mbkmeans

mbkmeans implements mini-batch k-means clustering to efficiently cluster large single-cell RNA sequencing (scRNA-seq) datasets and detect cellular subpopulations.

Key Features:

• Scalability and Efficiency: Scales to thousands-to-millions of cells by using mini-batch updates that avoid loading the entire dataset into memory.
• On-Disk Data Representation: Works with on-disk data representations such as HDF5 to process large scRNA-seq datasets without requiring full in-memory loading.
• Performance Optimization: Provides computational performance improvements over standard k-means implementations when applied to very large datasets (e.g., >1M cells).

Scientific Applications:

• Unsupervised clustering of scRNA-seq: Detects distinct cellular subpopulations from single-cell RNA-seq data through unsupervised clustering.
• Cellular heterogeneity and cell type/state identification: Facilitates identification of novel cell types or states and characterization of cellular heterogeneity.
• Large-scale gene expression analysis: Enables analysis of gene expression patterns at single-cell resolution for very large datasets.

Methodology:

Implements the mini-batch k-means algorithm that iteratively updates cluster centroids using small random mini-batches to reduce computational overhead and memory usage, and supports on-disk HDF5-backed data representations.