DLoopCaller

DLoopCaller predicts genome-wide chromatin loops by applying a deep learning model that integrates accessible chromatin landscapes with raw Hi-C contact maps to infer three-dimensional genome interactions.

Key Features:

• Integration of Data Types: Combines accessible chromatin landscapes with raw Hi-C contact maps to leverage one-dimensional and three-dimensional genomic information for loop prediction.
• Utilization of Orthogonal Data: Uses orthogonal datasets such as ChIA-PET, HiChIP, and Capture Hi-C to generate positive training samples and expand contact matrix coverage.
• Improved Accuracy: Demonstrates higher prediction accuracy in benchmark comparisons relative to Peakachu.
• Identification of Unique Interactions: Detects chromatin interactions not reported by HiCCUPS and Fit-Hi-C, providing additional interaction calls.

Scientific Applications:

• Understanding 3D Genome Organization: Provides genome-wide loop predictions to support analyses of spatial genome arrangement and gene regulation.
• Cell-Type Specificity Analysis: Enables analysis of cell-type specificity in chromatin loop formation across cell lines.
• Transcription Factor Motif Co-Enrichment: Facilitates investigation of transcription factor motif co-enrichment across cell lines and species.

Methodology:

Applies a deep learning model that integrates accessible chromatin landscapes with raw Hi-C contact maps and trains using positive samples derived from ChIA-PET, HiChIP, and Capture Hi-C.