Unsupervised cell functional annotation for single-cell RNA-seq.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Accurate cell type assignment is crucial for single-cell RNA sequencing (scRNA-seq) data analysis.
• Current methods often involve separate clustering and annotation steps, which can be affected by noise and reduce accuracy.

Purpose of the Study:

• To develop a novel neural network method, UNIFAN, for simultaneous clustering and annotation of scRNA-seq data.
• To improve the accuracy and ease of cell type assignment in scRNA-seq analysis.

Main Methods:

• UNIFAN utilizes a neural network architecture that integrates low-dimensional gene representations with cell-specific gene set activity scores.
• The method performs simultaneous clustering and annotation, leveraging known gene sets for improved performance.

Main Results:

• UNIFAN was applied to human and mouse scRNA-seq datasets from various organs, demonstrating superior performance compared to existing methods.
• The gene sets identified by UNIFAN provide strong evidence for cell type identity, facilitating easier annotation.

Conclusions:

• UNIFAN offers a significant advancement in scRNA-seq data analysis by enabling accurate and integrated cell clustering and annotation.
• The method's reliance on gene set knowledge enhances the interpretability and reliability of cell type assignments.