DeepEnhancerPPO: An Interpretable Deep Learning Approach for Enhancer Classification.

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Enhancers are crucial non-coding genomic elements regulating gene expression.
• Current methods for classifying enhancer categories and strengths are limited.
• Accurate enhancer classification is vital for understanding transcriptional regulation.

Purpose of the Study:

• To develop a novel deep learning model for accurate enhancer classification.
• To improve the classification of enhancer categories and regulatory strengths.
• To enhance the interpretability of enhancer classification results.

Main Methods:

• Proposed DeepEnhancerPPO, an end-to-end deep learning architecture.
• Integrated ResNet and Transformer modules for feature extraction.
• Employed Proximal Policy Optimization (PPO) for feature dimensionality reduction.

Main Results:

• DeepEnhancerPPO achieved superior performance on independent datasets.
• Outperformed existing models by 6.7% in enhancer category classification accuracy.
• Demonstrated robust enhancer strength classification, ranking among the top classifiers.

Conclusions:

• DeepEnhancerPPO offers a robust and accurate framework for enhancer classification.
• The model enhances understanding of gene regulation through improved classification.
• Proximal Policy Optimization improves both performance and interpretability.