EPI-Trans: an effective transformer-based deep learning model for enhancer promoter interaction prediction

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Enhancer-promoter interactions (EPIs) are vital for gene regulation and human development.
• Experimental identification of EPIs is resource-intensive.
• Current computational methods often fail to capture long-range sequence dependencies.

Purpose of the Study:

• To introduce EPI-Trans, a novel transformer-based deep learning model for predicting EPIs.
• To address limitations in existing computational approaches for EPI recognition.
• To develop a transferable model for diverse cell line applications.

Main Methods:

• Utilized a transformer architecture with a multi-head attention mechanism to learn inter-sequence relationships.
• Developed a generic pre-trained model for broad applicability.
• Fine-tuned the model using specific cell line datasets for enhanced performance.

Main Results:

• EPI-Trans demonstrated superior performance compared to existing methods across six benchmark cell lines.
• Achieved an average AUROC of 95.7% and AUPR of 79.6% with the best model configuration.
• The generic model showed a competitive average AUROC of 95%.

Conclusions:

• EPI-Trans offers a powerful deep learning solution for predicting enhancer-promoter interactions.
• The model effectively captures long-range dependencies crucial for accurate EPI recognition.
• EPI-Trans outperforms current state-of-the-art techniques in computational EPI prediction.