A versatile and scalable single-cell data integration algorithm based on domain-adversarial and variational...

Area of Science:

• Single-cell genomics
• Computational biology
• Bioinformatics

Background:

• Single-cell technologies offer high-resolution profiling of transcriptomic, epigenetic, and spatial data in complex tissues.
• Integrating diverse single-cell datasets is crucial for understanding cellular heterogeneity and composition.
• Existing methods face challenges in harmonizing data across different samples, technologies, and modalities.

Purpose of the Study:

• To develop a unified computational strategy for integrating multiple single-cell datasets.
• To enable the integration of paired single-cell ATAC and transcriptome profiles.
• To create a scalable and efficient tool for large-scale single-cell data analysis.

Main Methods:

• Domain-adversarial and variational approximation (DAVAE) framework.
• Mini-batch stochastic gradient descent for scalability.
• GPU acceleration for computational efficiency.

Main Results:

• DAVAE successfully integrated single-cell datasets across samples, technologies, and modalities.
• Demonstrated effectiveness in batch-effect removal and transfer learning.
• Achieved accurate cell-type predictions on multiple real-world datasets.

Conclusions:

• DAVAE provides a robust and scalable solution for multi-modal single-cell data integration.
• The method enhances biological insights by harmonizing diverse datasets.
• The 'scbean' toolkit package offers accessible implementation for researchers.