A new graph-based clustering method with application to single-cell RNA-seq data from human pancreatic islets.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Bulk RNA sequencing provides limited resolution for complex tissues like human pancreatic islets.
• Single-cell RNA sequencing (scRNA-seq) offers a higher resolution for transcriptional profiling of individual cells.
• Clustering algorithms are crucial for identifying cell types and subtypes from heterogeneous scRNA-seq data.

Purpose of the Study:

• To develop a robust clustering framework for analyzing the heterogeneity of scRNA-seq data.
• To improve the accuracy of cell type and subtype discovery in scRNA-seq datasets.
• To provide a practical computational tool for scRNA-seq data analysis.

Main Methods:

• A novel graph-based clustering framework utilizing various dissimilarity measures.
• Incorporation of log-ratio transformations to account for the compositional nature of scRNA-seq data.
• Application of the framework to centered log-ratio-transformed scRNA-seq data from human pancreatic islets.

Main Results:

• The proposed method demonstrates practical merit in identifying cell types and subtypes.
• Performance comparisons show advantages over existing single-cell clustering methods.
• The framework effectively handles the compositional data and heterogeneity inherent in scRNA-seq.

Conclusions:

• The developed graph-based clustering framework offers an effective approach for scRNA-seq data analysis.
• Log-ratio transformations are beneficial for addressing the compositional nature of scRNA-seq data.
• The R-package LrSClust facilitates the application of this advanced clustering method.