iEnhance: a multi-scale spatial projection encoding network for enhancing chromatin interaction data resolution

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• High-throughput chromatin interaction data are crucial for understanding 3D genome architecture.
• Data sparseness and high signal-noise ratio limit the precision of structural elements derived from these methods.
• Existing tools struggle to accurately recover both short-range and long-range interaction patterns.

Purpose of the Study:

• To develop a novel computational tool, iEnhance, for enhancing the resolution of chromatin interaction data.
• To improve the precision of 3D genome structural elements by overcoming data limitations.
• To create a versatile tool applicable to various chromatin interaction datasets and cell types.

Main Methods:

• iEnhance employs a multi-scale spatial projection and encoding network.
• It extracts multi-scale global and local features from input data using matrix projection.
• Features are hierarchically fused via an attention mechanism, followed by dense channel encoding and residual channel decoding.

Main Results:

• iEnhance significantly outperforms existing state-of-the-art Hi-C resolution enhancement tools.
• The method precisely recovers both short-range structural elements and long-range interaction patterns.
• iEnhance demonstrates robust performance across diverse datasets, including single-cell Hi-C and Micro-C, and is transferable to new cell lines.

Conclusions:

• iEnhance provides a robust and accurate method for high-resolution chromatin interaction data enhancement.
• The tool overcomes limitations of sparse and noisy data, enabling more precise 3D genome architecture analysis.
• iEnhance offers broad applicability and transferability for various genomic studies.