GILoop

GILoop detects CTCF-mediated chromatin loops from Hi-C contact maps using a dual-branch neural network that integrates graph-based and image-based representations.

Key Features:

• Dual-Branch Neural Network Architecture: Processes Hi-C contact maps simultaneously as graph-based and image-based inputs to leverage complementary information for loop detection.
• Integration of Multiple Data Views: Synergizes graph and image views of Hi-C contact maps to improve identification of genome-wide CTCF-mediated loops.
• Robustness Against Low-Quality Data: Maintains detection performance on low-quality Hi-C libraries.
• Insights into Matrix Density Preferences: Identifies distinct matrix density preferences between graph-based and image-based models in Hi-C data.
• Transfer Learning Capabilities: Applies transfer learning to generalize CTCF-mediated looping patterns across different cell lines.

Scientific Applications:

• Chromatin loop detection: Accurate identification of genome-wide CTCF-mediated chromatin loops from Hi-C contact maps.
• Genome organization analysis: Characterization of higher-order chromatin architecture and spatial genome organization.
• Gene regulation and epigenetics: Study of relationships between CTCF-mediated loops, gene regulation, and epigenetic mechanisms.
• Cross–cell-line comparative analysis: Transfer-learning-enabled comparison of CTCF loop organization and function across cell lines.

Methodology:

A dual-branch neural network that processes Hi-C contact maps as graph-based and image-based representations, using transfer learning to model CTCF-mediated loop patterns and to compare matrix density preferences.