Attention-based deep clustering method for scRNA-seq cell type identification.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution insights into cellular heterogeneity.
• Analyzing scRNA-seq data relies on accurate subpopulation assignment, often using unsupervised clustering.
• Traditional clustering methods struggle with scRNA-seq data's sparsity and high dimensionality.

Purpose of the Study:

• To develop a novel deep learning-based clustering method for scRNA-seq data.
• To address the limitations of existing methods in handling data sparsity and dimensionality.
• To improve the accuracy, stability, and robustness of scRNA-seq data clustering.

Main Methods:

• Proposed AttentionAE-sc, integrating zero-inflated negative binomial (ZINB) and graph autoencoder (GAE) methods via an attention mechanism.
• Employed iterative fusion of denoising and topological embeddings for cell representation learning.
• Evaluated performance on 16 diverse scRNA-seq datasets against state-of-the-art methods.

Main Results:

• AttentionAE-sc demonstrated superior clustering performance across 16 datasets without requiring predefined group numbers.
• The method generated improved, clustering-friendly cell representations with enhanced stability and robustness.
• Achieved significant cell subtype identification in a breast cancer scRNA-seq atlas, offering valuable biological insights.

Conclusions:

• AttentionAE-sc provides a powerful and versatile tool for scRNA-seq data analysis and cell subpopulation identification.
• The attention-based fusion approach effectively handles data challenges, leading to more reliable clustering.
• This method advances the understanding of cellular heterogeneity in complex biological systems.