hicGAN

hicGAN employs generative adversarial networks to infer high-resolution Hi-C contact matrices from low-resolution Hi-C data, improving resolution for 3D chromatin conformation analysis.

Key Features:

• Generative Adversarial Networks (GANs): Uses a GAN architecture with generator and discriminator components to synthesize high-resolution Hi-C contact matrices from low-resolution inputs.
• Resolution Enhancement: Produces high-resolution Hi-C matrices that closely match empirical high-resolution datasets to support more accurate downstream analyses.
• Low-resolution/low-coverage inference: Infers high-resolution contact patterns from low-resolution or low-coverage Hi-C data to mitigate limited availability of high-resolution datasets.

Scientific Applications:

• 3D Chromatin Conformation Analysis: Enables improved identification of topologically associating domains (TADs) and chromatin loops from enhanced-resolution Hi-C matrices.
• Functional Genomics Insights: Facilitates linking distal regulatory elements to putative target genes by resolving finer-scale chromatin contacts.
• Cross-tissue and Cross-cell Type Analysis: Supports comparative chromatin conformation studies when high-resolution Hi-C data are unavailable for specific tissues or cell types.

Methodology:

hicGAN trains a GAN in which a generator produces high-resolution-like Hi-C contact matrices and a discriminator evaluates them against real high-resolution Hi-C data, using iterative adversarial training to refine outputs for consistency with empirical high-resolution datasets.