background-peak-selection-normalization

background-peak-selection-normalization

Summary

Select GC content and accessibility-matched background peaks to normalize bias-corrected deviation scores in chromatin accessibility analysis. This normalization step removes technical artifacts and ensures that motif deviation scores reflect true biological variability rather than GC or sequencing depth confounds.

When to use

After computing expected accessibility from filtered peak and sample counts, and before computing final deviation scores. Use this skill when working with sparse ATAC-seq or DNase-seq data where GC bias and accessibility depth are known confounders of motif-associated variability.

When NOT to use

• Input peaks are already aggregated or bulk-level; background-peak normalization is designed for single-cell or sample-level sparse data.
• No GC bias annotations are available in rowData; addGCBias() must be applied first.
• You are computing raw accessibility counts rather than motif-associated variability; use filterPeaks and filterSamples alone if the goal is QC, not deviation scoring.

Inputs

• SummarizedExperiment object with filtered peak counts and rowData containing GC bias annotations
• Pre-computed expected accessibility values (from computeExpectations)
• Matrix of motif-to-peak matches (from matchMotifs)

Outputs

• chromVARDeviations object (SummarizedExperiment) with two assays: deviations and deviationScores
• Normalized motif deviation scores (rows = motifs, columns = samples/cells)

How to apply

Generate background peak sets using getBackgroundPeaks() to obtain peaks matched for GC content and accessibility level to each observed peak. These background peaks serve as control annotations for normalization. Pass the background peaks to computeDeviations() along with the motif-peak matches, filtered counts, and pre-computed expectations. The function will compute deviation scores as the difference between observed motif variability and the expected variability estimated from background peaks, thereby correcting for GC bias and sequencing depth effects. Validate that the deviation assay shows a reasonable distribution (e.g., mean near 0, standard deviation around 1) and that results are reproducible across independent background peak selections.