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.

Related tools

• chromVAR (provides filterSamples() function and SummarizedExperiment container for ATAC-seq counts) — https://github.com/GreenleafLab/chromVAR
• SummarizedExperiment (data structure holding counts matrix, rowData (peak annotations), and colData (sample metadata))
• BiocParallel (optional parallel backend for computationally intensive downstream steps after filtering)

Examples

counts_filtered <- filterSamples(example_counts, min_depth = 1500, min_in_peaks = 0.15)

Evaluation signals

• Post-filter sample count is strictly less than pre-filter count; no samples are added.
• All retained samples have total depth ≥ min_depth threshold (sum of their count matrix columns).
• All retained samples have in-peak fraction ≥ min_in_peaks (sum of counts in peaks / total counts per sample).
• Row dimension (peaks) is unchanged; only columns (samples) are subsetted.
• colData and rowData structures remain intact and consistent with filtered count matrix.

Limitations

• Thresholds (min_depth, min_in_peaks) are dataset and cell type–dependent; chromVAR documentation recommends min_depth=1500 for single-cell ATAC but this may be too stringent for sparse or rare cell populations.
• Filtering is univariate per threshold; joint optimization of both criteria (e.g., trading depth for peak fraction) is not performed automatically.
• No correction for batch effects or cell cycle stage; low-quality samples may cluster with dropout rather than true biology if not addressed separately.
• Removal of samples can inflate power in small cohorts; consider downsampling depth-biased cohorts rather than outright filtering in some contexts.

Evidence

• [intro] filterSamples_description: "it is advisable to filter out samples with insufficient reads using filterSamples"
• [other] filterSamples_parameters: "Filter samples using filterSamples() with min_depth=1500 and min_in_peaks=0.15 to remove low-quality cells"
• [readme] filterSamples_documentation: "counts_filtered <- filterSamples(example_counts, min_depth = 1500, min_in_peaks = 0.15)"
• [intro] output_type_singleexperiment: "The function computeDeviations returns a SummarizedExperiment with two "assays""
• [other] workflow_ordering: "Filter samples using filterSamples() with min_depth=1500 and min_in_peaks=0.15 to remove low-quality cells. 5. Filter peaks using filterPeaks()"