Denoising adaptive deep clustering with self-attention mechanism on single-cell sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular diversity and function.
• Clustering is essential for identifying cell types in scRNA-seq data.
• High dimensionality, noise, and dropout rates in scRNA-seq data pose significant challenges for accurate clustering.

Purpose of the Study:

• To propose a novel adaptive fuzzy clustering model, scDASFK, for scRNA-seq data analysis.
• To enhance the accuracy and robustness of cell type identification from noisy and high-dimensional scRNA-seq datasets.
• To develop a method that integrates denoising and clustering effectively.

Main Methods:

• scDASFK utilizes a denoising autoencoder and a self-attention mechanism for data preprocessing.
• Comparative learning is employed to integrate cell similarity information into the clustering process.
• An adaptive feedback mechanism supervises denoising using clustering results for improved latent feature representation.

Main Results:

• scDASFK demonstrates strong performance across 16 real-world scRNA-seq datasets.
• The model shows excellent clustering accuracy, scalability, and stability.
• Experimental results validate the effectiveness of the proposed denoising and adaptive clustering approach.

Conclusions:

• scDASFK is an effective and robust clustering model for scRNA-seq data analysis.
• The model offers significant potential for advancing the study of cellular heterogeneity.
• The scDASFK code is publicly available for research use.