DeepHiC

DeepHiC predicts high-resolution Hi-C contact maps from low-coverage sequencing data using a generative adversarial network to improve detection of chromatin loops and topologically associating domains (TADs).

Key Features:

• Generative Adversarial Network (GAN) Architecture: Uses a GAN with adversarial training to generate high-resolution Hi-C contact maps that resemble experimentally obtained high-resolution matrices.
• Low-Coverage Data Enhancement: Reconstructs high-resolution contact maps from low-coverage or downsampled sequencing data, including demonstrations with as little as 1% downsampled reads.
• Improved Biological Interpretability: Restores fine-grained patterns in Hi-C matrices to enhance the accuracy of chromatin loop identification and topologically associating domain (TAD) detection.

Scientific Applications:

• Chromatin Loop Detection: Provides enhanced resolution to support more precise identification of chromatin loops from Hi-C data.
• TAD Detection: Improves the detection and boundary definition of topologically associating domains (TADs) in contact maps.
• Developmental Studies: Facilitates analysis of dynamic chromatin architecture changes in contexts such as mouse embryonic development using low-coverage Hi-C data.

Methodology:

A generative adversarial network (GAN) trained with an adversarial framework to predict high-resolution Hi-C contact maps from low-coverage or downsampled Hi-C sequencing data, including tests with 1% downsampled reads.