EnHiC

EnHiC predicts high-resolution Hi-C contact maps from low-resolution Hi-C matrices to recover fine-scale chromatin interactions for studying three-dimensional genome architecture.

Key Features:

• Generative Adversarial Network (GAN) framework: Uses a GAN-based approach to predict high-resolution Hi-C matrices from low-resolution inputs.
• NMF-inspired approach: Leverages principles from non-negative matrix factorization to exploit intrinsic properties of Hi-C matrices.
• Rank-1 feature extraction from multi-scale matrices: Extracts rank-1 features from multi-scale low-resolution matrices to enhance resolution.
• Performance on human Hi-C datasets: Demonstrated superior resolution enhancement compared with other GAN-based models on three human Hi-C datasets.

Scientific Applications:

• Detection of Topologically Associated Domains (TADs): Facilitates accurate identification of TADs from enhanced-resolution contact maps.
• Fine-scale chromatin interaction analysis: Enables analysis of fine-scale chromatin interactions to inform studies of genome organization and regulation.

Methodology:

Applies a generative adversarial network tailored to Hi-C data and integrates rank-1 feature extraction inspired by non-negative matrix factorization from multi-scale low-resolution matrices.