Semi-supervised disentangled representation learning for single-cell RNA sequencing data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) data is high-dimensional and complex.
• Current analysis methods often result in entangled low-dimensional representations, hindering biological interpretation.
• Existing disentanglement methods require extensive labeled data or are limited to few categories.

Purpose of the Study:

• To develop a semi-supervised method for learning disentangled representations from scRNA-seq data.
• To enhance the interpretability of scRNA-seq data by separating biological and technical factors.
• To address the limitations of existing methods regarding data annotation requirements and factor complexity.

Main Methods:

• Proposes SCDRL (Semi-Supervised Disentangled Representation Learning).
• Utilizes gene expression profiles and a small proportion of labeled samples.
• Learns representations that disentangle batch effects, cell types, and other biological signals.

Main Results:

• SCDRL effectively separates batch effects and biological signals.
• The method generalizes to complex settings with over 10 cell types.
• Demonstrates superior performance compared to existing methods on simulated and real-world datasets, even with only 5% labeled data.

Conclusions:

• SCDRL offers an effective solution for learning disentangled representations in scRNA-seq data.
• The method enhances data interpretability and overcomes limitations of current approaches.
• SCDRL shows promise for broader applications in single-cell data analysis with limited annotations.