pyranges-peak-gene-linkage

Compute Pearson correlation between each peak's accessibility and each nearby gene's expression across cells sharing both modalities. Restrict peaks to a genomic window around each gene, then filter by p-value and rank by correlation coefficient.

Use snap.pp.make_gene_matrix from SnapATAC2 with a genome annotation and upstream window, or mu.atac.pp.gene_activity from muon. Both sum peak counts over gene bodies plus flanking regions to produce per-gene ATAC scores.

Use ±2 kb around the TSS for promoter-only linkage, ±100 kb for standard cis-regulatory analysis, or ±500 kb to include long-range enhancers. Novel analyses should sweep multiple windows and report sensitivity.

Near-zero correlations usually mean peaks are too sparse. Aggregate peaks into broader bins, restrict to highly variable peaks, or correlate within individual cell types to remove cell-type-driven noise.

No, this approach requires paired cells with both modalities. For unpaired RNA and ATAC datasets, first integrate them with an unpaired multiomics integration method such as scglue before attempting linkage.

Apply Benjamini-Hochberg correction across all tested peak-gene pairs in the window. Raw p-value filtering without correction inflates false positives when testing thousands of pairs genome-wide.