densvis

densvis implements density-preserving modifications to t-SNE and UMAP for non-linear dimensionality reduction of high-dimensional sequencing data.

Key Features:

• den-SNE: Implements a density-preserving modification of t-SNE that incorporates local density of points in the original high-dimensional space into the optimisation process.
• densMAP: Implements a density-preserving modification of UMAP that incorporates local density of points in the original high-dimensional space into the optimisation process.
• Density preservation: Produces lower-dimensional embeddings that accurately represent the heterogeneity of the original high-dimensional space, enabling identification of homogeneous and heterogeneous cell states.

Scientific Applications:

• Single-cell RNA-seq visualization: Summarises complex high-dimensional single-cell RNA sequencing data using lower dimensional representations.
• Transcriptional state identification: Enables visualisation of discrete transcriptional states in reduced-dimensional embeddings.
• Developmental trajectory analysis: Enables visualisation of continuous trajectories, such as those in early development.
• Heterogeneity assessment: Enables more accurate visual interpretation of high-dimensional datasets by preserving density information that reflects transcriptional heterogeneity.

Methodology:

The method includes a density-preserving term in the optimisation process that considers the local density of points in the original high-dimensional space, as described by Narayan et al. (2020).