Introduction

• Single-cell technology development
  • larger and larger numbers of cells assayed per experiment
  • the scalability leveraging on-disk data processing (avoid “out-of-memory”)
• Genomic Data Structure (GDS) files
  • an alternative to HDF5 & TileDB
  • hierarchical structure to store array-oriented data sets
  • compress & decompress data internally
  • out-of-memory data storage and manipulation in R
• SCArray
  • applies GDS to single-cell data manipulation & analysis
  • utilizes DelayedArray & SingleCellExperiment in Bioconductor
  • reuse existing analysis packages efficiently (e.g., scater) via internal “seed-aware” functions

Workflow & Data Structure

Key Functions in SCArray

Example: Small-size Dataset

Example: Large-size Dataset (1.3M mouse brain cells)

Discussion

• SCArray
  • under development
  • GDS as a file-based representation
    • a DelayedArray backend
    • for large-scale single-cell data storage & manipulation
  • leverage existing analysis R packages (e.g., scater)
    • via DelayedMatrix
• Plans
  • further integrate with Bioconductor infrastructure
    • GDSArray, DelayedMatrixStats, …
  • reimplement some memory-intensive algorithms

Acknowledgements

• Genomics Research Center (GRC), AbbVie
  • Astrid Wachter
  • Priyanka Vijay
  • Yating(Claire) Chai
  • Zheng Zha
• Bioconductor
  • Qian Liu (Roswell Park Comprehensive Cancer Center)
• National Center for Supercomputing Applications (NCSA)
  • University of Illinois at Urbana-Champaign (UIUC)