bootstrap-confidence-interval-computation

bootstrap-confidence-interval-computation

Summary

Compute bootstrap confidence intervals around variability metrics (e.g., standard deviations of z-scores) for genomic annotations by resampling cells or samples with replacement. This skill provides uncertainty quantification for ranking and hypothesis testing of motif variability across chromatin accessibility datasets.

When to use

When you have computed z-score deviations for genomic annotations (e.g., motifs) across multiple cells or samples and need to quantify uncertainty in their variability rankings before performing differential or comparative analyses. Use this when sample size is small or variability estimates may be unstable.

When NOT to use

• Input is already a pre-computed confidence interval table or posterior distribution — do not resample.
• Sample size is very large (n >> 1000) and asymptotic methods are more efficient.
• Variability is computed on aggregated bulk data with no cell/sample structure to resample.

Inputs

• chromVARDeviations object (bias-corrected z-score deviations for annotations × samples)
• Variability metric (standard deviation of z-scores per annotation)
• Annotation labels (e.g., motif IDs)
• Sample/cell metadata (colData)

Outputs

• Bootstrap confidence intervals (lower and upper bounds per annotation)
• Bootstrap replicate distributions (variability scores across resamples)
• Ranked annotations by variability with uncertainty bands

How to apply

After computing variability scores (e.g., standard deviation of z-scores across samples for each motif), resample cells or samples with replacement multiple times (bootstrap replicates) to recompute the variability metric for each annotation in each replicate. Collect the distribution of bootstrap estimates and compute percentile-based confidence intervals (e.g., 2.5th and 97.5th percentiles for 95% CI). These intervals quantify the range of plausible variability values and support downstream hypothesis testing against a null variability threshold (e.g., 1.0 for normalized deviations).