SamSPECTRAL

SamSPECTRAL applies spectral clustering augmented with information-preserving sampling and post-processing to identify biological cell populations in large, high-dimensional flow cytometry datasets.

Key Features:

• Information-Preserving Sampling: Implements a sampling technique that reduces dataset size while retaining critical information for downstream clustering.
• Post-Processing Stage: Refines initial clustering results to improve the precision of identified cell populations.
• Graph Construction and Conductance Weighting: Constructs a graph from data points and weights edges using conductance to represent relationship strengths.
• Spectral Clustering Integration: Applies spectral clustering on the constructed weighted graph to detect cluster structures via graph spectral properties.
• Scalability to High-Dimensional Large Datasets: Operates on high-dimensional data with the ability to handle datasets containing up to hundreds of thousands of points.

Scientific Applications:

• Flow Cytometry Data Analysis: Applied to flow cytometry datasets for identification and characterization of cellular populations.
• Identification of Complex Population Shapes: Detects non-elliptical cell populations and low-density populations adjacent to dense ones.
• Detection of Minor and Rare Subpopulations: Identifies minor subpopulations within major populations and rare cell populations.
• Immunology and Cancer Research: Supports studies in immunology and cancer research that require precise cell population analysis.

Methodology:

Information-preserving data sampling; graph construction from sampled points with edges weighted by conductance; spectral clustering on the weighted graph using eigenvalues and eigenvectors; and post-processing refinement of clusters.