tile-matrix-generation-counting-strategy-selection

tile-matrix-generation-counting-strategy-selection

Summary

Select and apply an appropriate counting strategy (e.g., paired-insertion counting) when generating a tile matrix from ATAC-seq fragment data in SnapATAC2. This skill bridges fragment-level preprocessing and downstream spectral embedding by converting raw alignments into a discrete feature matrix suitable for dimension reduction.

When to use

After importing fragment files into AnnData using pp.import_fragments and before performing spectral embedding (tl.spectral) or other dimension reduction. Apply this skill when you have paired-end ATAC-seq data with properly formatted fragment coordinates and need to represent chromatin accessibility as a tile-by-cell count matrix for clustering and visualization.

When NOT to use

• Input is already a peak-by-cell or gene-by-cell count matrix (use directly for embedding instead).
• Fragment data contains single-end reads without paired mate information (requires alternative counting strategy or realignment).
• Analysis goal is peak-level rather than genome-wide accessibility profiling (use pp.make_peak_matrix instead).

Inputs

• AnnData object with imported fragment data (adata with obs column containing cell barcodes and var containing genomic coordinates from pp.import_fragments)
• Reference genome assembly or chrom.sizes file defining tile boundaries

Outputs

• AnnData object with tile matrix added as a sparse count matrix (adata.X or named layer)
• Tile coordinates in adata.var indexed by genomic position (chr:start-end)

How to apply

Use SnapATAC2's pp.add_tile_matrix function with paired-insertion counting strategy, which counts the number of Tn5 insertions falling within non-overlapping genomic tiles (typically 5 kb). The paired-insertion strategy correctly handles paired-end reads by counting each valid fragment pair once, avoiding double-counting and artificial noise from single-end artefacts. This produces a sparse, binary or count matrix indexed by tile coordinates and cell barcodes. Validate that the resulting matrix has non-zero coverage across cell populations and that tile counts correlate with expected chromatin accessibility patterns before proceeding to embedding.