EnsembleSE: identification of super-enhancers based on ensemble learning

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Super-enhancers (SEs) are critical regulatory elements driving high gene expression, essential for understanding gene networks and biological processes.
• Traditional experimental SE identification is laborious and costly.
• Existing sequence-based deep learning methods lack interpretability and struggle with limited data.

Purpose of the Study:

• To develop a more effective and interpretable computational model for identifying super-enhancers.
• To improve the generalization ability and accuracy of SE prediction by integrating diverse biological features.
• To provide a tool for discovering cell-specific and species-specific SE patterns.

Main Methods:

• Developed an ensemble model (EnsembleSE) using an integration strategy to enhance generalization.
• Implemented a multi-angle feature representation combining local and global sequence information.
• Integrated physicochemical properties with sequence data for improved feature extraction and interpretability.

Main Results:

• EnsembleSE demonstrated improved performance on human and mouse datasets compared to state-of-the-art models.
• Achieved an average improvement of 4.5% in F1 score and 8.05% in recall.
• The model's feature representation enhances effectiveness and interpretability, supporting the discovery of SE patterns.

Conclusions:

• The proposed ensemble model, EnsembleSE, offers a robust and adaptable approach for accurate super-enhancer identification.
• Integrating diverse biological features significantly improves prediction accuracy and model interpretability.
• EnsembleSE provides valuable technical support for advancing research in gene regulation and disease mechanisms.