Linear-time cluster ensembles of large-scale single-cell RNA-seq and multimodal data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates massive datasets, necessitating scalable analysis methods.
• Current clustering methods often use sampling, which can lead to inaccurate cell groupings and missed rare cell types.
• Efficiently analyzing ultralarge scRNA-seq data is crucial for biological discovery.

Purpose of the Study:

• To develop a scalable and accurate method for clustering ultralarge scRNA-seq datasets.
• To improve the identification of transcriptionally distinct cell populations, including rare cell types.
• To leverage multimodal omics data for enhanced cell subpopulation resolution.

Main Methods:

• Specter employs a landmark-based approach for sparse data representation.
• It extends fast spectral clustering algorithms to achieve linear-time complexity.
• A cluster ensemble scheme is utilized to enhance accuracy and sensitivity.

Main Results:

• Specter scales linearly to millions of cells, offering fast computation times.
• The method demonstrates improved accuracy in cell grouping compared to existing approaches.
• Specter successfully identifies rare cell types with high sensitivity.
• It effectively resolves subtle transcriptomic differences using multimodal CITE-seq data.

Conclusions:

• Specter provides a scalable and accurate solution for clustering ultralarge scRNA-seq data.
• The landmark-based spectral clustering approach overcomes limitations of data sampling.
• Specter enhances the discovery of cellular heterogeneity and rare cell populations.