scDHMap

scDHMap maps single-cell genomics data into a low-dimensional hyperbolic manifold to capture hierarchical cell differentiation pathways and enable visualization and analysis of scRNA-seq and single-cell ATAC-seq data.

Key Features:

• Hyperbolic Manifold Learning: Operates in low-dimensional hyperbolic space to represent hierarchical, tree-like and branched structures without distortion.
• Deep Learning Integration: Implements a model-based deep learning framework to process sparse single-cell count data from scRNA-seq for dimensionality reduction and visualization.
• Optimization for Sparse Data: Handles the sparsity and high dropout rates typical of scRNA-seq count matrices.
• Denoising: Performs denoising of count matrices to improve data quality for downstream analyses.
• Batch Correction: Corrects batch effects across multi-experiment datasets.
• Extension to single-cell ATAC-seq: Applies the same hyperbolic manifold visualization approach to single-cell Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) data.

Scientific Applications:

• Trajectory Analysis: Reveals trajectory branches and cell development pathways to study differentiation processes.
• Batch Correction: Ensures consistency across multi-experiment datasets by correcting batch effects.
• Denoising Capabilities: Reduces dropout-related noise in count matrices to improve downstream inference.
• Visualization of scATAC-seq Data: Extends visualization and hierarchical representation to single-cell ATAC-seq datasets.

Methodology:

Combines a model-based deep learning approach with hyperbolic manifold learning for dimensionality reduction that preserves hierarchical relationships in single-cell data; validated using simulations and real-world experiments.