EnhancerDetector : Enhancer Discovery from Human to Fly via Interpretable Deep Learning.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Enhancers are crucial non-coding DNA elements regulating gene expression.
• Accurate identification of enhancers is a significant challenge in genomics.

Purpose of the Study:

• To introduce EnhancerDetector, a novel framework for cross-species enhancer prediction.
• To combine high accuracy with biological interpretability in enhancer identification.

Main Methods:

• Development of a convolutional neural network-based framework (EnhancerDetector).
• Training on human data and testing across human, mouse, and fly datasets.
• Application of class activation maps for identifying predictive sequence regions.

Main Results:

• EnhancerDetector outperforms existing methods in precision and F1 score across species.
• The framework demonstrates effective generalization to diverse experimental assay datasets.
• Experimental validation in transgenic flies confirmed predictive power, with 5/6 candidates showing reporter expression.

Conclusions:

• EnhancerDetector provides a practical, accurate, and interpretable framework for enhancer discovery.
• The tool is adaptable for new species and effective with limited genomic data.
• Identified sequence features define a characteristic "enhancerness" signature.