single-cell-chromatin-sample-filtering

single-cell-chromatin-sample-filtering

Summary

Remove low-quality single cells or bulk samples from ATAC-seq or DNAse-seq chromatin accessibility data prior to downstream analysis. This quality control step eliminates samples with insufficient sequencing depth or low peak coverage that would compromise motif deviation or clustering accuracy.

When to use

Apply this skill after loading fragment counts into a SummarizedExperiment object (e.g., via getCounts) but before motif matching or deviation computation. Use it when working with sparse, sample-wise chromatin accessibility data where sequencing depth and in-peak read fraction vary substantially across cells or samples—typical in single-cell ATAC-seq workflows.

When NOT to use

• Samples are already pre-filtered by the upstream alignment pipeline or cell caller—applying additional filtering may remove too few cells to affect power.
• Input is a pre-aggregated bulk sample matrix without per-sample depth metadata—thresholds become arbitrary.
• Analysis goal is to benchmark filtering sensitivity; use unfiltered data explicitly as a control.

Inputs

• SummarizedExperiment object with fragment count matrix (from getCounts or similar)
• colData metadata defining sample identifiers and optional grouping variables

Outputs

• Filtered SummarizedExperiment object with same structure but reduced column dimension (fewer samples)
• Integer vector of retained sample indices (implicit)

How to apply

Use the filterSamples() function from chromVAR to remove samples falling below user-specified thresholds on two key metrics: (1) min_depth, the minimum total fragment count per sample (e.g., 1500 for single cells), and (2) min_in_peaks, the minimum fraction of reads mapping to called peaks (e.g., 0.15). The rationale is that samples with low sequencing depth produce unreliable motif deviation estimates, while low in-peak fractions indicate either poor ATAC-seq quality or failed library preparation. Filter samples before filterPeaks() to ensure expectations and background peak matching operate on a clean, consistent sample cohort. Validate the filter by comparing pre- and post-filter sample counts and checking that remaining samples meet both criteria.