Accurate integration of multiple heterogeneous single-cell RNA-seq data sets by learning contrastive biological...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Integrating multiple heterogeneous single-cell RNA sequencing (scRNA-seq) datasets is vital for biological and medical research.
• Current methods struggle to effectively integrate diverse scRNA-seq datasets due to biological and technical variations.

Purpose of the Study:

• To introduce single-cell integration (scInt), a novel method for robustly integrating multiple scRNA-seq datasets.
• To provide a flexible approach for transferring knowledge between integrated reference and query datasets.
• To demonstrate the superior performance of scInt compared to existing integration methods.

Main Methods:

• scInt utilizes accurate cell-cell similarity construction.
• Employs unified contrastive biological variation learning for multi-dataset integration.
• Enables knowledge transfer from reference to query datasets.

Main Results:

• scInt outperforms 10 other state-of-the-art integration methods on both simulated and real scRNA-seq data.
• Demonstrates superior performance, especially with complex experimental designs.
• Successfully integrates developmental trajectories from different stages in mouse tracheal epithelial data.
• Identifies functionally distinct, condition-specific cell subpopulations in heterogeneous samples.

Conclusions:

• scInt offers a flexible and effective solution for integrating diverse scRNA-seq data.
• The method accurately handles biological and technical variations, improving data integration.
• scInt advances the analysis of complex biological systems and condition-specific cell populations.