maxATAC: Genome-scale transcription-factor binding prediction from ATAC-seq with deep neural networks

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Transcription factors (TFs) are crucial regulators of gene expression, connecting DNA sequence to cellular behavior.
• Predicting TF binding sites (TFBS) is essential for understanding gene regulatory networks.
• While Assay for Transposase-Accessible Chromatin sequencing (ATAC-seq) is now the primary method for chromatin accessibility profiling, current TFBS prediction methods from ATAC-seq are often suboptimal.

Purpose of the Study:

• To develop state-of-the-art TFBS prediction models utilizing ATAC-seq data.
• To provide the scientific community with a comprehensive and user-friendly resource for TFBS prediction.
• To enable accurate TFBS prediction across diverse cell types, including single-cell resolution.

Main Methods:

• Curated an extensive benchmark dataset for training TFBS prediction models using ATAC-seq data, covering 127 human TFs.
• Developed "maxATAC", a suite of deep neural network models for genome-wide TFBS prediction from ATAC-seq.
• Validated model performance on primary cells and single-cell ATAC-seq data.

Main Results:

• maxATAC provides the largest collection of high-performance TFBS prediction models specifically for ATAC-seq data.
• The models achieve state-of-the-art performance, significantly improving TFBS prediction accuracy.
• Demonstrated the utility of maxATAC in identifying TFBS associated with genetic risk loci for atopic dermatitis.

Conclusions:

• maxATAC represents a significant advancement in TFBS prediction from ATAC-seq data.
• The tool enhances our ability to study gene regulation and cellular function.
• maxATAC facilitates improved understanding of genotype-phenotype relationships, particularly in complex diseases.