bias-corrected-z-score-interpretation

bias-corrected-z-score-interpretation

Summary

Interpret deviation z-scores computed by chromVAR as bias-corrected measures of individual or sample-level deviations from expected chromatin accessibility patterns at annotated genomic features (motifs or kmers). This skill enables ranking and differential testing of annotations by their variability across cells or samples.

When to use

After computeDeviations has generated a SummarizedExperiment object with z-score assays reflecting bias-corrected deviations of observed vs. expected accessibility at motif or kmer sites. Use this skill when you need to (1) rank annotations by their across-sample/cell variability, (2) identify which annotations show statistically significant differential usage between cell types or conditions, or (3) understand which individual samples/cells deviate most from the accessibility baseline at a given annotation.

When NOT to use

• Input counts have not been filtered for GC bias using addGCBias — bias correction requires explicit bias annotation in the input object
• Raw count matrix is the input; z-scores must first be computed via computeDeviations before interpretation
• Analyzing bulk ATAC-seq without single-cell resolution, where sample-level deviation interpretation differs fundamentally from cell-level interpretation

Inputs

• chromVARDeviations object (SummarizedExperiment with z-score assays from computeDeviations)
• Cell or sample metadata (colData) including grouping variables (e.g., cell_type, sample_id)
• Optionally: pre-defined grouping variable name for differential testing (e.g., 'Cell_Type')

Outputs

• Variability scores per annotation (standard deviation of z-scores across samples)
• Bootstrap confidence intervals around variability estimates
• Ranked list of annotations sorted by variability
• Differential deviation test results (p-values, effect sizes, adjusted p-values per annotation and group pair)
• Visualization of ranked annotations (via plotVariability)