COTAN

Workflows

Standard Workflow

Clean raw single-cell UMI count datasets by identifying and removing doublets, dead cells, and low-efficiency cells.

library(COTAN)
library(zeallot)

# Resolve operator precedence conflicts
conflicted::conflict_prefer("%<-%", "zeallot")

# 1. Create COTAN object and initialize metadata
rawDataset <- data.frame(Gene1 = c(1, 0, 5), Gene2 = c(0, 2, 0)) # Mock raw counts
obj <- COTAN(raw = rawDataset)
obj <- initializeMetaDataset(
  obj, 
  GEO = "GSM2861514", 
  sequencingMethod = "Drop_seq", 
  sampleCondition = "mouse_cortex_E17.5"
)

# 2. Add condition metadata
data(vignette.cells.origin)
obj <- addCondition(obj, condName = "origin", conditions = vignette.cells.origin)

# 3. Drop problematic cells and run cleaning
obj <- dropGenesCells(obj, cells = c())
obj <- clean(obj)

# 4. Generate diagnostic plots
c(pcaCellsPlot, pcaCellsData, genesPlot, UDEPlot, nuPlot, zoomedNuPlot) %<-% cleanPlots(obj)

Note on inputs/outputs: Input is a raw UMI count matrix; output is a cleaned and filtered COTAN object with estimated cell efficiency (nu) parameters.

Cell Clustering And Dea

Cluster cells based on expression uniformity and perform differential expression analysis (DEA) on the resulting clusters.

library(COTAN)

# 1. Load pre-calculated clusterizations
data("vignette.split.clusters", package = "COTAN")
vignette.split.clusters <- asClusterization(
  vignette.split.clusters, 
  allCells = getCells(obj)
)

# 2. Execute Differential Expression Analysis on clusters
vignette.split.coexDF <- DEAOnClusters(obj, clusters = vignette.split.clusters)

# 3. Save outputs to the COTAN object
obj <- addClusterization(
  obj, 
  clName = "split", 
  clusters = vignette.split.clusters, 
  coexDF = vignette.split.coexDF, 
  override = FALSE
)

Note on inputs/outputs: Input is a COTAN object and a clusterization vector; output is a COTAN object updated with cluster-specific differential expression analysis (DEA) coefficients.

Gene Coexpression And Clustering

Calculate gene-pair co-expression (COEX), calculate Global Differentiation Index (GDI), and establish gene clusters.

library(COTAN)

# 1. Calibrate the COTAN model and estimate dispersion
obj <- proceedToCoex(
  obj, 
  calcCoex = FALSE, 
  optimizeForSpeed = TRUE, 
  cores = 3L
)

# 2. Retrieve and manipulate clusters
splitClusters <- getClusters(obj, clName = "split")
splitClustersAsList <- toClustersList(splitClusters)

Note on inputs/outputs: Input is a cleaned COTAN object; output is a COTAN object with estimated dispersion parameters and cluster lists.

When to Use

• Gene Co-expression Analysis: Assess correlated or anti-correlated expression of gene pairs at the single-cell level using zero UMI count distributions.
• Dataset Cleaning: Identify and drop low-quality cells, doublets, or cells with high mitochondrial percentages using dropGenesCells and clean.
• Differential Expression Analysis (DEA): Assess gene enrichment or depletion across cell clusters using DEAOnClusters.
• Cluster Reordering: Reorder cluster labels based on transcriptional similarity using reorderClusterization.

When NOT to Use

• Standard Highly Variable Gene Selection: For traditional highly variable gene selection and PCA-based clustering without the zero-count contingency framework, use Seurat or scran.
• Trajectory-Based Differential Expression: For modeling continuous gene expression changes along pseudotime, use tradeSeq or condiments.

Data Requirements

• Raw Counts: A matrix or data frame of raw, unnormalized UMI counts (rows as genes, columns as cells).
• Example Structure:

  rawDataset <- read.csv(datasetFileName, sep = "\t", row.names = 1L)
  obj <- COTAN(raw = rawDataset)
  

Key Parameters

• raw: A data frame or matrix of raw UMI counts passed to COTAN().
• calcCoex (TRUE/FALSE): Logical indicating whether to evaluate the gene-pair COEX matrix in proceedToCoex().
• optimizeForSpeed (TRUE): Logical to optimize calculations for speed in proceedToCoex().
• cores (1L): Number of cores to use for multi-processing.
• clName: Name of the clusterization to add or retrieve from the COTAN object.
• override (FALSE): Logical indicating whether to overwrite an existing clusterization in addClusterization().

Best Practices

• Initialize Metadata: Always run initializeMetaDataset immediately after creating the COTAN object to store experimental metadata.
• Quality Control Plots: Check cellSizePlot, genesSizePlot, and mitochondrialPercentagePlot to determine appropriate thresholds before dropping cells.
• Re-run Clean: Run clean() after dropping any genes or cells to re-estimate the cell efficiency (nu) parameters.
• Check UDE Correlation: Verify that cell efficiency (UDE) does not strongly correlate with principal components using UDEPlot.

Common Pitfalls

• Operator Precedence Conflicts: When using multi-assignment (%<-%), resolve operator precedence conflicts by calling conflicted::conflict_prefer("%<-%", "zeallot").
• High Memory/Time Usage: Calculating the full gene-pair COEX matrix can be resource-intensive; set calcCoex = FALSE in proceedToCoex() if only cluster-level DEA is needed.
• Out-of-Date Nu Parameters: Forgetting to run clean() or estimateNuLinear() after dropping cells will lead to inaccurate downstream analyses.

Alternatives

• Seurat: For general single-cell RNA-seq analysis, clustering, and visualization.
• scater / scran: For standard single-cell quality control, normalization, and highly variable gene selection.

Citations

• Yuzwa et al. 2017, Cell Reports (Mouse cortex dataset source)

References

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