]. - While the development team acknowledges that the model is not perfect, it has been described by some researchers as a remarkable achievement and an important milestone [2].