Improving replicability in single-cell RNA-Seq cell type discovery with Dune.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-Seq) enables high-resolution biological investigations.
• A key goal of scRNA-Seq is accurate cell type classification.
• Current methods often rely on heuristics or user-defined parameters for clustering, impacting resolution and replicability.

Purpose of the Study:

• To develop a novel method, Dune, for optimizing the trade-off between cluster resolution and replicability in scRNA-Seq data.
• To provide an objective approach for refining cell cluster analysis.

Main Methods:

• Dune takes multiple clustering results (partitions) of a single dataset as input.
• It iteratively merges clusters within partitions to maximize concordance across them.
• The method aims to improve the replicability and biological relevance of cell clusters.

Main Results:

• Dune outperforms existing hierarchical merging techniques in cluster replicability.
• The method demonstrates higher concordance with ground truth cell types.
• Evaluated on diverse datasets from multiple platforms, Dune shows consistent improvements.

Conclusions:

• Dune refines clustering analyses, enhancing the robustness of cell type identification.
• It reduces the need for extensive parameter tuning in scRNA-Seq studies.
• Dune facilitates the generation of replicable clusters representing common biological features across datasets.