openTSNE

openTSNE implements scalable and extensible t-distributed stochastic neighbor embedding (t-SNE) for high-dimensional data visualization, enabling efficient dimensionality reduction and mapping of new samples for biological datasets.

Key Features:

• Modular and Extensible Design: Provides a modular architecture that allows integration of extensions to the core t-SNE algorithm.
• Scalability and Speed: Achieves speeds that surpass popular implementations, including scikit-learn, and can handle millions of data points.
• Mapping New Data: Supports mapping new data points onto existing embeddings to integrate new samples and facilitate addressing batch effects.
• Improved Global Alignment: Incorporates enhancements to improve the global alignment of visualizations for more accurate and interpretable plots.

Scientific Applications:

• High-dimensional biological data visualization: Enables visualization and exploratory analysis of high-dimensional biological datasets to identify patterns and clusters.
• Batch-effect mitigation and dataset integration: Allows integration of new samples into existing embeddings to assist in comparing datasets and mitigating batch effects.

Methodology:

Implements the t-SNE algorithm with explicit improvements for scalability, speed, and global alignment, using a modular design that supports extension and mapping of new data.