EPI-Mind: Identifying Enhancer-Promoter Interactions Based on Transformer Mechanism

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Enhancer-Promoter Interactions (EPIs) are crucial for gene regulation.
• Traditional methods for detecting EPIs are costly and time-consuming.
• Computational approaches are needed to efficiently study EPIs.

Purpose of the Study:

• To develop a novel deep learning model for predicting EPIs using only sequence features.
• To create a model capable of predicting EPIs across different cell lines.
• To improve upon existing computational methods for EPI prediction.

Main Methods:

• A deep learning framework, EPI-Mind, was developed using sequence-based features.
• Convolutional Neural Networks (CNNs) and transformer mechanisms were employed for feature extraction.
• Multiple model versions were trained, including cell-type specific (EPI-Mind_spe), general (EPI-Mind_gen), and an optimized version (EPI-Mind_best).

Main Results:

• EPI-Mind_spe achieved high performance (AUROC > 90%, AUPR > 70%) but was cell-type specific.
• EPI-Mind_gen demonstrated improved generalizability across cell lines, with a 4.8% increase in AUROC over EPI-Mind_spe.
• EPI-Mind_best outperformed state-of-the-art methods on benchmark datasets, excelling in multiple performance metrics.

Conclusions:

• The proposed EPI-Mind deep learning framework accurately predicts Enhancer-Promoter Interactions from sequence data.
• EPI-Mind provides a novel and efficient computational route for studying gene regulation mechanisms.
• This approach offers a valuable alternative to experimental methods for EPI detection.