deviation-score-computation-and-interpretation

Deviation-score computation and interpretation

Summary

Compute bias-corrected deviation scores that quantify motif-associated variability in chromatin accessibility across samples using chromVAR's computeDeviations function. This skill enables identification of transcription factor motifs driving cell-to-cell or sample-to-sample epigenetic heterogeneity in ATAC-seq data.

When to use

Apply this skill when you have filtered ATAC-seq peak counts, matched motifs to those peaks, and want to measure which transcription factor motifs show elevated or reduced accessibility relative to GC-content and accessibility-matched background expectations—particularly when annotating TF motif usage across cell populations or when exploring which regulatory elements drive chromatin variability.

When NOT to use

• Input peak counts are not filtered by sample quality (depth < 1500 reads or in-peak fraction < 0.15); filterSamples must precede this skill.
• Peaks have not been reduced to non-overlapping set; overlapping peaks violate the background-matching assumptions underlying bias correction.
• Motifs have not been matched to peaks; computeDeviations requires an explicit motif match matrix, not raw motif sequences.

Inputs

• SummarizedExperiment object with filtered peak counts (samples × peaks)
• GC bias annotations in rowData (output from addGCBias)
• Expected accessibility matrix (output from computeExpectations)
• Motif-to-peak match matrix (logical matrix from matchMotifs)
• Background peak indices (output from getBackgroundPeaks)

Outputs

• chromVARDeviations SummarizedExperiment object with two assays: 'deviations' (bias-corrected z-scores, motifs × samples) and 'deviationScores' (raw deviations, motifs × samples)
• Row names correspond to motif identifiers
• Column names correspond to sample/cell identifiers

How to apply

After filtering samples (min_depth ≥1500, min_in_peaks ≥0.15) and peaks (non-overlapping set), add GC content bias to rowData using addGCBias() with the reference genome, compute expected accessibility using computeExpectations() on filtered counts, and generate GC- and accessibility-matched background peaks using getBackgroundPeaks(). Then invoke computeDeviations() with the filtered SummarizedExperiment object, motif match matrix (from matchMotifs), background peaks, and expected accessibility; this returns a SummarizedExperiment with two assays: 'deviations' (bias-corrected z-scores) and 'deviationScores' (raw deviation magnitudes). Validate the output by checking that row count equals motif count, column count equals sample count, and that score distributions reflect expected variability patterns without extreme outliers.