SAFAARI: Contrastive Adversarial Open-set Domain Adaptation for Single-cell Integration & Annotation.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell sequencing reveals cellular heterogeneity but faces challenges in data integration and batch effect correction.
• Existing methods struggle with large, complex datasets and multi-modal data integration.

Purpose of the Study:

• To introduce SAFAARI, a unified deep learning framework for cell annotation, batch correction, and multi-omics integration.
• To address limitations in current single-cell data analysis, including batch effects and domain shifts.

Main Methods:

• SAFAARI utilizes supervised contrastive learning and adversarial domain adaptation for domain-invariant embeddings.
• The framework enables label transfer across datasets and mitigates class imbalance for rare cell type detection.

Main Results:

• SAFAARI demonstrated robust performance in cell annotation, batch correction, and cross-omics integration across diverse datasets.
• The method outperformed existing approaches in identifying novel cell types and handling heterogeneous data.

Conclusions:

• SAFAARI offers a scalable, accurate, and flexible solution for single-cell analysis.
• The framework has broad applicability in biological and clinical research, facilitating deeper understanding of cellular heterogeneity.