scTransSort: Transformers for Intelligent Annotation of Cell Types by Gene Embeddings.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding complex tissue composition.
• Manual cell type annotation from scRNA-seq data is time-consuming, irreproducible, and challenging due to data sparsity and scale.
• Existing methods struggle with the increasing volume and complexity of scRNA-seq datasets.

Purpose of the Study:

• To develop an automated and accurate cell type annotation method for scRNA-seq data.
• To address the limitations of manual annotation and data sparsity in single-cell analysis.
• To introduce scTransSort, a transformer-based approach for robust cell classification.

Main Methods:

• Applied transformer architecture to single-cell classification using scRNA-seq data.
• Developed scTransSort, a method pretrained on single-cell transcriptomics data.
• Represented genes as expression embedding blocks to reduce data sparsity and computational complexity.
• Implemented intelligent information extraction for automatic feature identification from unordered data.

Main Results:

• scTransSort demonstrated high accuracy and performance in cell type identification.
• The method effectively reduced data sparsity and computational complexity.
• Experiments on human and mouse tissues confirmed scTransSort's robustness and generalization ability.
• Achieved automated cell type identification without manual feature labeling.

Conclusions:

• scTransSort offers a powerful and efficient solution for cell type annotation in scRNA-seq data.
• The transformer-based approach significantly improves upon existing methods for single-cell classification.
• This method enhances the scalability and reliability of analyzing complex biological systems using transcriptomics.