scMAE: a masked autoencoder for single-cell RNA-seq clustering.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at the individual cell level.
• Clustering scRNA-seq data is crucial for identifying cell types and understanding cellular heterogeneity.
• Existing deep learning methods often fail to leverage gene correlations, limiting clustering accuracy.

Purpose of the Study:

• To develop a novel deep learning-based method for scRNA-seq data clustering.
• To improve cell subpopulation identification by effectively capturing gene-gene correlations.
• To enhance the detection of rare cell types within complex biological samples.

Main Methods:

• Proposed scMAE, a masked autoencoder-based approach for scRNA-seq data.
• scMAE reconstructs perturbed gene expression data to learn robust cell representations.
• A masking predictor within the autoencoder captures inter-gene relationships.

Main Results:

• scMAE demonstrated superior performance compared to state-of-the-art methods across 15 scRNA-seq datasets.
• The method effectively identified latent structures and dependencies within the gene expression data.
• scMAE successfully identified rare cell types, often missed by other clustering approaches.
• Biological validation confirmed the significance of the identified cell subpopulations.

Conclusions:

• scMAE offers an effective strategy for scRNA-seq data clustering by integrating gene correlation information.
• The method advances the identification of cell types and the exploration of cellular heterogeneity.
• scMAE shows promise for applications requiring sensitive detection of rare cell populations.