CAdir: Joint clustering of cells and genes for single-cell transcriptomics with visualization-driven cluster quality...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) clustering aims to group similar cells.
• Existing algorithms often fail to identify marker genes or determine cluster numbers unsupervisedly.
• Current quality assessment relies on potentially misleading low-dimensional embeddings (e.g., UMAP, t-SNE).

Purpose of the Study:

• To develop an interpretable clustering algorithm for scRNA-seq data.
• To enable unsupervised inference of cluster number and identification of cluster-specific genes.
• To provide diagnostic tools for evaluating clustering quality.

Main Methods:

• Developed CAdir, a novel clustering algorithm.
• Utilizes correspondence analysis (CA) geometry to cluster cells and genes.
• Employs cluster direction angles in CA space to infer and refine cluster numbers.
• Integrates diagnostic plots for feedback on clustering decisions.

Main Results:

• CAdir infers the number of clusters and identifies defining marker genes.
• The algorithm provides interpretable diagnostic plots for quality assessment.
• Benchmarking shows comparable performance to state-of-the-art methods.
• CAdir is scalable to large scRNA-seq datasets.

Conclusions:

• CAdir offers an improved, interpretable approach to scRNA-seq data clustering.
• It addresses limitations of existing methods by automating cluster number inference and marker gene identification.
• The diagnostic tools enhance the reliability of clustering results.