Tools — gatac.tl

Tools — `gatac.tl`#

The gatac.tl namespace provides downstream analysis tools: dimensionality reduction, peak calling, marker detection, motif scanning, chromVAR deviation scoring, and topic modelling.

Dimensionality reduction#

Compute a spectral decomposition of the cell × feature matrix — the standard entry point for UMAP and clustering in ATAC-seq workflows (spectral) — or model topics over the peak-accessibility matrix with GPU-accelerated mini-batch Online Variational Bayes (lda, MiniBatchLDA).

`spectral`	GPU-accelerated spectral embedding via Laplacian Eigenmaps.
`lda`	Learn topics from a binarized peak matrix using GPU-accelerated mini-batch LDA.
`MiniBatchLDA`	GPU-accelerated Mini-batch LDA via Online Variational Bayes.

Peak calling & marker peaks#

Call ATAC peaks per cell-type group using the MACS3 algorithm under the hood, merge them into a non-overlapping set, count fragments over peaks, and identify differentially accessible peaks between groups using a GPU- accelerated binomial test with Benjamini–Hochberg correction.

`call_peaks`	GPU-accelerated peak calling per cluster using gmacs algorithm.
`merge_peaks`	Merge peaks from different groups into fixed-width, non-overlapping peaks.
`make_peak_matrix`	Generate cell by peak count matrix.
`marker_peaks`	GPU-accelerated marker peak detection using binomial test.
`get_marker_peaks`	Extract marker peak names from stored results.

Motif analysis#

Read motifs from MEME-format files, test for over-representation in peak sets, and run GSEA on motif rankings.

`read_motifs`	Read motifs from a MEME format file.
`parse_meme`	Parse MEME format content into motifs.
`DNAMotif`	DNA motif represented as a position weight matrix (PWM).
`sample_gc_matched_background`	Sample background peaks whose GC-content distribution matches target peaks.
`motif_enrichment`	Identify enriched transcription factor motifs using GPU acceleration.
`gsea_motif_enrichment`	Run preranked GSEA to identify enriched TF motifs from a LogFC-ranked peak list.

chromVAR#

Compute transcription-factor activity deviation scores following the chromVAR algorithm. All compute-intensive steps are executed on GPU.

`chromvar`	Run the full chromVAR pipeline in a single call.
`compute_peak_bias`	Compute peak biases (GC content and/or CpG density) for background sampling.
`sample_bg_peaks`	Sample background peaks for chromVAR analysis.
`scan_motifs`	Scan peaks for motif matches and create a sparse motif match matrix.
`compute_deviations`	Compute chromVAR TF deviation scores.

To run the four steps individually, see the docstring of compute_deviations (which lists them end-to-end).