fragment-based-count-matrix-generation

fragment-based-count-matrix-generation

Summary

Generate tile-based count matrices from single-cell ATAC-seq fragment data using paired-insertion counting strategy in SnapATAC2. This skill quantifies chromatin accessibility at fixed genomic intervals and produces dense or sparse count matrices suitable for downstream dimensionality reduction and clustering.

When to use

You have a backed AnnData object containing processed fragment data (stored in .obsm['fragment_paired'] or .obsm['fragment_single']) from single-cell ATAC-seq and need to convert raw genomic fragments into a quantitative count matrix indexed by fixed genomic tiles for clustering, embedding, or differential analysis.

When NOT to use

• Input data is already in the form of a peak-by-cell matrix or gene-by-cell matrix; use pp.make_peak_matrix or pp.make_gene_matrix instead.
• Fragment file has not been imported or processed; first convert BAM to fragment file using pp.make_fragment_file or pp.import_fragments.
• You need counts at peak regions rather than fixed tiles; use pp.make_peak_matrix with called or external peak coordinates.

Inputs

• backed AnnData object with fragment data in .obsm['fragment_paired'] or .obsm['fragment_single']
• genomic fragment coordinates (chromosome, start, end, barcode)

Outputs

• count matrix (n_obs × n_vars) indexed by genomic tiles and cell barcodes
• tile coordinate metadata (chromosome, start, end positions for each tile)
• sparse matrix with paired-insertion counts per cell per tile

How to apply

Load the backed AnnData object containing fragment coordinates and invoke pp.add_tile_matrix with counting_strategy='paired_insertion' to aggregate fragment counts across non-overlapping genomic tiles (fixed intervals, typically 5 kb). The function processes paired-end fragments by counting insertions at each genomic position, then sums counts within each tile to generate a sparse count matrix with dimensions n_obs (cells) × n_vars (tiles). Verify the resulting matrix shape matches the number of cells and the expected number of tiles, and confirm that count values are non-zero and reasonably distributed across the tile-cell space (no unexpected sparsity or skew).