CeLLTra: aligning cell names with gene expression via a pathway-informed transformer.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-Seq) provides high-resolution gene expression data for cell type identification.
• Traditional cell annotation methods face limitations due to data quality and reference dataset completeness.

Purpose of the Study:

• To develop a novel framework, CeLLTra, for accurate cell type annotation using scRNA-Seq data.
• To enhance cell type prediction by integrating biological pathway information and advanced language models.

Main Methods:

• Developed CeLLTra, a contrastive learning framework utilizing a Transformer model.
• Integrated biological pathway information by grouping genes into super tokens.
• Combined the pathway-informed Transformer with a pretrained domain-specific language model for annotation.

Main Results:

• CeLLTra achieved superior performance in supervised and zero-shot cell-type prediction on a large human scRNA-Seq dataset.
• The framework demonstrated strong generalization to external datasets, improving clustering.
• Enabled enhanced characterization of cancerous cell states in tumor-infiltrating myeloid cells.

Conclusions:

• CeLLTra offers a robust and accurate solution for cell type annotation from scRNA-Seq data.
• The integration of pathway information and language models significantly advances cell annotation capabilities.
• CeLLTra has implications for cancer research, particularly in understanding tumor microenvironments.