Single-cell assignment using multiple-adversarial domain adaptation network with large-scale references.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates vast datasets for cell population characterization.
• Accurate cell-type annotation is hindered by inconsistent public references, batch effects, and identification of rare cell types.

Purpose of the Study:

• To develop an integrative and automatic framework, SELINA (single-cell identity navigator), for robust human cell-type annotation.
• To create a comprehensive and uniform reference atlas for scRNA-seq data.

Main Methods:

• SELINA utilizes a pre-curated reference atlas of 1.7 million cells across 230 human cell types.
• Employs a multiple-adversarial domain adaptation network to mitigate batch effects.
• Incorporates synthetic minority oversampling for rare cell type annotation and an autoencoder for data fitting.

Main Results:

• SELINA demonstrates robust and superior performance across diverse human tissues.
• Accurate annotation of cell types within various disease contexts was achieved.
• The framework successfully integrated and harmonized a large-scale reference atlas.

Conclusions:

• SELINA offers a comprehensive solution for human scRNA-seq data annotation.
• The framework is available as both Python and R packages, facilitating broader adoption.
• SELINA enhances the reliability and accuracy of cell-type identification in complex biological datasets.