scDART

scDART integrates single-cell RNA sequencing (scRNA-seq) and single-cell Assay for Transposase-Accessible Chromatin using sequencing (scATAC-seq) data into a shared low-dimensional latent space to preserve cell trajectory structures across continuous cell populations.

Key Features:

• Deep learning framework: Implements a deep learning model to learn representations from scRNA-seq and scATAC-seq data.
• Shared low-dimensional latent space: Embeds both modalities into a unified low-dimensional latent space for joint analysis.
• Simultaneous cross-modality relationship learning: Learns relationships between scRNA-seq and scATAC-seq modalities concurrently rather than sequentially.
• Avoids pre-defined gene activity matrices (GAMs): Does not rely on pre-defined GAMs to convert scATAC-seq into RNA-like features, reducing dataset-specific bias from GAMs.
• Preservation of trajectory structures: Maintains cell trajectory structures across continuous cell populations within the integrated representation.
• Scalability and batch adaptability: Scales across datasets and adapts across different batches of data to support multi-batch integration.
• Facilitates trajectory inference: Produces integrated embeddings conducive to downstream trajectory inference analyses.

Scientific Applications:

• Cellular differentiation analysis: Enables study of differentiation processes by integrating gene expression and chromatin accessibility across cells.
• Lineage relationship reconstruction: Supports inference of lineage relationships through preserved trajectory structures in the latent space.
• Dynamic regulation studies: Facilitates analysis of dynamic changes in gene expression and chromatin accessibility across cell states.
• Cell state transition and developmental pathway analysis: Enables integrated investigation of cell state transitions and developmental pathways using combined scRNA-seq and scATAC-seq data.

Methodology:

Applies a deep learning model that simultaneously learns cross-modality relationships and embeds scRNA-seq and scATAC-seq into a shared low-dimensional latent space without relying on pre-defined gene activity matrices (GAMs), while preserving trajectory structures and adapting across batches.