scDART: integrating unmatched scRNA-seq and scATAC-seq data and learning cross-modality relationship simultaneously.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Integrating single-cell RNA sequencing (scRNA-seq) and single-cell Assay for Transposase-Accessible Chromatin sequencing (scATAC-seq) data from different batches presents significant challenges.
• Current methods often rely on pre-defined gene activity matrices, which can be low-quality and fail to capture dataset-specific cross-modal relationships.

Purpose of the Study:

• To develop a novel deep learning framework, scDART, for robust integration of scRNA-seq and scATAC-seq data.
• To simultaneously learn cross-modal relationships between gene expression and chromatin accessibility.
• To enable trajectory inference on integrated single-cell data.

Main Methods:

• Proposed scDART, a deep learning framework designed for multimodal single-cell data integration.
• Developed methods to learn simultaneous cross-modal relationships without relying on pre-defined gene activity matrices.
• Validated the framework's ability to preserve cell trajectories in continuous cell populations.

Main Results:

• scDART successfully integrates scRNA-seq and scATAC-seq data, overcoming limitations of existing methods.
• The framework learns accurate dataset-specific cross-modal relationships.
• scDART effectively preserves cell trajectories, enabling downstream trajectory inference on integrated data.

Conclusions:

• scDART offers a powerful new approach for integrating scRNA-seq and scATAC-seq data.
• The framework's ability to learn cross-modal relationships and preserve cell trajectories enhances single-cell data analysis.
• scDART facilitates advanced analyses like trajectory inference on multimodal single-cell datasets.