peak-calling-pseudo-bulk-aggregation

peak-calling-pseudo-bulk-aggregation

Summary

Aggregate single-cell ATAC-seq fragments into pseudo-bulk samples grouped by cluster assignment, then call peaks using MACS3 to identify chromatin-accessible regions that are consistent within cell populations. This approach recovers peaks from sparse single-cell data by leveraging population-level signal.

When to use

After clustering single-cell ATAC-seq data (e.g., via Leiden clustering on spectral embeddings), use this skill to identify peaks within each cluster. Triggering conditions: (1) you have sparse, per-cell insertion counts organized in a tile matrix; (2) cells have been assigned to discrete clusters or cell-type annotations; (3) you want to call peaks that are reproducible within cell populations rather than in individual cells; (4) downstream analyses require cluster-specific peak lists for differential accessibility or motif enrichment.

When NOT to use

• Input is already a pre-computed peak set or peak matrix; peak calling has been performed elsewhere.
• Cluster assignments are unreliable or highly fragmented (many clusters with <100 cells each), reducing pseudo-bulk signal below MACS3 detection threshold.
• You need single-cell resolution peaks (e.g., to study peak heterogeneity within a cluster); pseudo-bulk aggregation will obscure cell-to-cell variation.

Inputs

• Single-cell ATAC-seq fragment file (imported and stored as paired-end insertions in AnnData)
• Cluster assignments for each cell (from tl.leiden or equivalent clustering method)
• Tile matrix or other sparse count matrix representing chromatin accessibility

Outputs

• Per-cluster peak list (BED format or equivalent, called by MACS3)
• Merged peak set (union or consensus of all cluster-specific peaks)
• Peak matrix (annotation of cells by peak presence/absence, via pp.make_peak_matrix)

How to apply

Group fragments by cluster assignment and aggregate them into a single pseudo-bulk BAM or fragment file per cluster. Pass each pseudo-bulk file to tl.macs3 with appropriate parameters (e.g., genome size, p-value threshold, minimum read count). MACS3 will perform peak calling on the aggregated signal from all cells within a cluster, improving signal-to-noise compared to single-cell calling. After calling peaks independently for each cluster, merge overlapping peaks across clusters using tl.merge_peaks to create a unified peak set. This strategy exploits the population-level chromatin accessibility pattern while avoiding the sparsity problem of individual-cell peak calling. Validate peak calls by comparing cluster assignments and UMAP coordinates before and after peak calling to ensure clustering quality is preserved.