mbkmeans: Fast clustering for single cell data using mini-batch k-means.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity.
• Unsupervised clustering is a common method for identifying cell subpopulations in scRNA-seq data.
• Current scRNA-seq datasets are growing rapidly, posing computational challenges for traditional clustering algorithms.

Purpose of the Study:

• To develop an efficient clustering method for large-scale scRNA-seq datasets.
• To address the memory and speed limitations of standard k-means clustering for millions of cells.
• To provide an open-source R/Bioconductor package for scalable single-cell data analysis.

Main Methods:

• Implementation of the mini-batch k-means algorithm in an R package named mbkmeans.
• Utilizing on-disk data representations, such as HDF5, to handle large datasets without full in-memory loading.
• Benchmarking mbkmeans performance against standard k-means and other popular single-cell clustering tools.

Main Results:

• The mbkmeans package successfully clusters large scRNA-seq datasets, including one with 1.3 million cells.
• mbkmeans demonstrates significantly improved computational performance compared to standard k-means for large datasets.
• The package offers a scalable and memory-efficient solution for single-cell data clustering.

Conclusions:

• mbkmeans provides an effective solution for clustering massive single-cell RNA sequencing datasets.
• The mini-batch k-means approach overcomes computational bottlenecks associated with large-scale single-cell data analysis.
• The availability of mbkmeans in Bioconductor facilitates its adoption in the research community for robust cell subpopulation identification.