AtacWorks

AtacWorks applies deep convolutional neural networks to denoise low-coverage or low-quality ATAC-seq (Assay for Transposase-Accessible Chromatin using sequencing) data and to improve identification of accessible chromatin for epigenomic analysis.

Key Features:

• Denoising: Uses a deep neural network trained on high-quality ATAC-seq data to transform noisy, low-coverage signals into cleaner accessibility profiles.
• Peak calling precision: Identifies accessible chromatin peaks from low-coverage bulk sequencing at high (base-pair) resolution across diverse cell types and conditions.
• Cross-modality prediction: Enables inference of transcription factor footprints and ChIP-seq peak signals from low-input ATAC-seq data.
• Single-cell aggregation application: Enhances aggregate single-cell ATAC-seq profiles to detect regulatory regions differentially accessible in rare subpopulations, including lineage-primed hematopoietic stem cells.

Scientific Applications:

• Low-coverage epigenomic profiling: Improves signal-to-noise in low-coverage or low-quality ATAC-seq datasets to enable more accurate chromatin accessibility analyses.
• Regulatory landscape mapping: Facilitates precise identification of regulatory elements and peak boundaries at base-pair resolution.
• Rare cell population analysis: Supports detection of differentially accessible regions in rare or limited samples using aggregate single-cell ATAC-seq profiles.
• Cross-modality inference: Allows prediction of transcription factor footprints and ChIP-seq–like peaks from ATAC-seq input for complementary regulatory interpretation.

Methodology:

Trains deep convolutional neural networks on paired noisy and high-quality ATAC-seq datasets from the same cell type and applies the trained model to new datasets to denoise signal and improve peak identification; the architecture is reported to generalize across cell types and experimental conditions.