DIMM-SC

DIMM-SC applies a Dirichlet mixture model to cluster droplet-based single-cell RNA sequencing (scRNA-Seq) UMI count data, enabling model-based identification of cell populations and quantification of clustering uncertainty.

Key Features:

• Dirichlet mixture modeling: Explicitly models UMI count data using a Dirichlet mixture model tailored for droplet-based scRNA-Seq.
• Model-based clustering: Provides a probabilistic clustering framework that characterizes variation across cell clusters.
• Quantification of clustering uncertainty: Produces measures of uncertainty for each cell's cluster assignment to support statistical interpretation.
• Improved accuracy and reduced variability: Demonstrated substantially improved clustering accuracy and lower variability compared with K-means, CellTree, and Seurat in simulations.
• Validation and benchmarking: Evaluated via simulation studies and benchmarked on public scRNA-Seq datasets with known cluster labels and an in-house systemic sclerosis dataset.

Scientific Applications:

• Cellular heterogeneity: Identifying and characterizing diverse cell populations within complex tissues using droplet-based scRNA-Seq UMI data.
• Disease mechanisms (systemic sclerosis): Analyzing cellular responses and population changes in systemic sclerosis at single-cell resolution.
• Developmental biology: Studying differentiation processes and lineage-related population structure in developmental systems.

Methodology:

Uses a Dirichlet mixture model that explicitly models UMI counts, quantifies uncertainty in cluster assignments, and was assessed by simulation studies and benchmarking against K-means, CellTree, and Seurat on public scRNA-Seq datasets with known labels and an in-house systemic sclerosis dataset.