Structure-preserving visualization for single-cell RNA-Seq profiles using deep manifold transformation with...

Area of Science:

• Computational Biology
• Bioinformatics
• Data Visualization

Background:

• Dimensionality reduction and visualization are crucial for interpreting complex biological datasets like single-cell RNA sequencing (scRNA-seq).
• Existing methods often fail to unify requirements for preserving data structure, handling batch effects, and applicability across diverse biological scenarios.
• A versatile visualization tool is needed for tasks such as cell clustering and trajectory inference.

Purpose of the Study:

• To introduce Deep Visualization (DV), a novel, unified framework for dimensionality reduction and data visualization.
• To develop a method capable of preserving inherent data structures and effectively correcting batch effects.
• To create a versatile tool applicable to various biological data analysis tasks and scales.

Main Methods:

• DV embeds data into 2D or 3D space using Euclidean or hyperbolic metrics based on task type (static or dynamic).
• The method learns a structure graph to represent sample relationships, ensuring geometric structure preservation.
• DV corrects batch effects and transforms data into visualization space in an end-to-end manner.

Main Results:

• DV demonstrated strong performance across nine complex tissue datasets from human and animal studies.
• The method successfully identified complex cellular relationships and uncovered temporal trajectories.
• DV effectively addressed challenges posed by complex batch factors in scRNA-seq data.

Conclusions:

• Deep Visualization (DV) provides a competitive and unified approach for biological data visualization.
• The framework's ability to preserve structure and handle batch effects makes it suitable for diverse applications.
• Preliminary results suggest potential for pre-training DV models for efficient visualization of new datasets.