EpiMCI: Predicting Multi-Way Chromatin Interactions from Epigenomic Signals

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• High-throughput Pore-C (HiPore-C) enables whole-genome high-order chromatin multi-way interaction identification, crucial for understanding 3D genome organization.
• Analyzing HiPore-C data presents significant computational challenges due to its high output.

Purpose of the Study:

• To develop an accurate computational model for predicting multi-way chromatin interactions.
• To address the data analysis challenges posed by high-throughput sequencing technologies like HiPore-C.

Main Methods:

• Proposed EpiMCI, a hypergraph neural network model utilizing epigenomic signals as input.
• Integrated separate hyperedge representations with coupling hyperedge information for enhanced prediction.

Main Results:

• EpiMCI achieved high performance with AUCs of 0.981 (GM12878) and 0.984 (K562), outperforming existing methods.
• Demonstrated efficacy in denoising HiPore-C data and improving data quality.
• Vertex embeddings from EpiMCI accurately reflect global chromatin architecture and align with genomic region activities.

Conclusions:

• EpiMCI provides an accurate and efficient method for predicting multi-way chromatin interactions.
• The model is valuable for studies focused on chromatin architecture and 3D genome organization, particularly when using HiPore-C data.