Deep learning-based enhancement of epigenomics data with AtacWorks

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Assay for Transposase-Accessible Chromatin sequencing (ATAC-seq) is crucial for mapping genome-wide chromatin accessibility.
• Data quality, sequencing depth, and signal-to-noise ratio can limit ATAC-seq's ability to identify active regulatory regions.

Purpose of the Study:

• To introduce AtacWorks, a deep learning toolkit designed to denoise ATAC-seq data and identify regulatory elements.
• To improve peak detection in low-cell count, low-coverage, or low-quality ATAC-seq experiments.

Main Methods:

• Development of a deep learning framework (AtacWorks) for processing ATAC-seq data.
• Training models to identify regulatory peaks at base-pair resolution.
• Validation across diverse sample preparations and experimental platforms.

Main Results:

• AtacWorks effectively denoises sequencing coverage and identifies regulatory peaks from challenging ATAC-seq datasets.
• Models demonstrate generalizability to unseen cell types and experimental conditions.
• AtacWorks significantly enhances the sensitivity of single-cell ATAC-seq, achieving comparable results to conventional methods with ~10x fewer cells.
• The framework facilitates cross-modality inference of protein-DNA interactions.

Conclusions:

• AtacWorks provides a robust solution for analyzing low-quality or low-coverage ATAC-seq data.
• The toolkit enables more sensitive detection of regulatory elements, particularly in single-cell applications.
• AtacWorks can lead to novel biological insights, such as identifying regulatory regions in rare cell populations like hematopoietic stem cells.