arcinstitute/evoeval
收藏资源简介:
evoeval 是一个基因组变异效应评估的数据集集合,包含多个子数据集,每个子集以GRCh38坐标作为键,存储为Parquet格式的变异表,并带有标签和注释列。数据集用于评估变异效应预测模型,涵盖不同任务:ClinVar用于区分致病/可能致病与良性/可能良性变异;SpliceVarDB用于区分剪接改变变异与正常变异;GPN-star benchmarks包括COSMIC体细胞癌症突变与中性变异的区分、OMIM疾病相关变异与常见变异的区分,以及gnomAD平衡基准测试;TraitGym包括GWAS精细映射的非编码变异和孟德尔疾病调控变异的因果变异与匹配对照的区分。每个子集都有标准列模式(如染色体、位置、参考等位基因、替代等位基因)和特定注释列,类平衡和分层信息详细提供。数据集旨在支持基因组学中的机器学习模型评估。
evoeval is a collection of datasets for genomic variant effect evaluation, comprising multiple sub-datasets. Each sub-dataset is keyed by GRCh38 coordinates, stored as Parquet-formatted variant tables, and includes label and annotation columns. This collection is intended to assess variant effect prediction models, covering a range of tasks: ClinVar for distinguishing pathogenic/likely pathogenic variants from benign/likely benign variants; SpliceVarDB for differentiating splice-altering variants from normal variants; GPN-star benchmarks, which encompass distinguishing somatic cancer mutations from neutral mutations in COSMIC, differentiating disease-associated variants from common variants in OMIM, and the gnomAD balancing benchmark; TraitGym, which involves distinguishing causal variants from matched controls for non-coding variants in GWAS fine-mapping and regulatory variants associated with Mendelian diseases. Each sub-dataset adheres to a standard column schema (e.g., chromosome, position, reference allele, alternative allele) with specialized annotation columns, and detailed class balance and stratification information is provided. This dataset collection aims to support the evaluation of machine learning models in genomics research.




