gpn-star-scores
收藏资源简介:
GPN-Star genome-wide scores 是一个全基因组范围的变异效应预测数据集,包含经过校准的 GPN-Star 模型分数。该数据集覆盖八个不同的物种模型与基因组组装组合,包括人类(三种不同进化约束的 200M 参数模型)、秀丽隐杆线虫、黑腹果蝇、鸡、拟南芥和小鼠。数据以两种主要形式提供:Parquet 格式的原始高精度分数文件,以及为基因组浏览器可视化优化的 BigWig 轨道文件。Parquet 文件包含两种配置类型:entropy 配置提供染色体位置、参考碱基和校准后的熵分数;llr 配置额外包含替代碱基、校准后的对数似然比分数及其绝对值。数据集规模庞大(10B<n<100B),适用于基因组学功能注释、进化约束分析、变异致病性预测等任务。用户可通过 UCSC Genome Browser 直接可视化数据,或使用 Polars 等工具进行高效的大规模基因组区间查询和变异注释分析。数据集采用 Apache-2.0 许可证发布。
GPN-Star genome-wide scores is a genome-wide variant effect prediction dataset containing calibrated GPN-Star model scores. It covers eight different species model and genome assembly combinations, including humans (three 200M parameter models with different evolutionary constraints), Caenorhabditis elegans, Drosophila melanogaster, chicken, Arabidopsis thaliana, and mouse. The data is provided in two main formats: raw high-precision score files in Parquet format, and BigWig track files optimized for genome browser visualization. Parquet files include two configuration types: the entropy configuration provides chromosome positions, reference bases, and calibrated entropy scores; the llr configuration additionally includes alternative bases, calibrated log-likelihood ratio scores, and their absolute values. The dataset is large-scale (10B<n<100B) and suitable for tasks such as genomic functional annotation, evolutionary constraint analysis, and variant pathogenicity prediction. Users can directly visualize the data via the UCSC Genome Browser or use tools like Polars for efficient large-scale genomic interval queries and variant annotation analysis. The dataset is released under the Apache-2.0 license.




