Asurat

ASURAT

Workflows

Standard Workflow

Perform single-cell RNA-seq preprocessing, functional sign creation from databases, sign-by-sample matrix generation, clustering, and multifaceted visualization.

library(ASURAT)
library(SingleCellExperiment)

# 1. Add metadata and perform quality control filtering
pbmc <- add_metadata(sce = pbmc, mitochondria_symbol = "^MT-")
pbmc <- remove_variables(sce = pbmc, min_nsamples = 10)
pbmc <- remove_samples(sce = pbmc, min_nReads = 5000, max_nReads = 20000, min_nGenes = 100, max_nGenes = 1e+10, min_percMT = 0, max_percMT = 10)
pbmc <- remove_variables_second(sce = pbmc, min_meannReads = 0.05)

# 2. Create and select signs from databases
pbmc <- remove_signs(sce = pbmc, min_ngenes = 2, max_ngenes = 1000)
pbmc <- cluster_genesets(sce = pbmc, cormat = cormat, th_posi = 0.30, th_nega = -0.30)
pbmc <- create_signs(sce = pbmc, min_cnt_strg = 2, min_cnt_vari = 2)

# 3. Generate sign-by-sample matrix (SSM)
pbmc <- makeSignMatrix(sce = pbmc, weight_strg = 0.5, weight_vari = 0.5)

# 4. Identify significant signs
pbmc <- compute_sepI_all(sce = pbmc, labels = labels, nrand_samples = 200)

Input: A raw count SingleCellExperiment object and knowledge-based databases. Output: A SingleCellExperiment object containing a sign-by-sample matrix (SSM) with computed separation indices.

When to Use

• To simultaneously perform unsupervised clustering and biological interpretation of single-cell RNA-seq data using functional gene sets from databases like Cell Ontology, Gene Ontology, or KEGG.
• To transform normalized-and-centered gene expression matrices into sign-by-sample matrices (SSMs) using makeSignMatrix.
• To identify significant biological markers (signs) for cell clusters using non-parametric separation indices computed via compute_sepI_all.
• To visualize multifaceted single-cell datasets across multiple sign-by-sample matrices and gene expression levels simultaneously using plot_multiheatmaps.

When NOT to Use

• For standard gene-level-only clustering without functional database integration, use scran or Seurat because ASURAT is specifically designed for sign-based (gene set) multivariate analysis.
• For datasets lacking reliable functional annotations or established gene sets, use standard unsupervised clustering because ASURAT relies on knowledge-based databases to define signs.

Data Requirements

• Input format: SingleCellExperiment object containing raw counts in assays(sce)$counts.
• Structure: Row names must be gene symbols (e.g., Symbol_TENx) and column names must be cell barcodes.
• Normalization state: Requires log-normalized and centered expression data (stored in assay(sce, "centered")) to compute gene expression correlation matrices.

Key Parameters

• mitochondria_symbol ("^MT-"): Regular expression to identify mitochondrial genes for quality control.
• min_nsamples (10): Minimum number of non-zero expressing cells required to keep a gene.
• min_meannReads (0.05): Minimum mean read count across samples required to keep a gene.
• min_ngenes (2): Minimal number of genes in a functional gene set to be kept.
• max_ngenes (1000): Maximal number of genes in a functional gene set to be kept.
• th_posi (0.30): Positive correlation coefficient threshold for clustering gene sets.
• th_nega (-0.30): Negative correlation coefficient threshold for clustering gene sets.
• weight_strg (0.5): Weight parameter for strongly correlated gene sets (SCG) in SSM creation.

Best Practices

• Add metadata to both rows and columns using add_metadata before performing quality control filtering.
• Perform log-normalization and center the data by subtracting the mean expression levels across cells before computing correlation matrices.
• Filter out redundant signs using semantic similarity matrices with remove_signs_redundant to simplify the downstream analysis.
• Use compute_sepI_all to find cluster-specific signs instead of standard parametric tests, as row vectors of SSMs are centered.

Common Pitfalls

• Using slot name "log-normalized" for altExp(sce): This can cause errors when converting to Seurat objects using as.Seurat. Fix by using standard slot names like "logcounts".
• Setting excessive cutoff values in remove_samples: This can produce biased results. Carefully inspect the distribution of nReads and nGenes before setting thresholds.

Alternatives

• Seurat for standard graph-based clustering and cell cycle scoring.
• scater for standard single-cell preprocessing and visualization.
• ComplexHeatmap for general high-dimensional heatmap plotting.

Citations

• Iida K (2026), "ASURAT: Functional sign-based single-cell analysis".

References

• Homepage: bioconductor.org/packages/asurat
• Vignette: bioconductor.org/packages/release/bioc/vignettes/asurat/inst/doc/asurat.html