Visualizing hierarchies in scRNA-seq data using a density tree-biased autoencoder.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution insights into cellular development.
• Cellular differentiation often follows a hierarchical, tree-like progression in gene expression space.
• Representing this tree structure in 2D is crucial for biological interpretation and analysis.

Purpose of the Study:

• To develop an approach for inferring meaningful tree structures from high-dimensional scRNA-seq data.
• To create a visualization method that preserves the inferred tree topology.
• To introduce a novel dimensionality reduction technique biased towards this tree structure.

Main Methods:

• Vector quantization of scRNA-seq data.
• Density-based maximum spanning tree for tree structure extraction.
• Density-tree biased autoencoder (DTAE) for dimensionality reduction and visualization.

Main Results:

• The density-based maximum spanning tree effectively captures biological information.
• DTAE successfully emphasizes the data's tree structure in low-dimensional space.
• The proposed method demonstrates superior performance compared to existing dimension reduction techniques, validated on real and synthetic datasets.

Conclusions:

• The developed approach accurately identifies and visualizes hierarchical structures in scRNA-seq data.
• DTAE offers a powerful tool for exploring cellular differentiation trajectories.
• This method enhances the biological interpretation of complex single-cell genomics data.