opt-SNE

opt-SNE optimizes t-distributed stochastic neighbor embedding (t-SNE) parameter settings to improve visualization and analysis of large, high-dimensional single-cell cytometry and transcriptomics datasets.

Key Features:

• Automated Parameter Optimization: Automates selection of t-SNE parameters such as early exaggeration and gradient descent iterations using real-time evaluation of Kullback-Leibler divergence tailored to each dataset.
• Enhanced Visualization Quality: Calibrates early exaggeration and adjusts the gradient descent learning rate to produce more accurate representations of biological populations in cytometry and transcriptomics datasets.
• Efficiency in Computation Time: Reduces the need for empirical parameter tuning to decrease computation time for massive datasets, including datasets with millions of cells.
• Overcoming Hard-Coded Parameter Limitations: Dynamically adjusts parameters based on dataset-specific characteristics to avoid poorly resolved or misleading t-SNE embeddings.

Scientific Applications:

• Single-cell cytometry: Improves visualization and resolution of cellular populations in high-dimensional cytometry datasets.
• Single-cell transcriptomics: Enhances embedding clarity for large transcriptomics datasets to aid interpretation of cell populations.
• Analysis of cellular heterogeneity and dynamics: Facilitates extraction of insights about cellular heterogeneity and dynamics from large single-cell datasets.

Methodology:

Modifies the standard t-SNE algorithm and uses real-time evaluation of Kullback-Leibler divergence to guide adjustments of early exaggeration, gradient descent iterations, and the gradient descent learning rate.