corral

corral performs correspondence analysis-based dimension reduction and integrative analysis of single-cell RNA sequencing (scRNAseq) count data to produce cell embeddings and visualize gene-cell associations for downstream clustering and comparative studies.

Key Features:

• Correspondence Analysis (CA): CA decomposes a chi-squared residual matrix as a count-based alternative to PCA, avoiding distortive log-transformation of scRNAseq data.
• Adaptations for scRNAseq Data: Five adaptations of CA address overdispersion and high sparsity typical of scRNAseq to generate fast, scalable cell embeddings with improved or comparable clustering accuracy.
• Freeman-Tukey Residuals (CA-FT): CA-FT applies Freeman-Tukey residuals as one adapted CA variant reported to yield superior performance across datasets.
• CA Biplots: CA biplots visualize associations between genes and cell populations to support interpretation of dimension-reduced embeddings.
• Multi-Table Analysis: Extends CA to integrative dimension reduction across multiple tables, sources, or experimental conditions.
• R/Bioconductor Implementation: Implemented as an R/Bioconductor package with integration for Bioconductor single-cell classes.

Scientific Applications:

• Cell Type Identification: Produces embeddings that improve clustering accuracy for identification of distinct cell types in heterogeneous scRNAseq samples.
• Gene Expression Analysis: Visualizes gene-cell associations to aid interpretation of regulatory relationships and functional genomics in single-cell data.
• Comparative Studies: Enables comparative and integrative analyses across conditions or datasets using multi-table dimension reduction.

Methodology:

Uses correspondence analysis by decomposing a chi-squared residual matrix; implements five CA adaptations for overdispersion and sparsity including a Freeman-Tukey residuals variant (CA-FT); generates CA biplots and supports multi-table integrative CA; implemented in R/Bioconductor with integration for Bioconductor single-cell classes.