ScaleSC

ScaleSC accelerates large-scale single-cell RNA-seq data analysis through GPU computing, delivering substantial speedups and improved scalability for datasets containing millions of cells and thousands of batches.

Key Features:

• GPU Acceleration: Leverages CuPy and CUDA to achieve over 20× speedup compared to CPU-based approaches.
• Large-Scale Scalability: Processes datasets of 10-20 million cells with over 1000 batches on a single A100 GPU by overcoming memory bottlenecks.
• Algorithmic Consistency: Resolves discrepancies between GPU and CPU algorithm implementations to ensure consistent results across computing platforms.
• Marker Gene Identification: Includes novel GPU-optimized implementations for identifying marker genes.
• Cluster Merging: Provides GPU-optimized tools for merging cell clusters.
• Scanpy Compatibility: Maintains similar syntax with Scanpy to support existing workflow integration.

Scientific Applications:

• Large-Scale Single-Cell Transcriptomics: Enables analysis of million-cell single-cell RNA-seq datasets that exceed the capacity of existing GPU-based tools such as Rapids-singlecell.
• Batch-Effect Handling: Supports integrated analysis of datasets comprising over 1000 batches.
• Cell Type Annotation: Facilitates marker gene-based identification and refinement of cell type clusters through GPU-optimized computational methods.

Methodology:

ScaleSC builds on Scanpy and Rapids-singlecell, with GPU acceleration implemented through CuPy and CUDA. GPU-optimized implementations were developed for core single-cell analysis steps, marker gene identification, and cluster merging, with explicit resolution of GPU-CPU algorithmic discrepancies to ensure consistency.