Predicting unrecognized enhancer-mediated genome topology by an ensemble machine learning model.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Transcriptional enhancers regulate gene expression over long genomic distances.
• Identifying enhancer-promoter interactions is crucial for understanding development and disease.
• Current proximity-ligation assays (e.g., HiChIP, ChIA-PET) are costly and complex.

Purpose of the Study:

• To develop a computational model, LoopPredictor, for predicting genome topology and enhancer-promoter loops.
• To enable the study of enhancer-mediated gene regulation in cell types lacking experimental contact maps.
• To facilitate the identification of disease-associated variants linked to distal regulatory elements.

Main Methods:

• Developed LoopPredictor, an ensemble machine learning model.
• Trained the model using H3K27ac and YY1 HiChIP data for functional loop enrichment.
• Integrated multi-omics features for loop identification and annotation.
• Evaluated cross-species prediction capabilities using human and mouse data.

Main Results:

• LoopPredictor efficiently predicts cell type-specific enhancer-promoter loops and promoter-promoter interactions.
• Model performance is comparable to experimental H3K27ac HiChIP data.
• Predicted enhancer loops show high conservation across species (human and mouse).

Conclusions:

• LoopPredictor overcomes limitations of experimental assays for studying genome topology.
• The model enables dissection of cell type-specific long-range gene regulation.
• LoopPredictor accelerates the identification of distal disease-associated risk variants.