scTrans: Sparse attention powers fast and accurate cell type annotation in single-cell RNA-seq data.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Cell type annotation is vital for single-cell RNA sequencing (scRNA-seq) data analysis, enabling biological insights.
• Current methods often use highly variable genes, risking information loss and limiting adaptability to new datasets.

Purpose of the Study:

• To develop an advanced computational model for accurate and efficient cell type annotation in scRNA-seq data.
• To address information loss and generalization issues inherent in existing annotation tools.

Main Methods:

• Introduced scTrans, a Transformer-based model leveraging sparse attention mechanisms.
• scTrans processes all non-zero genes, reducing dimensionality while preserving crucial information.

Main Results:

• Validated scTrans on 31 tissues from the Mouse Cell Atlas, demonstrating high speed and accuracy.
• scTrans efficiently annotated large datasets (nearly one million cells) with limited computational resources.
• The model exhibited strong generalization, accurately annotating novel datasets and producing high-quality latent representations for downstream analysis.

Conclusions:

• scTrans offers an effective solution for cell type annotation in scRNA-seq data, overcoming limitations of previous methods.
• The model's ability to utilize all non-zero genes enhances generalization and adaptability for biological research.