Complex hierarchical structures in single-cell genomics data unveiled by deep hyperbolic manifold learning.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell sequencing techniques have advanced cell development analysis.
• Existing methods often use Euclidean space, distorting hierarchical cell differentiation structures.
• Hyperbolic space methods show promise but are limited and not optimized for sparse single-cell data.

Purpose of the Study:

• To develop a novel deep learning approach for visualizing complex hierarchical structures in single-cell RNA-seq data.
• To address limitations of existing hyperbolic space methods for sparse count data.
• To improve the accuracy of cell development trajectory analysis.

Main Methods:

• Proposed scDHMap, a model-based deep learning method operating in low-dimensional hyperbolic space.
• Evaluated scDHMap using extensive simulations and real experimental datasets.
• Extended scDHMap for visualizing single-cell ATAC-seq data.

Main Results:

• scDHMap outperforms existing dimensionality-reduction methods for single-cell RNA-seq data.
• Demonstrated improved performance in revealing trajectory branches and batch correction.
• Showcased effective denoising of count matrices with high dropout rates.

Conclusions:

• scDHMap provides a superior approach for visualizing and analyzing complex hierarchical structures in single-cell data.
• The method enhances common analytical tasks for single-cell RNA-seq and ATAC-seq data.
• scDHMap offers a robust solution for handling sparse and noisy single-cell count data.