Dune

Dune optimizes concordance among multiple clustering partitions of single-cell RNA sequencing (scRNA-Seq) data by iteratively merging clusters to balance cluster resolution and replicability across datasets.

Key Features:

• Parameter-free optimization: Performs optimization without requiring user-defined tuning parameters to balance cluster resolution and replicability.
• Input flexibility: Accepts a set of clustering results (partitions) produced by arbitrary algorithms and parameter settings on a dataset.
• Iterative merging: Iteratively merges clusters within partitions to increase agreement across different clustering outcomes.
• Hierarchical merging: Applies hierarchical merging of clusters as the core merging strategy.
• Implementation: Provided as an R package for analysis of single-cell transcriptome data.
• Concordance maximization: Aims to maximize cluster concordance across partitions to produce more consistent cluster assignments.
• Consensus cluster identification: Identifies consensus clusters that are more likely to represent common biological features across datasets.
• Replicability improvement: Has been demonstrated to improve the replicability of merged clusters relative to existing techniques.

Scientific Applications:

• Cell type discovery: Facilitates classification of cells into putative novel cell types from scRNA-Seq datasets.
• Parameter selection assessment: Helps assess and mitigate the impact of clustering parameter choices on cluster granularity.
• Cross-dataset reproducibility: Derives consensus clusters that improve reproducibility of cell-type assignments across datasets.
• Comparative clustering analysis: Enables comparison and consolidation of results from multiple clustering algorithms applied to the same dataset.

Methodology:

Accepts multiple clustering partitions as input and iteratively performs hierarchical merging of clusters within partitions to maximize concordance across clustering outcomes.