An interpretable single-cell RNA sequencing data clustering method based on latent Dirichlet allocation.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptome data but often results in sparse and noisy datasets.
• Analyzing scRNA-seq data presents challenges in gene selection, cell clustering, annotation, and biological mechanism discovery.
• Existing methods struggle with the inherent complexity and noise in scRNA-seq data.

Purpose of the Study:

• To introduce a novel scRNA-seq analysis method utilizing the latent Dirichlet allocation (LDA) model.
• To address the difficulties in analyzing sparse and noisy scRNA-seq data by employing a 'cell-function-gene' framework.
• To improve cell clustering, annotation, and biological interpretation of scRNA-seq datasets.

Main Methods:

• Developed an scRNA-seq analysis method based on the latent Dirichlet allocation (LDA) model.
• Integrated a 'cell-function-gene' three-layer framework to discover latent gene expression patterns.
• Compared the LDA-based method against four classic methods on seven benchmark scRNA-seq datasets.

Main Results:

• The LDA-based method demonstrated superior performance in cell clustering accuracy and purity compared to existing methods.
• The method successfully distinguished cell types with functional specialization and reconstructed cell development trajectories on complex datasets.
• Accurately identified representative putative functions (PFs) and genes for cell types/stages, facilitating data-driven annotation and interpretation.

Conclusions:

• The proposed LDA-based method offers a robust framework for analyzing scRNA-seq data, overcoming common challenges.
• This approach enhances the biological interpretability of scRNA-seq data by uncovering latent functional patterns.
• The method effectively supports cell type identification, functional specialization analysis, and developmental trajectory reconstruction.