HuggingFaceBio/traitgym

该数据集是对songlab/TraitGym（Benegas, Eraslan & Song, bioRxiv 2025.02.11.637758）的重新打包，并增加了一个额外步骤：为每个变异预提取了以变异为中心的8,192 bp窗口（来自hg38参考基因组），以及用替代等位基因替换的相同窗口。变异、标签和匹配控制与原始songlab/TraitGym的_matched_9配置完全相同。配置包括：mendelian_traits（n = 3,380）：包含113个单基因孟德尔疾病（从OMIM整理）的338个推定因果非编码变异，以9:1的比例在染色体×后果×TSS距离上匹配gnomAD常见变异控制；complex_traits（n = 11,400）：包含来自83个多基因UK BioBank性状（统计精细映射PIP > 0.9）的1,140个推定因果非编码变异，以9:1的比例在染色体×后果×TSS距离×MAF×LD得分上匹配PIP < 0.01的控制。This dataset is a repackaging of songlab/TraitGym (Benegas, Eraslan & Song, bioRxiv 2025.02.11.637758), with an additional preprocessing step: pre-extracting for each variant two variant-centric 8,192 bp windows derived from the hg38 reference genome — one is the native reference window, and the other is an identical window with the alternative allele substituted. Variants, labels, and matched controls are exactly identical to those in the original _matched_9 configuration of songlab/TraitGym. The configuration includes: 1. mendelian_traits (n = 3,380): 338 putative causal non-coding variants associated with 113 monogenic Mendelian diseases curated from OMIM, matched to gnomAD common variant controls at a 9:1 ratio based on chromosome, consequence type, and TSS distance; 2. complex_traits (n = 11,400): 1,140 putative causal non-coding variants from 83 polygenic UK Biobank traits (statistical fine-mapping PIP > 0.9), matched to controls with PIP < 0.01 at a 9:1 ratio based on chromosome, consequence type, TSS distance, MAF, and LD score.