Dissecting Cellular Heterogeneity Based on Network Denoising of scRNA-seq Using Local Scaling Self-Diffusion

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Dissecting cellular heterogeneity is crucial for understanding biological systems.
• Unsupervised clustering is a key method for this analysis but faces challenges with noisy and redundant data.

Purpose of the Study:

• To develop a novel method, Local Scaling Self-Diffusion (LSSD), to enhance metric learning for cell similarities.
• To improve the accuracy and interpretability of clustering in single-cell RNA sequencing (scRNA-seq) data analysis.

Main Methods:

• Proposed Local Scaling Self-Diffusion (LSSD) to enhance cell similarity metric learning.
• Implemented self-diffusion on local scaling affinity matrices to refine cell similarities.
• Applied LSSD to two simulated and four real scRNA-seq datasets.

Main Results:

• LSSD demonstrated superior clustering performance compared to existing single-cell clustering methods.
• The method effectively reduced noise and redundant information in scRNA-seq data.
• Clustering results on colorectal tumor datasets showed strong biological interpretability.

Conclusions:

• LSSD is an effective approach for dissecting cellular heterogeneity and improving clustering in scRNA-seq data.
• The method offers enhanced biological interpretability of identified cell types.
• LSSD addresses key limitations of traditional unsupervised clustering in single-cell analysis.