Consensus clustering of single-cell RNA-seq data by enhancing network affinity.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-resolution cell subpopulation identification is crucial for single-cell RNA sequencing (scRNA-seq) data analysis.
• Existing unsupervised clustering methods struggle with scRNA-seq data's high variability, low capture efficiency, and dropout rates.

Purpose of the Study:

• To introduce SCENA (Single-cell Clustering by Enhancing Network Affinity), a novel unsupervised method for robust cell clustering.
• To improve the accuracy and efficiency of cell population identification in scRNA-seq data.

Main Methods:

• SCENA employs three key strategies: multiple gene set selection, enhanced local cell affinity, and consensus matrix clustering.
• The method utilizes Central Processing Units (CPUs) and Graphics Processing Units (GPUs) for heterogeneous parallel computing.

Main Results:

• SCENA demonstrated high accuracy in detecting cell populations across 13 diverse scRNA-seq datasets.
• The method proved robust against dropout noise, a common challenge in scRNA-seq experiments.
• Application to mouse brain cell data successfully identified known types and discovered novel interneuron populations.

Conclusions:

• SCENA offers a high-performance and efficient platform for analyzing large and complex scRNA-seq datasets.
• The method facilitates biological discovery by improving cell clustering accuracy and speed.
• SCENA addresses key limitations of existing methods, paving the way for deeper insights into cellular heterogeneity.