Human-scATAC-Corpus

Human-scATAC-Corpus compiles over 5.4 million human scATAC-seq cells from 35 datasets across 37 tissues or cell lines into a unified database for single-cell epigenomic research, method benchmarking, and foundation model pretraining.

Key Features:

• Multi-format Data Distribution: Each dataset is provided in three aligned formats: cell-by-candidate cis-regulatory element matrices for cross-dataset integration, raw fragment files for flexible processing, and cell-by-peak matrices for within-dataset analyses.
• Rich Metadata: The resource includes comprehensive metadata spanning diverse biological contexts to enable varied downstream analytical scenarios.
• Integrated Foundation Model: EpiAgent, a foundation model pretrained on Human-scATAC-Corpus, provides online analyses including reference mapping, embedding extraction, and cell type annotation.

Scientific Applications:

• Method Benchmarking and Development: The standardized, multi-format resource enables systematic evaluation and comparison of computational methods across single-cell epigenomic tasks.
• Foundation Model Pretraining: The large-scale, unified dataset serves as training data for deep learning models in single-cell epigenomics.
• Cross-dataset Integration: The aligned cell-by-candidate cis-regulatory element matrices facilitate integrative analysis across heterogeneous scATAC-seq datasets.
• Gene Regulation Research: The chromatin accessibility profiles support investigation of epigenomic landscapes that govern gene regulation in diverse cell types.

Methodology:

The database construction addressed heterogeneous feature spaces and complex processing pipelines to create an ensemble resource; datasets were processed to produce three aligned output formats from raw scATAC-seq data.