Capturing large genomic contexts for accurately predicting enhancer-promoter interactions.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Enhancer-promoter interactions (EPIs) are crucial for gene regulation but challenging to predict due to distant gene targeting.
• Existing machine learning models for EPI prediction often lack comprehensive feature utilization, leading to moderate performance, especially across different cell types.

Purpose of the Study:

• To develop a novel Transformer-based model, TransEPI, for enhanced prediction of enhancer-promoter interactions (EPIs).
• To leverage large genomic contexts for improved EPI prediction accuracy and robustness across diverse cell types.

Main Methods:

• Developed TransEPI, a Transformer-based deep learning model, utilizing extensive genomic sequence data.
• Trained and validated TransEPI on large-scale EPI datasets derived from Hi-C or ChIA-PET data across six cell lines.
• Employed rigorous testing on independent datasets with differing cell lines and chromosomes to prevent overfitting and assess generalization.

Main Results:

• TransEPI demonstrated consistent and superior performance across cross-validation and independent test datasets from various cell types.
• The model significantly outperformed existing state-of-the-art machine learning and deep learning methods in EPI prediction.
• Integration of large genomic contexts was identified as the key factor contributing to TransEPI's enhanced predictive power.

Conclusions:

• TransEPI offers a powerful new approach for accurate enhancer-promoter interaction prediction by incorporating broader genomic information.
• The model's effectiveness extends to identifying target genes for non-coding mutations linked to brain and neural diseases.
• TransEPI holds significant potential for advancing gene regulation studies and understanding the genetic basis of neurological disorders.