PARC

PARC clusters ultralarge single-cell datasets by constructing and partitioning high-dimensional cell similarity graphs to identify cellular populations, including rare cell types.

Key Features:

• Scalable Graph-Based Clustering: Processes datasets containing more than one million single cells without sub-sampling.
• Rare Population Detection: Identifies rare cell populations within large high-dimensional single-cell datasets.
• High-Dimensional Data Support: Clusters data derived from single-cell RNA sequencing, mass cytometry, and high-content imaging experiments.

Scientific Applications:

• Single-Cell Transcriptomics: Clusters single-cell RNA-seq gene expression profiles to identify cellular subpopulations.
• Mass Cytometry Analysis: Analyzes high-dimensional phenotypic data generated by single-cell mass cytometry experiments.
• Single-Cell Imaging Analysis: Processes large-scale imaging-derived single-cell biophysical datasets.

Methodology:

PARC constructs a hierarchical nearest-neighbor graph using Hierarchical Navigable Small World (HNSW), applies data-driven graph pruning, and performs clustering using the Leiden community-detection algorithm.