scMVAF: a multi-view adaptive fusion clustering approach for single-cell RNA-sequencing data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for identifying cellular heterogeneity.
• Clustering scRNA-seq data into subpopulations is vital for downstream analysis.
• Current clustering methods struggle with high-dimensional, sparse, and noisy scRNA-seq data.

Purpose of the Study:

• To develop a novel multi-view clustering framework, scMVAF, for scRNA-seq data.
• To improve the discriminative power of embedding representations for cell clustering.
• To address the limitations of single-perspective analysis in existing methods.

Main Methods:

• Generated multiple data views by feature down-sampling.
• Employed an autoencoder with a denoising zero-inflated negative binomial model for view-specific embedding.
• Introduced a multi-view fusion module to integrate embeddings and generate pseudo-labels for iterative refinement.

Main Results:

• scMVAF integrates information from multiple cell views to learn robust embedding representations.
• The multi-view fusion module effectively combines information across different views.
• Clustering performance is enhanced through iterative updates of embeddings and pseudo-labels.

Conclusions:

• scMVAF demonstrates superior performance compared to eight advanced methods across 16 real datasets.
• The proposed framework effectively captures cellular heterogeneity in scRNA-seq data.
• scMVAF provides a powerful tool for accurate cell subpopulation identification.