A scalable sparse neural network framework for rare cell type annotation of single-cell transcriptome data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis is crucial for understanding cellular heterogeneity.
• Current automatic cell type annotation methods struggle with imbalanced datasets and identifying rare cell populations, leading to errors.

Purpose of the Study:

• To introduce scBalance, a novel framework for accurate and efficient automatic cell type annotation in scRNA-seq data.
• To address the limitations of existing methods in handling imbalanced datasets and discovering rare cell types.

Main Methods:

• Developed scBalance, an integrated sparse neural network framework.
• Incorporated adaptive weight sampling and dropout techniques for auto-annotation.
• Validated on 20 scRNA-seq datasets with varying scales and imbalance levels.

Main Results:

• scBalance significantly outperforms current methods in both intra- and inter-dataset annotation tasks.
• Demonstrated scalability in identifying rare cell types in million-level datasets, exemplified by bronchoalveolar cell landscape analysis.
• Achieved faster computation times compared to commonly used tools.

Conclusions:

• scBalance offers a superior solution for scRNA-seq auto-annotation, particularly for imbalanced data and rare cell identification.
• The framework's speed, scalability, and user-friendly Python-based platform make it a valuable tool for researchers.