A clustering method for single-cell RNA sequencing data based on denoising and masking learning.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is vital for understanding cellular heterogeneity and disease.
• scRNA-seq data present challenges like high dimensionality, sparsity, noise, and dropout events.
• Conventional clustering methods struggle with these data characteristics.

Purpose of the Study:

• To introduce scDMAC, a novel clustering framework for scRNA-seq data.
• To enhance the accuracy and stability of scRNA-seq data clustering.
• To overcome limitations posed by sparsity and dropout events in scRNA-seq analysis.

Main Methods:

• scDMAC integrates a zero-inflated negative binomial (ZINB)-based denoising autoencoder with a masking autoencoder.
• The ZINB autoencoder models count distribution and dropout events for data denoising.
• A masking strategy is applied to denoised data to learn gene-wise correlations via reconstruction.

Main Results:

• scDMAC demonstrated superior clustering accuracy and stability on benchmark scRNA-seq datasets.
• The framework consistently improved clustering performance across diverse datasets.
• Results highlight scDMAC's robustness to noise and sparsity.

Conclusions:

• scDMAC effectively combines probabilistic denoising with masking-based representation learning.
• The framework offers a powerful solution for dropout and sparsity issues in scRNA-seq data.
• scDMAC enhances the extraction of biologically meaningful representations, advancing single-cell transcriptomic analysis.