DeepCAGE

DeepCAGE predicts genome-wide chromatin accessibility across human cell types by integrating DNA sequence with human core transcription factor (TF) expression and binding information using a densely connected deep convolutional neural network.

Key Features:

• Deep learning architecture: Employs a densely connected deep convolutional neural network (CNN) that automatically learns sequence signatures of chromatin-accessible regions.
• Integration with transcription factors: Incorporates expression levels and binding activities of human core TFs alongside sequence features to inform accessibility predictions.
• Prediction tasks: Performs both classification and regression of chromatin accessibility signals genome-wide.
• Motif discovery and TF contribution: Extracts novel binding motifs and quantifies the contribution of specific TFs to regulatory activity at particular loci across cell types.
• Variant prioritization from WGS: Applies to whole-genome sequencing data to prioritize putative deleterious variants associated with complex human traits.

Scientific Applications:

• Transcriptional regulation: Predicts accessibility to aid identification of regulatory elements and TF–target interactions.
• Epigenetics and chromatin dynamics: Enables genome-wide annotation of chromatin-accessible regions to study chromatin state changes across conditions and cell types.
• Complex trait genetics: Prioritizes non-coding variants from whole-genome sequencing for association with complex human traits.
• Regulatory landscape annotation: Provides interpretable annotations of accessibility and TF contributions across diverse human cell types.
• Disease mechanism investigation: Facilitates linking regulatory variation at loci to potential effects on gene regulation in disease contexts.

Methodology:

Trains a densely connected deep convolutional neural network on known chromatin-accessible regions using DNA sequence and TF expression/binding data, and performs classification and regression of accessibility signals.