scSemiPLC: a semi-supervised learning framework for annotating single-cell RNA-Seq data by generating pseudo-labels...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution cellular heterogeneity insights.
• Manual cell annotation is time-consuming and struggles with large-scale scRNA-seq datasets.
• Automated cell annotation methods are crucial for efficient biological research.

Purpose of the Study:

• To develop an efficient and accurate semi-supervised cell annotation framework, scSemiPLC.
• To leverage unlabeled data effectively in the cell annotation process.
• To enhance the utilization of pseudo-labels generated through clustering.

Main Methods:

• Proposed scSemiPLC, a semi-supervised annotation training framework.
• Employed clustering to generate pseudo-labels for unlabeled data.
• Utilized consistency regularization and pseudo-label weighting to improve annotation.

Main Results:

• scSemiPLC demonstrated superior annotation accuracy and stability compared to existing methods.
• The framework effectively extracts biologically meaningful representations from scRNA-seq data.
• scSemiPLC showed robustness across varying numbers of cell labels.

Conclusions:

• scSemiPLC offers a novel and effective approach for semi-supervised cell annotation.
• The method significantly outperforms classical automatic and mainstream semi-supervised techniques.
• This framework advances automated cell annotation in the field of single-cell genomics.