A neural network-based method for exhaustive cell label assignment using single cell RNA-seq data.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) allows for high-resolution transcriptome analysis of complex tissues.
• Accurate cell type annotation is a critical, yet challenging, step in scRNA-seq data analysis.
• Traditional methods are time-consuming, and existing supervised methods often label cells as 'unassigned' even when unknown cell types are not expected.

Purpose of the Study:

• To develop a novel neural network-based cell annotation method, NeuCA (Neural network-based Cell Annotation), for scRNA-seq data.
• To improve cell annotation accuracy by utilizing the hierarchical structure of cell types.
• To provide a robust tool for analyzing scRNA-seq data from well-studied tissues.

Main Methods:

• Developed NeuCA, a deep learning model for supervised cell type assignment in scRNA-seq data.
• Incorporated hierarchical cell type information into the neural network architecture.
• Evaluated NeuCA's performance against existing methods on multiple real-world scRNA-seq datasets.

Main Results:

• NeuCA achieved higher cell annotation accuracy compared to existing methods, particularly for datasets with closely related cell types.
• Demonstrated stable and reliable performance across intra-study, inter-study, and cross-condition scRNA-seq data.
• Successfully addressed the issue of excessive 'unassigned' cells in well-studied tissues.

Conclusions:

• NeuCA offers a significant advancement in scRNA-seq data annotation, enhancing accuracy and reliability.
• The method's ability to utilize cell type hierarchies makes it particularly valuable for complex biological systems.
• NeuCA is available as an R/Bioconductor package, facilitating its adoption in the research community.