qSNE

qSNE performs scalable t-distributed stochastic neighbor embedding (t-SNE) using quasi-Newton optimization and automatic perplexity tuning for dimensionality reduction of large-scale single-cell and mass cytometry datasets.

Key Features:

• Quasi-Newton Optimization: Employs a quasi-Newton optimizer with quadratic convergence to accelerate optimization relative to standard t-SNE.
• Automatic Perplexity Tuning: Incorporates an automatic perplexity optimizer that dynamically adjusts neighborhood scale during embedding.
• Scalability to Massive Datasets: Enables full-scale analysis of large datasets, including mass cytometry, without downsampling or interpolative approximations, preserving rare populations.
• Performance Improvements: Delivers substantially faster runtimes than existing t-SNE implementations for large single-cell datasets.

Scientific Applications:

• Single-cell and mass cytometry analysis: Enables visualization and exploratory analysis of single-cell datasets and mass cytometry to identify cellular populations.
• Genomics and proteomics dimensionality reduction: Applies dimensionality reduction in genomics and proteomics studies to reveal structure in high-dimensional measurements.
• Systems biology and disease research: Supports systems biology investigations to uncover cellular heterogeneity and potential disease mechanisms.

Methodology:

Applies t-distributed stochastic neighbor embedding (t-SNE) with a quasi-Newton optimizer and an automatic perplexity optimizer to compute embeddings without downsampling or interpolative approximations.