百亿参数人类基因组通用基础模型发布

Core Insights - The "Genos" universal foundational model for the human genome, developed by BGI Life Sciences Institute and ZhiJiang Lab, was officially launched at the 20th International Genomics Conference in Hangzhou, Zhejiang [1][2] - The model integrates data from 636 high-quality human genomes globally, aiming to reduce data bias and better represent human genetic diversity [1] - Genos can predict RNA expression profiles in seconds based solely on DNA sequences, significantly enhancing the speed of bioinformatics analysis [2] Research and Development - The foundational model's related paper was published in the international journal "GigaScience," marking a significant advancement in genomic research [1] - Existing models in genomics typically rely on one or two reference genomes, limiting their ability to capture genetic diversity [1] Clinical Applications - In clinical applications, Genos achieved a 92% accuracy rate in interpreting pathogenic mutations, which can be further improved to 98.3% when combined with ZhiJiang Lab's 021 scientific foundational model [2] - The model is positioned as an efficient tool for clinical diagnostics, enhancing the accuracy of disease diagnosis [2] Open Source Initiative - To promote global research collaboration, Genos will be fully open-sourced, offering two versions with 12 billion and 100 billion parameters to cater to different application scenarios [2] - The model's weights, architecture details, and complete training processes have been publicly released [2]