GATAC: GPU-Accelerated scATACseq Analysis#
GATAC is a GPU-accelerated toolkit for end-to-end ATAC-seq data processing. Starting from raw fragment files, it produces analysis-ready sparse matrices (tile, peak, and gene activity) that integrate directly with the scverse ecosystem [VBH+23] — AnnData, Scanpy [WAT18], and SnapATAC2 [ZZAR24]. Computations are offloaded to the GPU via RAPIDS cuDF, CuPy, and cuML, delivering 10–50× speedups over CPU-only workflows on typical single-cell datasets.
GATAC uses Apache Parquet as its staging format. Columnar layout and built-in compression let RAPIDS cuDF stream data directly into GPU memory with zero CPU round-trips — enabling aggregation over datasets that far exceed available GPU RAM.
GATAC reproduces the core operations of established tools like SnapATAC2 [ZZAR24], ArchR [GCP+21], MACS3 [ZLM+08], and chromVAR [SWBG17] within a unified framework. It produces standard AnnData objects fully compatible with the scverse ecosystem.
What can GATAC do?#
Convert raw TSV.GZ fragment files to columnar Parquet for fast GPU streaming, with optional barcode prefixing for multi-sample projects.
GPU QC metrics (TSSe, unique fragments, duplicate & mito fraction) plus threshold-based barcode filtering with a query on the metrics.
Flag doublet / multiplet cells via the AMULET Poisson overlap test.
Bin the genome into fixed-size tiles and produce a sparse cell × bin count matrix compatible with SnapATAC2.
Score gene activity from paired insertion counts over promoter + gene body regions, or compute ArchR-style distance-weighted gene scores — all from a GTF annotation.
GPU-accelerated selection of the most accessible genomic features, from a matrix or from the tile statistics, so that a cohort’s tile matrix can be built with the selected tiles only.
Embed cells for UMAP and clustering via spectral decomposition or ArchR-style (iterative) LSI with projection of new cells.
Call peaks per cell-type group, merge overlapping peaks across groups, and build a cell × peak count matrix.
Scan peaks for TF binding motifs (MEME format), run motif enrichment tests, and compute chromVAR deviation scores.