The art of using t-SNE for single-cell transcriptomics.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell transcriptomics generates large-scale RNA expression data.
• Dimensionality reduction, often using t-distributed stochastic neighbour embedding (t-SNE), is crucial for visualizing this high-dimensional data.
• Naive t-SNE applications can inaccurately represent global data structure.

Purpose of the Study:

• To develop a protocol for creating more faithful t-SNE visualizations.
• To address the shortcomings of standard t-SNE in representing global data structure.
• To improve the accuracy of dimensionality reduction for single-cell RNA sequencing data.

Main Methods:

• Protocol development incorporating PCA initialization, high learning rates, and multi-scale similarity kernels.
• Application of exaggeration and downsampling-based initialization for very large datasets.
• Validation using published single-cell RNA sequencing datasets.

Main Results:

• The developed protocol yields more faithful t-SNE visualizations compared to naive applications.
• Improved representation of both local and global data structures in high-dimensional single-cell data.
• Demonstrated superiority on published single-cell RNA-seq datasets.

Conclusions:

• The proposed protocol effectively circumvents common pitfalls in t-SNE application.
• This method enhances the accuracy and reliability of t-SNE for single-cell transcriptomics data visualization.
• The protocol offers a superior approach for analyzing and interpreting large single-cell datasets.