scSSA: A clustering method for single cell RNA-seq data based on semi-supervised autoencoder.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell sequencing technologies enable high-throughput analysis of the genome, transcriptome, and epigenome at the individual cell level.
• Traditional methods struggle with single-cell RNA data due to dropout events and the curse of dimensionality, limiting the understanding of cellular heterogeneity.
• Existing data-driven clustering methods often fail to fully leverage available biological information.

Purpose of the Study:

• To develop an advanced clustering model for single-cell RNA sequencing (scRNA-seq) data.
• To overcome the challenges of dropout events and high dimensionality inherent in scRNA-seq data.
• To improve the accuracy and applicability of cell type identification in single-cell analyses.

Main Methods:

• Proposed scSSA, a clustering model integrating a semi-supervised autoencoder, Fast Independent Component Analysis (FastICA), and Gaussian mixture clustering.
• Employed a semi-supervised autoencoder for data imputation and denoising, followed by dimensionality reduction to obtain a low-dimensional latent representation.
• Utilized FastICA and Gaussian mixture models for further dimensionality reduction and clustering of the latent representation.

Main Results:

• scSSA demonstrated superior performance in cell clustering compared to established methods like Seurat and CIDR across 10 public scRNA-seq datasets.
• The model effectively addressed challenges such as dropout events and the curse of dimensionality in scRNA-seq data.
• Accurate cell type identification was achieved, highlighting the model's robustness.

Conclusions:

• scSSA accurately identifies cell types and is broadly applicable to various single-cell datasets.
• The proposed model exhibits significant potential for advancing scRNA-seq data analysis.
• Code for scSSA is publicly available, facilitating further research and application.