scRSSL: Residual semi-supervised learning with deep generative models to automatically identify cell types.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell sequencing (scRNA-seq) enables cellular heterogeneity studies.
• Cell type identification is crucial in single-cell transcriptomics.
• Existing methods struggle with high dimensionality, sparsity, and sample imbalance in scRNA-seq data.

Purpose of the Study:

• To develop a robust method for cell type recognition in challenging single-cell datasets.
• To address limitations of traditional cell type identification approaches.
• To leverage semi-supervised learning for accurate cell classification with limited labels.

Main Methods:

• Proposed a deep residual generation model based on semi-supervised learning (scRSSL).
• Integrated residual networks into semi-supervised generative models.
• Utilized residual neural networks for cell type inference and local feature extraction.
• Employed semi-supervised learning to manage sample imbalance and leverage limited cell labels.

Main Results:

• The scRSSL model demonstrates effective handling of high dimensionality, sparsity, and sample imbalance.
• Achieved automatic and accurate prediction of individual cell types, even with minimal labeled data.
• Experimental results show superior performance compared to existing cell type recognition methods.

Conclusions:

• scRSSL offers an advanced solution for cell type identification in single-cell transcriptomics.
• The model's semi-supervised approach enhances accuracy and robustness in complex datasets.
• This method provides a valuable tool for analyzing cellular heterogeneity.