ChromeGCN

ChromeGCN predicts DNA chromatin profiles by integrating short DNA sequence data with 3D genome interactions using a graph convolutional network to model epigenetic states.

Key Features:

• Integration of Local and Long-Range Data: Fuses short DNA sequence windows with long-range 3D genome interactions to represent spatial dependencies in chromatin state prediction.
• Graph Convolutional Network Architecture: Uses a graph convolutional network that explicitly incorporates known long-range genomic interactions to model complex dependencies influencing epigenetic states.
• Improved Predictive Performance: Outperforms state-of-the-art deep learning methods in chromatin profile prediction across three evaluation metrics, with notable gains on DNA windows with high interaction degree.
• Application to Epigenetic Analyses: Captures both local and long-range interactions to improve analysis of transcription factor binding and other epigenetic phenomena.

Scientific Applications:

• Transcription Factor Binding and Epigenetic Mechanisms: Supports identification and prediction of transcription factor binding and epigenetic modifications by incorporating spatial genome organization.
• Gene Regulation and Therapeutic Target Studies: Aids studies of gene regulatory mechanisms and identification of potential therapeutic targets influenced by complex DNA interactions.

Methodology:

Constructs a graph-based representation combining sequence information and spatial 3D interactions and applies a graph convolutional network, relaxing the independent and identically distributed assumption of local windows.